Keywords: Intelligent and Flexible Manufacturing, Logistics
Abstract: Manufacturing and logistics systems operate for decades and need to evolve to manufacture new products, increase quality, energy, or overall efficiency. Consequently, automation hardware and software adaptation in the operation phase is crucial. Means to design such automation architectures compliant to Industry 4.0 are of high economic interest. The talk will introduce means to analyze existing automation architectures as a first step to refactoring. In the second step, the integration of AI into such architectures will be discussed. Finally, automation architectures that ease the adaptation of physical manufacturing and logistics systems will be presented.

Keywords: AI-Based Methods, Machine learning
Abstract: Artificial Intelligence (AI) uses computer algorithms to simulate human intelligence, mainly focused on learning and decision-making processes. Due to the maturity of the area and the new advances in AI branches such as machine and deep learning, the industrial applications of AI have been increasing rapidly. Today AI is present in the algorithms to drive cars, land planes, render images, make decisions, among many other applications. This growth makes us wonder, what is the future of AI in autonomous vehicles (cars, planes)? How will AI solve problems in artificial vision or in autonomous surgical systems?, and also reflect on some regulations and ethical questions that must be addressed when using AI to solve critical problems. This panel looks at how Continental, Intel and Wizeline view the use of AI to solve industrial problems.

Keywords: Modelling, Simulation and Optimization in Healthcare, AI and Machine Learning in Healthcare, Data fusion
Abstract: Physiological status diagnosis plays an important role in clinical practice. Different personal information hinders the practical application heavily. To address this issue, we propose a tensor-based physiological status diagnosis approach, fused the subject-variant information with physiological data. The subject-variant information guided similarity information matrix is employed to regularize the tensor-based formulation so that the subject-variant information can be appropriately adopted. We proposed an alternating direction method of multipliers (ADMM) inbuilt with the block coordinate descent (BCD) algorithm to solve this formulation. A real-case dataset has been used to validate the proposed diagnosis method, which shows satisfactory results compared with other existing methods.

Keywords: AI and Machine Learning in Healthcare, AI-Based Methods, Big-Data and Data Mining
Abstract: Diabetic retinopathy (DR) is a microvascular complication of diabetes and is a leading cause of vision loss and blindness. Screening and early detection of DR is critical but current screening methods rely on eye care experts and expensive medical equipment, which are not available in medically underserved communities. The non-image-based, machine-learning approach in this study aims to detect DR in the early stage using demographics, comorbidities, and routine lab results data, which are widely available for diabetic patients. We develop different temporal deep learning models to analyze a real-world, large-scale dataset and compare performances of these models. Experimental results show that temporal models outperform baseline random forest models in metrics of AUPRC and recall.

Keywords: Causal Models, AI and Machine Learning in Healthcare, Health Care Management
Abstract: Imaging phenotypes extracted via radiomics of magnetic resonance imaging has shown great potential at predicting the treatment response in breast cancer patients after administering neoadjuvant systemic therapy (NST). Existing machine learning models are, however, limited in providing an expert-level interpretation of these models, particularly interpretability towards generating causal inference. Causal relationships between imaging phenotypes, clinical information, molecular features, and the treatment response may be useful in guiding the treatment strategies, management plans, and gaining acceptance in medical communities. In this work, we leverage the concept of counterfactual explanations to extract causal relationships between various imaging phenotypes, clinical information, molecular features, and the treatment response after NST. We implement the methodology on a publicly available breast cancer dataset and demonstrate the causal relationships generated from counterfactual explanations. We also compare and contrast our results with traditional explanations, such as LIME and Shapley.

Keywords: Factory Automation, Cloud Computing For Automation
Abstract: This session explores the latest trends in adoption of robotics, artificial intelligence, machine vision and related automation. Real world examples of leading applications in major industries such as manufacturing, warehousing & distribution, and more will be discussed, as well as the impacts of increased automation on jobs.

Keywords: Computer Vision in Automation, Collision Avoidance, Industrial and Service Robotics
Abstract: A fully automated and reliable picking of a diverse range of previously unseen objects in clutter is a challenging problem. This becomes even more difficult given the inherent uncertainty in sensing, control, and interaction-physics. This paper presents a robust method for a stable and collision- free grasp planning, given a cluttered heap of novel objects of different varieties. Our grasp planning pipeline leverages a novel grasp pose ranking method and a pose refinement method that ensures collision-free gripping and stable contact between gripper-fingers and the target object. Often, a grasp planning algorithm may not be able to find a valid grasp pose due to the tightly packed configuration of the objects. In such situations, our method directs the robot to perform a clutter removal action using a linear push policy. On a physical robot with a two-fingered parallel-jaw gripper and a depth sensor, our method is able to consistently clear-up the pile of upto 20 objects with 95% reliability.

Keywords: Failure Detection and Recovery, Machine learning, Big-Data and Data Mining
Abstract: A timely fault detection method is crucial to ensure the safety and reliability of the battery pack for electric vehicles in real-life applications especially under dynamic environmental and driving conditions. This paper presents a physics guided data-driven approach to robustly detect battery system faults in field fleet. The proposed approach integrates the physical knowledge and data insights to improve the fault detection accuracy by extracting operating condition-aware health indicators and identifying the battery faults through a data-driven outlier detection model. The proposed approach is validated using the data collected from a field fleet, and the results demonstrate the effectiveness of the design.

Keywords: Robotics and Automation in Life Sciences, Computer Vision for Automation
Abstract: Transparent objects are widely used in our daily lives and therefore robots need to be able to handle them. However, transparent objects suffer from light reflection and refraction, which makes it challenging to obtain the accurate depth maps required to perform handling tasks. In this paper, we propose a novel affordance-based framework for depth reconstruction and manipulation of transparent objects, named A4T. A hierarchical AffordanceNet is first used to detect the transparent objects and their associated affordances that encode the relative positions of an object's different parts. Then, given the predicted affordance map, a multi-step depth reconstruction method is used to progressively reconstruct the depth maps of transparent objects. Finally, the reconstructed depth maps are employed for the affordance-based manipulation of transparent objects. To evaluate our proposed method, we construct a real-world dataset TRANS-AFF with affordances and depth maps of transparent objects, which is the first of its kind. Extensive experiments show that our proposed methods can predict accurate affordance maps, and significantly improve the depth reconstruction of transparent objects compared to the state-of-the-art method, with the Root Mean Squared Error in meters significantly decreased from 0.097 to 0.042. Furthermore, we demonstrate the effectiveness of our proposed method with a series of robotic manipulation experiments on transparent objects. See supplementary video and results at https://sites.google.com/view/affordance4trans

Keywords: Automation at Micro-Nano Scales, Mechatronics in Meso, Micro and Nano Scale
Abstract: Autonomous manipulation of micro- and nanoscale objects is of major interest for various research applications. Precise manipulation of micro- and nanowires through visual feedback is challenging because of the difficulty of observing their motion along the line-of-sight of microscopes. In this paper, we present a novel auto-focusing and visual posture estimation strategy for identifying three-dimensional (3D) poses for one or more moving micro- and nanowires under bright-field microscopes. The proposed method integrates classic passive auto-focusing algorithms, rule-based hill-climb methods, and an automatic and efficient scheme to estimate the positions and orientations of moving wires. Extensive experimental results demonstrate high accuracy and efficiency of the tracking and 3D pose estimation compared to traditional methods. This work lays the foundation for automated control of micro- and nano-robots in 3D microfluidic environments.

Keywords: Inventory Management, Motion and Path Planning, Optimization and Optimal Control
Abstract: Shelves are commonly used to store objects in homes, stores, and warehouses. We formulate the problem of Optimal Shelf Arrangement (OSA), where the goal is to optimize the arrangement of objects on a shelf for access time given an access frequency and movement cost for each object. We propose OSA-MIP, a mixed-integer program (MIP), show that it finds an optimal solution for OSA under certain conditions, and provide bounds on its suboptimal solutions in general cost settings. We analytically characterize a necessary and sufficient shelf density condition for which there exists an arrangement such that any object can be retrieved without removing objects from the shelf. Experimental data from 1,575 simulated shelf trials and 54 trials with a physical Fetch robot equipped with a pushing blade and suction grasping tool suggest that arranging the objects optimally reduces the expected retrieval cost by 60-80% in fully-observed configurations and reduces the expected search cost by 50-70% while increasing the search success rate by up to 2x in partially-observed configurations.

Keywords: Collaborative Robots in Manufacturing, Motion Control, Industrial and Service Robotics
Abstract: A wheeled mobile manipulator, consisting of a mobile platform and a manipulator arm, can be used to provide assistive handling of loads without requiring large investments or changes in factory layout. This paper introduces a novel constraint-based control architecture in which such assistive tasks can be defined. From the same constraint-based task specification two constraint-based controllers are generated in such a way that the more accurate and higher bandwidth manipulator arm can compensate for the inaccuracies and lower bandwidth introduced by the mobile platform. The effectiveness of this approach is validated using a null-space motion task and a guided peg insertion task. The experiments clearly show an improvement in the accuracy of the null-space motion task and an increased comfort-level of the operator in the guided peg insertion task.

Keywords: Motion Control, Learning and Adaptive Systems, Industrial and Service Robotics
Abstract: In this paper, we propose to use a nonlinear adaptive PID controller to regulate the joint variables of a mobile manipulator. The motion of the mobile base forces undue disturbances on the joint controllers of the manipulator. In designing a conventional PID controller, one should make a trade-off between the performance and agility of the closed-loop system and its stability margins. The proposed nonlinear adaptive PID controller provides a mechanism to relax the need for such a compromise by adapting the gains according to the magnitude of the error without expert tuning. Therefore, we can achieve agile performance for the system while seeing damped overshoot in the output and track the reference as close as possible, even in the presence of external disturbances and uncertainties in the modeling of the system. We have employed a Bayesian optimization approach to choose the parameters of a nonlinear adaptive PID controller to achieve the best performance in tracking the reference input and rejecting disturbances. The results demonstrate that a well-designed nonlinear adaptive PID controller can effectively regulate a mobile manipulator's joint variables while carrying an unspecified heavy load and an abrupt base movement occurs.

Keywords: Motion Control, Robotics and Automation in Life Sciences
Abstract: In this paper, an Observer-Free Output Feedback (OF2) tracking controller is formulated for a robotic manipulator, in order to improve performance during human-robot collaboration. The OF2 controller is based on a set of filtered error dynamics that avoids the need for direct speed measurements or observer design. The main advantage of this method is that it is model-free and robust to changes in operating conditions often present in environments where humans and robots work together. Moreover, OF2 controller is demonstrably stable, thus safe, and a Lyapunov stability proof is offered using a nominal dynamic model of the robot. Collaborative robots have highly nonlinear and uncertain dynamic models and are ideal candidates for our controller. The controller can be used to not only compensate for the unknown system parameters, but also reject external disturbances, such as human or environmental forces. Tracking performance of our controller has been tested experimentally on the Baxter collaborative robot under different trajectory tracking and impact experiments with and without payload. The results have been compared with the factory built-in and pre-tuned PID controller. Results indicate that our controller shows an order of magnitude improvement in the trajectory tracking performance and a reduction in joint efforts required during learning from demonstration and assembly tasks.

Keywords: Motion Control, Sensor-based Control, Formal Methods in Robotics and Automation
Abstract: The target of this paper is to model and to design a controller for a piezoelectric actuator that is devoted to robotic hand. The actuator is characterized by the following properties simultaneously: strong asymmetrical hysteresis nonlinearity, creep nonlinearity, and oscillations in its dynamics. Moreover, the actuator is also exhibited to external disturbance which is the manipulation force. To account for these specificities, we propose to use the generalized Bouc-Wen model for the hysteresis, along with a lumped disturbance term that encompasses the creep and the external force. We also propose to use a high order model for the dynamics. Then, an active disturbance rejection in combination with an extended state observer is suggested and designed as control law. Experiments are carried out and demonstrate the efficiency of the proposed approach.

Keywords: Hydraulic/Pneumatic Actuators, Robust/Adaptive Control, Optimization and Optimal Control
Abstract: This paper presents an Adaptive Model Predictive Control (AMPC) approach for the position tracking and force control of a hydraulic actuator (HA). Due to its nonlinear dynamics, the iterative linearization paradigm is employed to approximate the HA system by a linear time-varying model. Such a representation is used as the internal plant model of the predictive controller to effectively make predictions on the system state. The effectiveness of the proposed AMPC architecture is shown through numerical experiments addressing the control of a real HA on different scenarios. Finally, a comparative analysis on several values of sampling time, prediction and control horizon is carried out in order to investigate the effect of the parameters tuning on the performance of the closed-loop control system.

Keywords: Diagnosis and Prognostics
Abstract: In this paper, we consider input-constrained differential-drive robots whose setpoint (reference) signals are sent by a remote control station via a wireless, and possibly insecure, communication channel. The objective is to develop an anomaly detector capable of revealing the presence of false data injections on the setpoint signal. Here, this is achieved by exploiting Command Governor (CG) supervisor module arguments and a detection unit. Specifically, two CG units, installed at the two ends of the communication channel, use a feedback linearized model of the vehicle dynamics to filter the reference signal according to the setpoints complying with the vehicle's constraints. The detection unit, directly placed on the vehicle, takes advantage of the CGs actions to detect anomalies in the received reference signal. Moreover, a setpoint randomization procedure is proposed in order to avoid the existence of stealthy attacks. Simulation results involving a differential-drive robot are presented to show the effectiveness of the proposed solution.

Keywords: Modelling, Simulation and Validation of Cyber-physical Energy Systems
Abstract: The increasing penetration of renewable energy generation (REG) introduces an amount of uncertainty and power electronics devices into power grid that severely affects the physical responses in the evolution of cascading failure. However, existing cascading failure models cannot accurately describe the physical responses affected by the features of REG and their interaction in the propagation of cascading failure. In this paper, a cascading failure model integrating the multiply physical responses is proposed. In this model, the physical responses affected by REG are described, including relay protection, frequency regulation and dispatching strategy. According to the proposed model, a simulation algorithm of cascading failure process is also developed. In the experiment, our model is compared with the model in traditional power grid. The experimental results show that the penetration of REG will change the evolution path of failure, and the probability of blackouts will increase with the increasing penetration level of REG.

Keywords: Modelling, Simulation and Validation of Cyber-physical Energy Systems, Demand Side Management, Smart Grids
Abstract: Renewable energy sources and electric vehicles provide an effective way to reduce the energy cost of an enterprise microgrid. However, the uncertainties of renewable energy sources and the time coupling characteristic of electric vehicles bring great challenges of non-anticipativity and feasibility for supply-demand coordination. To satisfy the non-anticipativity, we develop a supply-demand coordination optimal model using pre-scheduling method with virtual re-scheduling. In this model, the current decision only depends on the current and past realizations of random variables. Furthermore, we enhance the model with time-coupled robust constraints to guarantee the feasibility of the strategy under all possible realizations of the random variables. These time-coupled robust constraints bring high computational complexity to solve this model. So, we develop the method of combining forward recursion and backward recursion to decouple these time-coupled robust constraints in time. In this way, the coordination model is transformed to a mixed integer linear programming (MILP) model which can be efficiently solved. Finally, numerical test based on a real case is analysed and the results show that the energy cost of the enterprise is about 136129 if the flexible load is about 20% and load shifting and generators rescheduling can reduce the energy cost more than 6%.

Keywords: Modelling, Simulation and Validation of Cyber-physical Energy Systems, Smart Home and City, Sustainability and Green Automation
Abstract: Edge nodes (ENs) in mobile edge computing can support current delay-sensitive applications of the Industrial Internet of Things. ENs are deployed in the network edge and can execute computational tasks offloaded from users’ mobile devices (MDs) in a timely way. However, their computing and communication resources are limited and cannot execute all offloaded tasks. Thus, a cloud data center (CDC) is highly needed and hybrid cloud-edge systems emerge to provide low-delay services. This work investigates a joint optimization problem of task offloading, task partitioning, and user association to minimize the total cost of the system. This work focuses on applications that can be split into multiple dependent subtasks, each of which can be completed in MDs, ENs, and CDC. Specifically, a mixed integer nonlinear program is formulated to minimize the total cost. Then, a hybrid algorithm named Genetic Simulated-annealing-based Particle Swarm Optimizer (GSPSO) is designed to solve it. GSPSO yields a close-to-optimal strategy to jointly optimize connections among MDs and ENs, and allocation ratios among MDs, ENs, and CDC. Experimental results demonstrate that compared with benchmark methods, GSPSO decreases the total cost while fully meeting the completion time requirements of user tasks.

Keywords: Smart Grids, Modelling, Simulation and Validation of Cyber-physical Energy Systems, Power and Energy Systems automation
Abstract: Voltage deviation is one of the most concerning safety problems in the distribution network, which can cause severe damage to electrical equipment and affect its regular operation. Many works have proposed voltage regulation methods based on reactive power injection of distributed energy resources. However, they also provide practical analysis tools for voltage problems and malicious attacks. From the perspective of security, this paper reveals a potential threat scenario of voltage deviation. Adversaries could manipulate the power injection of non-local nodes to cause additional voltage deviation of a chosen node. The attack strategy could be formulated based on the principle of power flow analysis and voltage regulation. Simulation results show that the attack can make the voltage deviation of some nodes exceed the allowable range with limited attack resources. To alleviate this threat, we analyze the voltage safety of the distribution network and provide some defensive strategies for the design and operation of the network.

Keywords: Intelligent and Flexible Manufacturing, Sustainable Production and Service Automation
Abstract: Manufacturers are pursuing energy-efficient production in response to the fluctuating energy price, growing global competition, rigorous international laws and severe environmental crisis. This paper proposes to boost the energy efficiency of production systems by controlling the production. It extends the existing energy saving control research by presenting an integrated modeling, analyzing, and controlling approach. The work starts from the modeling of the production systems, and establishes an analytical model to systematically quantify the production loss resulted from energy saving control and the various disruptions. A dynamic control algorithm is proposed to reduce the energy consumption and maintain the desirable productivity.

Keywords: Intelligent and Flexible Manufacturing, Robust Manufacturing, Hybrid Strategy of Intelligent Manufacturing
Abstract: To improve production efficiency and reduce energy costs in steel manufacturing, this paper investigates the integrated production scheduling problem between continuous casting and hot rolling (IPSP-CCHR) and formulates it as a bi-level optimization model. The upper-level one is the production planning match problem (PPMP) between continuous casting and hot rolling processes, which aims to maximize productivity. The lower-level one is the reheating furnace scheduling problem (RFSP), which aims to minimize the total residence time. Next, we propose a stratified sampling bi-level optimization (SSBLO) algorithm to solve the IPSP-CCHR, which applies Latin Hypercube sampling (LHS) for solving the upper-level problem and estimation of distribution algorithm (EDA) for solving the lower-level problem. In the experiments, computational results via well-synthetic data show the bi-level optimization model and the proposed algorithm is effective and is expected to apply in realistic industrial cases.

Keywords: Machine learning, Failure Detection and Recovery, Big-Data and Data Mining
Abstract: In imbalanced datasets, where the majority class has significantly more instances than the minority class, conventional classification methods exhibit poor minority-class detection performance because they tend to classify most instances as majority instances. To address this problem, this paper presents a general-purpose imbalanced-data preprocessing method that combines two instance-selecting techniques to obtain a clean and balanced set of training instances. The first technique, ensemble outlier filtering, removes outlier instances from both majority and minority classes. The second technique, normalized-cut sampling, samples the majority class aiming to preserve its distribution across the majority region. Our proposed data preprocessing method uses these two techniques and can be combined with any general classification methodology on the sub-sampled data to construct a classification model. Computational results show the proposed method outperforms several widely used imbalanced-data classification methods.

Keywords: Deep Learning in Robotics and Automation, Autonomous Agents, Reinforcement
Abstract: Object Goal Navigation requires a robot to navigate to an instance of a target out-of-view object class in a previously unseen environment. The framework described in this paper, first builds a semantic map of the environment gradually over time, and then repeatedly selects a long-term goal based on the semantic map to locate the target object instance. The long-term goal – ‘where to go’ is formulated as a vision-based deep reinforcement learning problem. Specifically, an Encoder Network is trained to process a semantic map, extract high-level features, and select a long-term goal. In addition, we incorporate data augmentation and Q-function regularization to make the long-term goal selection process more effective. We report experimental results using the photo-realistic Gibson benchmark dataset in the AI Habitat 3D simulation environment to demonstrate that our framework substantially improves performance on standard measures in comparison with state of the art baseline.

Keywords: Deep Learning in Robotics and Automation, Reinforcement, Simulation and Animation
Abstract: On the journey of true autonomous robotics, applying deep reinforcement learning (DRL) techniques to solve complex robotics tasks has been a growing interest in academics and the industry. Currently, numerous simulation frameworks exist for evaluating DRL algorithms with robots, and they usually come with prebuilt tasks or provide tools to create custom environments. Among these, one of the highly sought approaches is using Robot Operating System (ROS) based DRL frameworks for simulation and deployment in the real world. The current ROS-based DRL simulation frameworks like openai_ros or Gym-gazebo provide a framework for creating environments; however, they do not support training with vectorised environments for speeding up the training process and parallel simulations for testing and evaluating meta-learning, multi-task learning and transfer learning approaches. Therefore, we present MultiROS, a 3D robotic simulation framework with a collection of prebuilt environments for deep reinforcement learning (DRL) research to overcome these limitations. This package interfaces with the Gazebo robotic simulator using ROS and provides a modular structure to create ROS-based RL environments. Unlike the others, MultiROS provides support for training with multiple environments in parallel and simultaneously accessing data from each simulation. Furthermore, since MultiROS uses the popular OpenAI Gym interface, it is compatible with most OpenAI Gym based reinforcement learning algorithms that use third-party python frameworks.

Keywords: Deep Learning Methods, Reinforcement Learning, Software Architecture for Robotic and Automation
Abstract: Reinforcement learning (RL) has become an interesting topic in robotics applications as it can solve complex problems in specific scenarios. The small amount of RL-tools focused on robotics, plus the lack of features such as easy transfer of simulated environments to real hardware, are obstacles to the widespread use of RL in robotic applications. FRobs_RL is a Python library that aims to facilitate the implementation, testing, and deployment of RL algorithms in intelligent robotic applications using robot operating system (ROS), Gazebo, and OpenAI Gym. FRobs_RL provides an Application Programming Interface (API) to simplify the creation of RL environments, where users can import a wide variety of robot models as well as different simulated environments. With the FRobs_RL library, users do not need to be experts in ROS, Gym, or Gazebo to create a realistic RL application. Using the library, we created and tested two environments containing common robotic tasks; one is a reacher task using a robotic manipulator, and the other is a mapless navigation task using a mobile robot.

Keywords: Deep Learning in Robotics and Automation, AI-Based Methods, Machine learning
Abstract: Making accurate predictions about the dynamic environment is crucial for the trajectory planning of mobile robots. Predictions are by nature uncertain, and for motion prediction multiple futures are possible for the same historic behavior. In this work, the objective is to predict possible future positions of the target for the collision avoidance purpose for mobile robots by considering different uncertainty by combining a sampling-based idea with data-driven methods. More specifically, we propose a major improvement on a loss function for multiple hypotheses and test it with convolutional neural networks on motion prediction problems. We implement post-processing heuristics that produce multiple Gaussian distribution estimations, and show that the result is suitable for trajectory planning for mobile robots. The method is also evaluated with the Stanford Drone Dataset.

Keywords: Deep Learning in Robotics and Automation, Simulation and Animation, Learning and Adaptive Systems
Abstract: Simulation-to-reality transfer has emerged as a popular and highly successful method to train robotic control policies for a wide variety of tasks. However, it is often challenging to determine when policies trained in simulation are ready to be transferred to the physical world. Deploying policies that have been trained with very little simulation data can result in unreliable and dangerous behaviors on physical hardware. On the other hand, excessive training in simulation can cause policies to overfit to the visual appearance and dynamics of the simulator. In this work, we study strategies to automatically determine when policies trained in simulation can be reliably transferred to a physical robot. We specifically study these ideas in the context of robotic fabric manipulation, in which successful sim2real transfer is especially challenging due to the difficulties of precisely modeling the dynamics and visual appearance of fabric. Results in a fabric smoothing task suggest that our switching criteria correlate well with performance in real. In particular, our confidence-based switching criteria achieve average final fabric coverage of 87.2-93.7% within 55-60% of the total training budget. See https://tinyurl.com/lsc-case for code and supplemental materials.

Keywords: Sensor Fusion, Industrial and Service Robotics, Autonomous Vehicle Navigation
Abstract: A visual localization method utilizing a fisheye monocular camera is proposed to enhance navigation accuracy of autonomous mobile robots in indoor environments for warehouse or service robotics applications. Existing visual infrastructure-aided localization algorithms take advantage of uniquely colored or lit robots that limit their application to ideal lighting conditions, occlusion-free scenarios or multi-modal fusion with stereo vision, LiDAR, and inertial sensors which inevitably increases their complexity. Using fisheye monocular vision imposes challenges such as depth estimation, frame warping, and low accuracy of the state estimation for far objects. The proposed augmented localization framework includes an uncertainty-aware state observer employing a motion model with a learning-based input estimator and point cloud clusters over a region of interest, to estimate the position of a robot while maintaining effective computational efficiency. Observability of the developed state estimator and asymptotic stability of the estimation error dynamics are also studied. Various tests including occlusion, low visibility for far objects, and noisy depth estimation (from the clustered region of interest), have been conducted in indoor settings to validate the method. The tests confirm robust performance of the augmented visual localization framework in presence of intermittent measurements due to environmental conditions or low reliability of vision-based depth estimation. Furthermore, a significant increase in accuracy and consistency of visual localization is shown without using additional stereo, inertial, or LiDAR measurements.

Keywords: Sensor Networks, Sensor Fusion, Data fusion
Abstract: Autonomous robots operating in indoor and GPS denied environments can use LiDAR for SLAM instead. However, LiDARs do not perform well in geometrically-degraded environments, due to the challenge of loop closure detection and computational load to perform scan matching. Existing WiFi infrastructure can be exploited for localization and mapping with low hardware and computational cost. Yet, accurate pose estimation using WiFi is challenging as different signal values can be measured at the same location due to the unpredictability of signal propagation. Therefore, we introduce the use of WiFi fingerprint sequence for pose estimation (i.e. loop closure) in SLAM. This approach exploits the spatial coherence of location fingerprints obtained while a mobile robot is moving. This has better capability of correcting odometry drift. The method also incorporates LiDAR scans and thus, improving computational efficiency for large and geometrically-degraded environments while maintaining the accuracy of LiDAR SLAM. We conducted experiments in an indoor environment to illustrate the effectiveness of the method. The results are evaluated based on Root Mean Square Error (RMSE) and it has achieved an accuracy of 0.88m for the test environment.

Keywords: Sensor-based Control, Computer Vision in Automation, Deep Learning in Robotics and Automation
Abstract: In this work, we present a method that can learn to model dynamic and arbitrary 3D scenes, purely from 2D visual observations. Our approach uses a keypoint-conditioned Neural Radiance Field (KP-NeRF) to capture and model these scenes with the overarching goal of supporting image-based robot manipulation. Differentiating this from previous methods, which typically condition the model on generic embedding vectors for representation, our implicit neural radiance function is conditioned on a set of keypoints that are inferred from a learned encoder given imagery observations. This implicitly separates the visual modeling components into object appearances and object pose configurations. Such inductive bias built into the architecture encourages discovered keypoints to capture state transitions in the robot's environment across time and space. We then learn a forward prediction model of the encoded keypoints, constructed over the keypoint representation space, and perform MPC control for challenging manipulation tasks including block pushing and door closing. We evaluate the performance of our method through various tasks: novel scene view synthesis, action-conditioned forward prediction, and robot manipulation tasks.

Keywords: Simulation and Animation, Assembly, Compliant Assembly
Abstract: Kinematic and dynamic simulations of colliding and sliding bodies have traditionally been based on triangulated object representations, which lead to numerical inaccuracies due to wrong surface normals and inaccurate surface representations. Moreover, contact layers are needed to be able to do efficient simulations. For simulating tight fitting assembly operations, this is particularly critical as the size of numerical inaccuracies and the contact layer are similar to the size of the slack during insertion. Simulations using mathematical surface representations such as planes, cylinders, ellipsoids and B-splines are becoming increasingly popular, and are not suffering from these flaws. It is typically more expensive to find contact points, but once they are found, the contact points can be tracked efficiently if the extrema is unique (singularity free). In this paper, we extend this tracking method for efficiently handling, and also tracking in (near)-singular situations. We outline how this can be done and illustrate it on two cases with singularities.

Keywords: Computer Vision for Manufacturing, Deep Learning in Robotics and Automation
Abstract: The 3D shape and oscillation of weld pool surface contain abundant significant information related to the welding penetration state, which provides clues to control the welding process. In this paper, a biprism stereo vision system based on a single camera was established to sense the 3D surface of weld pool under different penetration states during the pulsed gas metal arc welding process with a V shape groove. We utilize Pyramid Stereo Matching Network to calculate a high-quality disparity map for weld pool image pair, which fuses features from different levels to improve the disparity estimation accuracy in high light and textureless regions. The disparity map contains two and three dimensions information directly related to the penetration state. Then the disparity maps were input into the Residual Neural Network to extract deeper-level features for training and testing. In our experiments, the welding penetration states were classified as four classes, i.e. partial penetration, full penetration, over penetration and burn through. Testing experiments demonstrated 99.7% as the penetration state identification accuracy.

Keywords: Computer Vision for Transportation, Intelligent Transportation Systems, Deep Learning in Robotics and Automation
Abstract: Single object tracking in point clouds is a fundamental component in enabling autonomous vehicles to understand dynamic traffic environments. Earlier tracking confidence only relies on IoU between two static boxes, ignoring the motion properties of objects, which may weaken the association abilities of trackers. To comprehensively associate an object with the estimated motion state, we introduce a directed representation. This representation factorizes the box of an object into its central position and orientation. To handle under-detection and over-detection problems, we also present an undirected range suppression mechanism that automatically enlarges and stabilizes the view field at the current time step. As a result, we build a single object tracking system that achieves high accuracy and real-time performance. On both KITTI and nuScenes tracking datasets, we demonstrate that our system outperforms other recent single object trackers in both accuracy and speed. Besides, we also validate the superiority of our approach compared to multiple object tracking methods.

Keywords: Planning, Scheduling and Coordination, Discrete Event Dynamic Automation Systems, Cyber-physical Production Systems and Industry 4.0
Abstract: Automated guided vehicle (AGV) systems are widely used in different industrial environments. The performance of these systems depends heavily on the control strategies used, among others to ensure all movements are executed in a deadlock-free manner. In this paper, we propose a deadlock avoidance algorithm that is proven to result in deadlock-free behavior for limited known future movements of all AGVs. The algorithm can be applied to a system with a tessellated layout, where the drivable space is discretized into tiles. There are no restrictions on the shapes and sizes of both tiles and AGVs, hence such a system is suitable for controlling a heterogeneous fleet. Tiles need to be reserved for an AGV before the AGV can move over them. The deadlock avoidance algorithm is called each time the central controller wants to reserve tiles for an AGV; the reservation is only allowed if, within limited known future movements of all AGVs, at least one order of movements exists such that the system remains deadlock-free.

Keywords: Planning, Scheduling and Coordination, Foundations of Automation, Optimization and Optimal Control
Abstract: We present a new Monte Carlo Tree Search (MCTS) algorithm to solve the stochastic orienteering problem with chance constraints, i.e., a version of the problem where travel costs are random and one is given a bound on the tolerable probability of exceeding the budget. The algorithm we present is online and anytime, i.e., it alternates planning and execution and the quality of the solution it produces increases as the allowed computational time increases. Differently from most former MCTS algorithms, for each action available in a state the algorithm maintains estimates of both its value, and the probability that its execution will eventually result in a violation of the chance constraint. Then, at action selection time, our proposed solution prunes away trajectories that are estimated to violate the failure probability. Extensive simulation results show that this novel approach is capable of producing solutions better than former work, while offering an anytime performance.

Keywords: Planning, Scheduling and Coordination, Manufacturing, Maintenance and Supply Chains, Optimization and Optimal Control
Abstract: Job shops are an important production environment for low-volume high-variety manufacturing. Its scheduling has recently been formulated as an integer linear programming (ILP) problem to take advantages of popular mixed-integer linear programming (MILP) methods, e.g., branch-and-cut. With a large number of parts, MILP methods may experience difficulties. To address this, a critical but much overlooked issue is formulation tightening. The idea is that if the constraints can be transformed to directly delineate the convex hull, then a solution can be obtained by using linear programming (LP) methods. The tightening process, however, is fundamentally challenging because of integer variables. In this article, an innovative and systematic approach is established for the first time to tighten the formulations of individual parts, each with multiple operations. It is a major advancement of our previous work on problems with binary and continuous variables to integer variables. The idea is to first link integer variables to binary variables by innovatively combining constraints so that the integer variables are uniquely determined by the binary variables. With binary and continuous variables only, it is proved that the vertices of the convex hull can be obtained based on vertices of the LP problem after relaxing binary requirements. These vertices are then converted to tightened constraints for general use. This approach significantly improves our previous results on tightening individual operations. Numerical results demonstrate significant benefits on solution quality and computational efficiency. This approach also applies to other complex ILP and MILP problems with similar characteristics and fundamentally changes the way how such problems are formulated and solved.

Keywords: Planning, Scheduling and Coordination, Motion and Path Planning, Optimization and Optimal Control
Abstract: This paper studies the problem of rendezvous scheduling between an energy-limited aerial robot (AR) and a mobile ground robot (MGR). If the AR cannot complete its mission in a single sortie, it has to recharge at scheduled locations and times with the MGR acting as a mobile charging station. Differing from previous work, the visit order of the waypoints is assumed to be determined a priori using one of the available algorithms for pathfinding or area coverage. We consider two alternative cases depending on whether the AR can land prior to the arrival of the MGR or it has to hover in the air and wait for its arrival. Our approach to each is motivated by the principle of optimality - namely the corresponding constrained optimization problem is decomposed into smaller problems whose solutions are then integrated together. Their solutions are found from a binary search tree of charge selections that evolves iteratively using a tree sort algorithm. The advantage of this approach is that as the focus is on the scheduling aspect, the algorithm becomes relatively more tractable and realistically applicable. Our simulation results demonstrate that the optimal rendezvous schedule can be determined in realizable time even for large-scale missions.

Keywords: Planning, Scheduling and Coordination, Reinforcement Learning, Deep Learning Methods
Abstract: The travelling salesperson problem (TSP) is a classic resource allocation problem used to find an optimal order of doing a set of tasks while minimizing (or maximizing) an associated objective function. It is widely used in robotics for applications such as planning and scheduling. In this work, we solve TSP for two objectives using reinforcement learning (RL). Often in multi-objective optimization problems, the associated objective functions can be conflicting in nature. In such cases, the optimality is defined in terms of Pareto optimality. A set of these Pareto optimal solutions in the objective space form a Pareto front (or frontier). Each solution has its trade off. We present Pareto frontier approximation network (PA-Net), a network that generates good approximations of the Pareto front for the bi-objective travelling salesperson problem (BTSP). Firstly, BTSP is converted into a constrained optimization problem. We then train our network to solve this constrained problem using the Lagrangian relaxation and policy gradient. With PA-Net we improve the performance over an existing deep RL-based method. The average improvement in hypervolume metric which is used to measure optimality of the Pareto front is 2.3%. At the same time, PA-Net has 4.5× faster inference time. Finally, we present the application of PA-Net to find optimal visiting order in a robotic navigation task/coverage planning.

Keywords: Smart Grids, Plug-in Electric Vehicles, Optimization and Optimal Control
Abstract: Nowadays, power grids are facing reduced total system inertia as traditional generators are phased out in favor of renewable energy sources. This issue is expected to deepen with the increasing penetration of electric vehicles (EVs). The influence of a single EV on power networks is low; nevertheless, the aggregate impact becomes relevant when they are properly coordinated. In this context, we consider the frequent case of a group of EVs connected to a parking lot with a photovoltaic facility. We propose a novel strategy to optimally control their batteries during the parking session, which is able to satisfy their requirements and energy constraints. EVs participate in a noncooperative energy market based on a smart pricing mechanism that is designed in order to increase the predictability and flexibility of the aggregate parking load. Differently from the existing contributions, we employ a novel approach to minimize the degradation of batteries. The effectiveness of the proposed method is validated through numerical experiments based on a real scenario.

Keywords: Deep Learning in Robotics and Automation, Computer Vision in Automation, Agricultural Automation
Abstract: Agricultural datasets for crop row detection are often bound by their limited number of images. This restricts the researchers from developing deep learning based models for precision agricultural tasks involving crop row detection. We suggest the utilization of small real-world datasets along with additional data generated by simulations to yield similar crop row detection performance as that of a model trained with a large real world dataset. Our method could reach the performance of a deep learning based crop row detection model trained with real-world data by using 60% less labelled realworld data. Our model performed well against field variations such as shadows, sunlight and growth stages. We introduce an automated pipeline to generate labelled images for crop row detection in simulation domain. An extensive comparison is done to analyze the contribution of simulated data towards reaching robust crop row detection in various real-world field scenarios.

Keywords: Environment Monitoring and Management, Robotics and Automation in Agriculture and Forestry, Agricultural Automation
Abstract: In this work, we present a Multispectral-Depth (MS-D) system for close range multispectral imaging comprised of a multispectral camera and an RGB-D camera. The proposed system outputs a multispectral point cloud, enabling robust and accurate calculation of numerous vegetation indices. The accuracy of the MS-D system is quantitatively compared with the state-of-the-art methods for close range multispectral imaging. The analysis shows that MS-D outperforms the state-of-the-art approaches with additional advantage of high invariance to the environmental conditions. Contrast to the state-of-the-art feature matching approaches, this system, once calibrated, is applicable as is, without the need for recalibration. The accompanying method for autonomous extrinsic calibration of the system, which extends the capabilities of previous approaches, is also presented.

Keywords: Planning, Scheduling and Coordination, AI-Based Methods, Robotics and Automation in Agriculture and Forestry
Abstract: In this paper, we propose a distributed multi-stage optimization method for planning complex missions for heterogeneous multi-robot systems. This class of problems involves tasks that can be executed in different ways and are associated with cross-schedule dependencies that constrain the schedules of the different robots in the system. The proposed approach involves a multi-objective heuristic search of the mission, represented as a hierarchical tree that defines the mission goal. This procedure outputs several favorable ways to fulfill the mission, which directly feed into the next stage of the method. We utilize a distributed metaheuristic based on evolutionary computation to allocate tasks and generate schedules for the set of chosen decompositions. The solution is evaluated in a simulation setup of an automated greenhouse use case, where we demonstrate the system's ability to adapt the planning strategy depending on the available robots and the given optimization criteria.

Keywords: Process Control, Agricultural Automation
Abstract: The aim of this study is the design and implementation of an automated lighting system to compensate the necessary illumination of a hydroponic indoor-greenhouse in spaces with low illumination. The proposed system implements the Kalman Filter for noise reduction and an anti-windup PID tuned with PSO to reach a desired value of illumination in face of illuminance changes. In addition, an IoT web interface for monitoring the current illumination on plants is implemented, which shows a plot of the illumination throughout the day.

Keywords: Sustainability and Green Automation, Mechatronics in Meso, Micro and Nano Scale, Agricultural Automation
Abstract: The Black Soldier Fly (BSF), HERMETIA ILLUCENS, can be an effective alternative to traditional disposal of food and agricultural waste (biowaste) such as landfills because its larvae are able to quickly transform biowaste into ready-to-use biomass. However, several challenges remain to ensure that BSF farming is economically viable at different scales and can be widely implemented. Manual labor is required to ensure optimal conditions to rear the larvae, from aerating the feeding substrate to monitoring abiotic conditions during the growth cycle. This paper introduces a proof-of-concept automated method of rearing BSF larvae to ensure optimal growing conditions while at the same time reducing manual labor. We retrofit existing BSF rearing bins with a "smart lid," named as such due to the hot-swappable nature of the lid with multiple bins. The system automatically aerates the larvae-diet substrate and provides bio-information of the larvae to users in real time. The proposed solution uses a custom aeration method and an array of sensors to create a soft real time system. Growth of larvae is monitored using thermal imaging and classical computer vision techniques. Experimental testing reveals that our automated approach produces BSF larvae on par with manual techniques.

Keywords: Agricultural Automation, Diagnosis and Prognostics, Deep Learning in Robotics and Automation
Abstract: Accurate objective, automated limb lameness detection and pose estimation play an important role for animal well-being and precision livestock farming. We present a wearable sensor-based limb lameness detection and pose estimation for horse walk and trot locomotion. The gait event and lameness detection are first built on a recurrent neural network (RNN) with long short-term memory (LSTM) cells. Its outcomes are used in the limb pose estimation. A learned low-dimensional motion manifold is parameterized by a phase variable with a Gaussian process dynamic model. We compare the RNN-LSTM-based lameness detection method with a feature-based multi-layer classifier (MLC) and a multi-class classifier (MCC) that are built on support vector machine/K-nearest-neighbors and deep convolutional neural network methods, respectively. Experimental results show that using only accelerometer measurements, the RNN-LSTM-based approach achieves 95% lameness detection accuracy and also outperforms the feature-based MLC or MCC in terms of several assessment criteria. The pose estimation scheme can predict the 24 limb joint angles in the sagittal plane with average errors less than 5 and 10 degs under normal and induced lameness conditions, respectively. The presented work demonstrate the successful use of machine learning techniques for high performance lameness detection and pose estimation in equine science.

Keywords: Robotics and Automation in Construction
Abstract: For infrastructure health monitoring, defects in concrete are periodically inspected by hammering test. Conventional hammering inspection is manually conducted to detect internal concrete defects by recognizing the sounds generated from the hammer strike. Therefore, a robot-based hammering inspection system can realize the inspection automatically. However, those robots have several elements such as noise-resistant acoustic analysis, weight reduction of the robot, constant striking mechanism, and stable robot movement; so integrated research has not been achieved. This paper introduces a hammering inspection system using robot that solves these problems. In the proposed system, a multi-copter type mobile robot realizes a stable attitude on a structure using thrust force to press the robot body onto the structure surface, while a lightweight hammering mechanism constantly strikes a wall. During inspection, noises from multi-copter affect acoustic analysis. Therefore, the proposed method clarifies the features of hammering sound and propeller noise and analyzes hammering sounds. This developed hammering inspection system is the first to realize all functions required for comprehensive hammering inspection, and its accuracy to detect concrete defects from hammering sounds was 81.7% through the experiments.

Keywords: Collision Avoidance, Automation in Construction, Calibration and Identification
Abstract: This paper proposes a real-time collision avoidance system with automatic Light Detection and Ranging (LiDAR) calibration as a means of meeting the increasing demand for safety in construction automation. Although a typical system requires object detection to prevent collisions with obstacles in the workspace, practical safety performance relies heavily on detection accuracy and processing time delays. To achieve both robustness and operational efficiency while increasing safety, we propose a system that determines the possibility of a collision from the observed point cloud and the posture of an excavator without detecting objects. This is achieved by introducing an excavator model synchronized with a real one as a digital twin and evaluating the overlap between the volume occupied by the model and the point cloud observed by the three-dimensional (3D) LiDAR sensor. Moreover, the algorithm to estimate the position and orientation of the 3D LiDAR was developed utilizing a digital twin and the probabilistic sequential estimation technique. The proposed system was successfully demonstrated through experiments using a real excavator, making us confident that deploying the system, from the installation of LiDAR to normal operation, could be fully automated.

Keywords: Building Automation
Abstract: Starting from an application of a real medium size university building, the present paper focuses on the comparison among different ways to synthesize a predictive control scheme to improve the energy performance for the heating, ventilation and air conditioning system of the building. The main motivation is the comparison among a nonlinear predictive control structure previously developed (based on first principle equations) with a predictive control whose prediction model is an artificial neural network. Particular emphasis is given on how to tune the neural network so as to gain good closed loop performance. Twenty-one different networks are designed and tuned in order to correlate their closed loop performance with type and length of training data set, for building energy efficiency applications. Finally, a linear time-variant predictive control is given, obtained as analytical linearization, along the future estimated system trajectory, of the nonlinear equations of the neural network model. The goal is to add to the comparison a low computational burden (linear controller) still derived from nonlinear data-driven methods.

Keywords: Smart Home and City, Sensor Networks, Automation Technologies for Smart Cities
Abstract: To deal with the degraded performance of Global Navigation Satellite Systems (GNSS) in indoor environments, Indoor Positioning Systems (IPS) have been developed. The rapid proliferation of smartphones has led to many IPSs that utilize positioning technologies that are readily available on modern smartphones; including Bluetooth Low Energy (BLE). Using radio signals such as BLE in indoor environments comes with a number of challenges that can limit the reliability of the signal. In dealing with these challenges, most existing BLE-based IPSs introduce undesired drawbacks such as an extensive and fragile calibration phase, strict hardware requirements, and increases in the system’s complexity. In this paper, an IPS is developed and evaluated that requires minimal setup for indoor environments and has a sufficiently low complexity to be run locally on a modern smartphone. An extensive exploration of the IPS’ parameters was performed. The best performing parameter combinations resulted in a median positioning error of 1.48 ± 0.283 meters, while using the log-distance path loss model for distance estimation and Weighted Centroid Localization with a weight exponent between 2.0 and 3.5 for position estimation.

Keywords: Diagnosis and Prognostics, Factory Automation
Abstract: Predictive Maintenance (PdM) is a one of the core topics for Industry 4.0 and entitled as “Predictive Maintenance 4.0.” The main tasks of PdM are to monitor production tool health and then issue an alert when a maintenance is necessary. PdM has become a top priority as it can optimize tool utility. The so-called iFA system platform, realized by integrating several intelligent services including Intelligent Predictive Maintenance (IPM), was proposed to accomplish the goal of Zero-Defect Manufacturing. However, the current algorithm in IPM did not provide a feasible aging feature extraction procedure. Thus, once the aging features cannot be acquired adequately, the monitoring accuracy will become poor. To remedy the above-mentioned problem, the automated Aging Feature Extraction Scheme (AFES) is proposed in this paper to perform analysis of process data for remote health prediction. This automated AFES is packed as an application module and plugged in the cyber physical agent of iFA. The proposed architecture, which integrates iFA, Resource Agent (RA), message broker, and automated Production System, is also designed to effectively monitor tool health status and predict the remaining useful life via the automated AFES. The experimental results indicate that the proposed architecture can not only enhance the performance of the IPM algorithm, but also feed-back the tool health indexes to RA via comprehensive system integration, such that the goal of optimized/maximum OEE can be accomplished. This work was submitted alongside another paper to CASE2022, conceptualizing a data exchange infrastructure and its impact on dependability characteristics of the technical process.

Keywords: Sustainable Production and Service Automation, Energy and Environment-aware Automation
Abstract: The energy saving topic is becoming increasingly important in industry: environmental impact and sustainability of processes are, nowadays, considered critical factors for this field. Manufacturing processes sustainability can be improved by controlling machine state with energy-efficient control policies that switch off/on the device. This approach can be also applied to a workstation composed of parallel and identical machines. This work is focused on this type of configuration, presenting a novel model to identify energy-efficient control policies for two-stage production line with parallel and identical machines. The proposed model reduces energy consumption while assuring a target level on production performance indicators. Numerical results confirm model effectiveness and benefits when implemented on general cases of production systems.

Keywords: Industrial Robots, Factory Automation
Abstract: Autonomous docking is a fundamental requirement for mobile robots working in modern warehouses in order to allow them to operate continuously without human intervention. This paper presents a laser intensity based autonomous docking approach, which uses self-adhesive retro-reflective tapes as reflectors. Sufficient and necessary conditions for successful reflector detection are derived, which provide a systematic guidance for deployment of laser intensity based reflectors. A laser intensity based reflector detection technique was proposed, allowing for reliable identification of artificial landmarks in dynamic and unstructured environments. To demonstrate the application of the proposed approaches, extensive experiments were conducted to control a differential drive wheeled mobile robot for autonomous docking and recharging.

Keywords: Industrial Robots, Factory Automation
Abstract: This study proposes a method for generating the optimum grasp position of a robotic hand. First, the shape of the object was approximated as multi-primitive to simplified. Subsequently, two parallel planes are extracted from the multi-primitive object, various grasping positions were computed for the two planes, and grasping evaluations were performed for the grasping direction to generate the optimum grasping position. As a result of the experiment, we found that 10 grasps were successfully obtained with seven types of object and posture.

Keywords: Industrial and Service Robotics, Intelligent and Flexible Manufacturing, Computer Vision in Automation
Abstract: Autonomous robotic grasping is a fundamental skill for the next generation of robots. It is a challenging problem as it requires many steps to succeed, ranging from detecting the target location to selecting the grasp pose configuration to have a stable grasp. Grasping fragile objects or objects with variable shapes is a more complex task with respect to the traditional pick and place of solid objects, and robots typically do not have a reliable sense of touch. In such cases, a retention action, which can be obtained by electro-adhesion, is typically preferred over a compression force in order to avoid damaging the objects. The proposed work presents a robotic manipulation grasping system that leverages a gripper realized with the flexible thin-film electro-adhesive (EA) devices technology and a vision pipeline based on an RGB-D camera to detect the grasp pose configuration and track the target during the holding phase to check whether the task has been successfully completed. Thanks to the properties of the EA gripper, vision is the only perception cue needed to successfully grasp the target without damaging it since the gripper can automatically adapt its shape to the surface of the target delicately wrapping its two fingers around the object. Several tests have been done to assess the capabilities of the proposed robotic system, picking and placing deformable objects, comparing the EA gripper with a traditional parallel jaw gripper. The gripper is particularly suitable to handle objects where two parallel flat surfaces are available for grasping. Future works will attempt to improve the grasping of non flat objects.

Keywords: Industrial and Service Robotics, Robust/Adaptive Control
Abstract: This paper presents the implementation of a robust control technique applied within a standard multi-objective robot controller for humanoid robots. The challenge is to regulate the angular position of a seesaw actuated by a humanoid on it, which is a complex multitask involving the humanoid balance, the motion to project its weight into its feet to move the seesaw, and the seesaw motion itself. An accurate model of the seesaw is not available, therefore, a model-free approach based on super twisting sliding mode control is proposed; besides, the stability of the closed-loop system with the proposed technique is proved using the Lyapunov's direct method. The proposed controller was applied to a multi-robot model compound by the HRP-4 humanoid and a passive seesaw model, and it was tested using a physics-engine simulation environment. The presented results shows that the challenge of regulating the seesaw was successfully achieved by means of the proposed robust control.

Keywords: Medical Robots and Systems
Abstract: Vitreoretinal surgery consists of procedures carried out on the retina. It requires extreme precision and presents challenges due to the very small available workspace. For these reasons, it is a prime candidate for robot-assisted surgery, which enables sub-millimeter precision. During surgery, it is necessary for surgical instruments to be pivoted about trocars while also dealing with eye movement, which will in turn move the said trocar. To ensure that no damaging forces are exerted on the eye, the instrument's Remote Centre of Motion (RCM) has to be estimated and the trocar tracked. This short paper focuses on the RCM estimation aspect. A state-of-the-art RCM estimation method consisting of an Extended Kalman Filter (EKF) was applied, and its suitability for this application was investigated. The method required the instrument's pose data, which was supplied using an electro-magnetic sensor. For experimental validation of the setup, a customised calibration procedure was developed. Results showed a maximum median absolute error of 3.5mm. These results show promise, but necessitate further investigation to meet the requirement of an absolute error below 1.4mm, which corresponds to the trocar's radius. Future work will ensure that the majority of the error is within this margin, and limit its maximum attained value.

Keywords: Telerobotics and Teleoperation, Medical Robots and Systems, Virtual Reality and Interfaces
Abstract: Remote telesurgery can enable expert surgeons to operate on patients in distant or underserved locales. However, network instability and delays hamper long-distance communication. To address this, we explore how a “digital twin,” a 3D simulator that actively mirrors a real environment, can be applied to telesurgery. We focus on the Fundamentals of Laparoscopic Surgery peg transfer surgical training task. We present a framework that enables a teleoperator to perform this task over unstable or low-bandwidth communication channels using a digital twin. The surgeon remotely teleoperates the robot in our simulator, which abstracts their motions into commands and transmits them to the real robot for semi-autonomous execution. The system executes the transfer and then sends the real state of the pegboard back to the simulator. We present experiments that demonstrate that the operation of each portion of the framework in isolation maintains a high task success rate, and that the success rate of the digital twin framework is robust to network transmission instability and delays.

Keywords: Computer Vision in Automation, Autonomous Agents, Agent-Based Systems
Abstract: Object-based SLAM has been widely studied in recent years. While many research works focus on representing objects as ellipsoids, the initialization is still an open problem. The traditional approach using 2D detected bounding boxes from multiple views fails in straight motion scenarios, because the object observation is limited to a few frames. In this work, we present a novel approach to recover object pose from a single RGB image by decoupling the translation, orientation and dimension. Under the assumption that the object dimension is already known, the translation can be approximately computed by back-projection of 2D bounding box center. Finally, the optimal rotation angle is calculated to satisfy the bounding-box constraints and image lines alignment. The proposed method shows an efficient solution to leverage the 2D information for 3D object initialization. Besides, the initialized object is integrated into a simultaneous localization and mapping system (SLAM) for a further optimization. We evaluate our novel initialization method and object-based SLAM system on public indoor and outdoor dataset, the results show that our system reconstructs the object with a higher accuracy.

Keywords: Computer Vision in Automation, Autonomous Agents, Collision Avoidance
Abstract: We propose a novel flow synthesis based visual servoing framework enabling long-range obstacle avoidance for Micro Air Vehicles (MAV) flying amongst tall skyscrapers. Recently deep learning based frameworks use optical flow to do high precision visual servoing. In this paper, we explore the question: can we design a surrogate flow for these high precision visual-servoing methods which leads to obstacle avoidance? We revisit the concept of saliency for identifying high-rise structures in/close to the line of attack of amongst other competing skyscrapers and buildings as a collision obstacle. A synthesised flow is used to displace the salient object segmentation mask. This flow is so computed that the visual servoing controller maneuvers the MAV safely aroung the obstacle. In this approach, we use a multi-step Cross-Entropy Method (CEM) based servo control to achieve flow convergence, resulting in obstacle avoidance. We use this novel pipeline to successfully and persistently maneuver high-rises and reach the goal in photo-realistic and simulated real-world scenes. We conduct extensive experimentation and compare our approach with optical flow and short-range depth-based obstacle avoidance methods to demonstrate the proposed framework's merit. Additional Visualisation and code can be found at https://sites.google.com/view/monocular-obstacle-avoidance/ home

Keywords: Computer Vision in Automation, Autonomous Agents, Industrial and Service Robotics
Abstract: Loop closure can effectively eliminate the accumulated error in Simultaneous Localization and Mapping (SLAM). Appearance-based localization methods tend to fail under large viewpoint changes. In this paper, we propose a monocular SLAM system with object-based loop closure against viewpoint variation to achieve global localization. Objects are represented as cuboids and inferred from 2D object observation. On this basis, we construct a semantic topology graph from the object-oriented map and propose an efficient graph matching method with a directional histogram descriptor to detect the loop. Objects are matched if they satisfy general, graph and geometry verifications. By aligning the matched objects, the accumulated errors can be corrected, and the map can be updated. Experimental results demonstrate that the proposed method shows high accuracy and robustness under large viewpoint differences.

Keywords: Computer Vision in Automation, Diagnosis and Prognostics, AI-Based Methods
Abstract: Material attribution is an integral part of product life cycle management. In the apparel fashion industry, material attribution activities are error prone because of their manual and monotonic nature. As a part of intelligent process automation for material attribution, we are proposing a model that uses deep neural networks to automate the classification of apparels based on attributes such as gender, category, subcategory, and color, when an image of an apparel is passed to the model. Our model assures process improvement by accurately extracting all the attributes in one go by using a computationally efficient algorithm that also minimizes the carbon footprint.

Keywords: Computer Vision in Automation, Industrial and Service Robotics, Calibration and Identification
Abstract: Cameras becomes more ubiquitous on mobile devices which pave the way for Augment Reality (AR) applications. AR's enabling technology is the underlying visual-inertial Simultaneous Localization and Mapping (SLAM) package which requires a precise camera model for mapping purpose. Due to manufacturing inconsistency and device aging over time, the preciseness is often hard to maintain over time. On the other hand, those cameras are often equipped with optical image stabilization (OIS) system. OIS changes camera intrinsic parameters and being aware of its existence is important before a high-order SLAM model is applied. Here we present a two-step approach to detect if an image conforms to a given camera model (distortion coefficient and intrinsic matrix) by developing two statistical hypothesis testings. We have implemented the algorithm and test it in physical experiments. Results show that our algorithm successfully detects model inconsistency and the existence of OIS system with 85.4% recall and 100% precision.

Keywords: Computer Vision for Automation, AI-Based Methods, Big Data in Robotics and Automation
Abstract: We propose a rotated bounding box (RBB) detector without the annotation of object orientation that detects an axis-aligned bounding box (AABB) from rotated images. Both AABB and RBB detection are fundamental technologies used in robotics applications. When focusing on robots that perform picking tasks, the RBB is required rather than the AABB. However, the annotation costs of RBBs are large compared with those of AABBs, and datasets for RBBs are quite scarcer as well. To remove the necessity of annotating rotation information and to utilize existing AABB datasets, we consider a method for obtaining RBBs using only AABB data without annotation for rotation information. First, the proposed method generates sequential images by rotating an input image and tracks estimated target AABBs among the sequentially-rotated images using an AABB detector for each target object. Then, it selects the AABB with the smallest area among the tracked AABBs and inversely rotates the selected AABB using the image rotation angle to obtain the RBB for the target object. We confirmed that the proposed method can estimate RBBs with high accuracy on a real dataset containing daily objects with simple shapes.

Keywords: Deep Learning Methods
Abstract: Localization of robots is vital for navigation and path planning, such as in crises where a map of the environment is needed. The use of Ultra-Wideband (UWB) for indoor location systems have been gaining popularity over the years with the introduction of low-cost UWB modules providing centimetre-level accuracy. However, in the presence of obstacles in the environment, Non-Line-Of-Sight (NLOS) measurements from the UWB will produce inaccurate results. Low-cost UWB does not provide channel information that could be used to determine NLOS conditions. We propose an approach to decide if a measurement is within Line-Of-Sight (LOS) or not by using some signal strength information provided by the UWB modules through a Neural Network (NN) model. The results of this model were then used for localization through Weighted-Least-Square (WLS) method to improve localization results. By identifying the probability of ranging measurements being LOS and including them in WLS localization, we improve the accuracy by 16.93% on the lobby testing data and 27.97% on the corridor testing data using the NN model trained with all extracted inputs from the office training data.

Keywords: Reinforcement, Deep Learning in Robotics and Automation, Autonomous Agents
Abstract: In modern robotics, effectively computing optimal control policies under dynamically varying environments poses substantial challenges to the off-the-shelf parametric policy gradient methods, such as the Deep Deterministic Policy Gradient (DDPG) and Twin Delayed Deep Deterministic policy gradient (TD3). In this paper, we propose a systematic methodology to dynamically learn a sequence of optimal control policies non-parametrically, while autonomously adapting with the constantly changing environment dynamics. Specifically, our non-parametric kernel-based methodology embeds a policy distribution as the features in a non-decreasing Euclidean space, therefore allowing its search space to be defined as a very high (possible infinite) dimensional RKHS (Reproducing Kernel Hilbert Space). Moreover, by leveraging the similarity metric computed in RKHS, we augmented our non-parametric learning with the technique of AdaptiveH— adaptively selecting a time-frame window of finishing the optimal part of whole action-sequence sampled on some preceding observed state. To validate our proposed approach, we conducted extensive experiments with multiple classic benchmarks and one simulated robotics benchmark equipped with dynamically changing environments. Overall, our methodology has outperformed the well-established DDPG and TD3 methodology by a sizeable margin in terms of learning performance.

Keywords: Reinforcement, Manufacturing, Maintenance and Supply Chains, Optimization and Optimal Control
Abstract: Multi-echelon supply chains (SC) are highly complex systems with many inherent uncertainties, including customer demands and transportation time from one location to another. In addition to the complex SC structures, inventory decisions made at different stages affect each other. Maintaining global-level optimal inventory levels along the entire supply chain that is robust to changing business needs remains challenging. In this work, we have developed a simulation environment, GymSC, for reinforcement learning of multi-echelon SC inventory policies. Using a series of model-free deep reinforcement learning algorithms, we trained dynamic SC policies that have significantly improved performance when compared with popularly used heuristics. The robustness of the learned policy is demonstrated by its adaptability to a previously unseen environment with non-stationary customer demands. The presented work showcases the effectiveness and robustness of deep reinforcement learning in solving important practical SC optimization problems. In addition, we present this configurable simulation environment as a platform for testing existing and new algorithms to develop robust inventory policies with a goal to encourage further integration of deep reinforcement learning into the SC management problems as a promising alternative for addressing inventory optimization with uncertainties.

Keywords: Autonomous Agents, Deep Learning in Robotics and Automation, Probability and Statistical Methods
Abstract: This paper describes a framework for the object- goal navigation (ObjectNav) task, which requires a robot to find and move to an instance of a target object class from a random starting position. The framework uses a history of robot trajectories to learn a Spatial Relational Graph (SRG) and Graph Convolutional Network (GCN)-based embeddings for the likelihood of proximity of different semantically-labeled regions and the occurrence of different object classes in these regions. To locate a target object instance during evaluation, the robot uses Bayesian inference and the SRG to estimate the visible regions, and uses the learned GCN embeddings to rank visible regions and select the region to explore next. This approach is tested using the Matterport3D (MP3D) benchmark dataset of indoor scenes in AI Habitat, a visually realistic simulation environment, to report substantial performance improvement in comparison with state-of-the-art baselines.

Keywords: Deep Learning in Robotics and Automation, Autonomous Vehicle Navigation, Intelligent and Flexible Manufacturing
Abstract: We propose a novel factory automation method that simultaneously optimizes a logical factory production line, such as inventory and production amount, and the physical path planning of robot vehicles. Traditionally, path planning for robot vehicles and overall factory line optimization have been studied independently. However, actual factory production lines require a use case for path planning that considers the balance between inventory control, production maximization, and coordination with assembly workers. Therefore, we developed a novel approach for the mobile simulation of a logistic-physical factory line and devised a deep reinforcement learning method based on high-level rewards. This method is capable of sequential path planning when considering the balance between the inventory and product number as well as the coordination of agents among the production lines. Consequently, our mobile agent successfully learns to plan the shortest route without a bottleneck in the factory production line during any given episode. Moreover, the mobile agent appropriately adjusts the route when a bottleneck occurs and inventory is excessive. This suggests that path planning for robot vehicle agents can be achieved based on indicators that optimize the entire factory, which we expect to be a novel application of robot vehicle coordination that operates along the entire production line.

Keywords: Deep Learning in Robotics and Automation, Reinforcement, Assembly
Abstract: This paper investigates the advantages and bound- aries of actor-critic reinforcement learning algorithms in an industrial setting. We compare and discuss Cycle of Learning, Deep Deterministic Policy Gradient and Twin Delayed Deep Deterministic Policy Gradient with respect to performance in simulation as well as on a real robot setup. Furthermore, it emphasizes the importance and potential of combining demon- strated expert behavior with the actor-critic reinforcement learning setting while using it with an admittance controller to solve an industrial assembly task. Cycle of Learning and Twin Delayed Deep Deterministic Policy Gradient showed to be equally usable in simulation, while Cycle of Learning proved to be best on a real world application due to the behavior cloning loss that enables the agent to learn rapidly. The results also demonstrated that it is a necessity to incorporate an admittance controller in order to transfer the learned behavior to a real robot.

Keywords: Motion and Path Planning, Energy and Environment-aware Automation
Abstract: A waypoint trajectory is different from a continuous-path trajectory, e.g., a circular trajectory since the path between the waypoints is not constrained. However, continuous-path trajectories suffer from a major limitation, which is the inability to set any desired kinematic constraints on the trajectory other than the ones defined by the standard equation of the continuous-path trajectory, e.g., the circle equation. Therefore, this paper proposes the kinematically-constrained continuous-path polynomial trajectory to overcome this limitation. First, a framework to generate polynomial trajectories of any degree with an arbitrary number of waypoints in any dimensional space is presented. Second, a dynamic feasibility condition of the proposed trajectory for quadrotors applications is introduced. Third, the required quadrotor's thrust is compared to the corresponding continuous-path trajectory to examine if the proposed trajectory requires larger thrust and consequently more energy consumption. Fourth, two different cases with arbitrary kinematic constraints and a piecewise kinematic profile are studied via simulation to show the effectiveness of the proposed approach. Finally, the results are validated experimentally. The results show effectiveness and the feasibility of the proposed approach.

Keywords: Motion and Path Planning, Motion Control, Intelligent Transportation Systems
Abstract: This paper presents an approach for smooth trajectory planning in semi-rigid nonholonomic mobile robot formations using Bezi´er-splines. Unlike most existing approaches, the focus is on maintaining a semi-rigid formation, as required in many scenarios such as object transport, handling or assembly. We use a Relaxed A* planner to create an optimal collision-free global path and then smooth this path using splines. The smoothed global path serves to create target paths for every robot in the formation. From these paths, we then calculate the trajectories for each robot. In an iterative process, we match the velocities of the robots so that all trajectories are synchronized, and the dynamic limits of all robots are maintained. We provide experimental validation, which confirms no violation of the dynamic limits and shows an excellent control performance for a system of three robots moving at 0.3 m/s.

Keywords: Motion and Path Planning, Optimization and Optimal Control, Robust/Adaptive Control
Abstract: The actual paper proposes a robust optimal control strategy via an Extended-State Integral Linear Quadratic Gaussian (ES-iLQG) controller meant to drive the quadrotor motion to track a time-parametrized trajectory in presence of exogenous and endogenous disturbances. The herein enhanced LQG controller, includes an Extended-State Linear Kalman Filter (ES-LKF) utilised as a disturbance estimator, and an integral Linear Quadratic Regulator (iLQR). Results from a simulation stage exhibit the effectiveness of the proposed control scheme for trajectory tracking purposes. In this regard, promising experimental results were obtained from two scenarios: Trajectory tracking of an elliptical helix-shaped and an 8-shaped trajectories. It is noteworthy that the control law is computed onboard and relies on optical flow for translational motion control.

Keywords: Motion and Path Planning, Surveillance Systems, Reactive and Sensor-Based Planning
Abstract: Coverage Path Planning (CPP) requires planning collision-free paths for a robot that observes all reachable points of interest in an environment. Most popular CPP approaches are hierarchical and decomposition-based, involving three steps: (1) decomposing the environment into sub-regions (rectangles or polygons) that simplify the generation of space-filling paths, (2) determining a visitation order over these sub-regions via graph search or a Traveling Salesman Problem (TSP) solver, and (3) generation of space-filling paths in each sub-region. This approach requires significant processing of the environment and the availability of suitable TSP solvers. Furthermore, step (1) can sometimes fail in non-convex environments or lead to “over-decomposition” in cluttered environments. To the best of our knowledge, existing decomposition-free approaches are heuristic or random, and therefore typically inefficient and probabilistically complete. We present a resolution-complete decomposition-free coverage path planner that effectively folds steps (1) and (2) above into a single online search routine, making it significantly easier to integrate into existing robot architectures and applicable to a larger set of environments. Our approach leverages a precomputed library of space-filling coverage patterns and automatically determines where to apply them. We evaluate our approach on a variety of environments to demonstrate its benefits and provide an open-source implementation at https://github.com/ktushar14/cdf_cpp.

Keywords: Swarms, Collision Avoidance, Motion and Path Planning
Abstract: This paper describes an efficacious virtual-region-based shape control scheme for exploring an unknown occluded dynamic environment, likely to have multiple targets, by a swarm of robots. The traditional leader-follower strategy has been intertwined with the shape control technique to achieve group-splitting of the robotic swarm in multi-target or multi-path scenarios. If multiple passages are detected by the parent swarm during the navigational process, several sub-swarms with proportionate agents will be created for progressing through these separate paths. The splitting philosophy is dependent on the fitment of the virtual elliptical regions in each of the paths found ahead, which will guide their respective sub-swarms to navigate further. This work, subsequently, conceives a realistic condition where two or more pathways merge to form a single lane, leading to a unique target. In such situations, the sub-swarms (operating in those paths) rejoin and proceed to the goal as a unified unit. Therefore, in this work, two different features of a swarm robotics framework, namely textit{splitting and merging}, have been studied. The scalability control plan ensures strict cohesiveness amongst the agents (inside the mother swarm or any sub-swarm) during the exploration steps. To substantiate the efficacy of the proposed technique, simulation results along with hardware experiment are duly furnished in this article.

Keywords: Agent-Based Systems, Collision Avoidance, Motion and Path Planning
Abstract: Path smoothing is an important operation in a number of path planning applications. While several approaches have been proposed in the literature, a lack of simple and effective methods with quality-based termination conditions can be observed. Among the many available methods, the traditional random shortcuts approach represents a popular and effective solution. However its random selection of shortcuts may miss tight areas difficult to be sampled. This may lead to sharp corners in tight areas making it difficult to achieve termination conditions based on path quality.

In this paper we propose a method that overcomes these limitations and is able to include user-specified termination conditions based on solution quality. At each iteration, our method first identifies a vertex on the path that has the most potential for smoothing, and then applies one of two possible shortcut-based smoothing operation. As a result, our prioritized shortcut selection and quality-based termination conditions result in a method that outperforms the random shortcuts approach both in path length and worst-case angle. We present several benchmarks demonstrating that, for the same amount of smoothing time, our method produces higher-quality paths when compared to the traditional random shortcuts approach.

Keywords: Intelligent and Flexible Manufacturing, Agent-Based Systems, Planning, Scheduling and Coordination
Abstract: Due to the advancements in manufacturing system technology and the ever-increasing demand for personalized products, there is a growing desire to improve the flexibility of manufacturing systems. Multi-agent control is one strategy that has been proposed to address this challenge. The multi-agent control strategy relies on the decision making and cooperation of a number of intelligent software agents to control and coordinate various components on the shop floor. One of the most important agents for this control strategy is the product agent, which is the decision maker for a single part in the manufacturing system. To improve the flexibility and adaptability of the product agent and its control strategy, this work proposes a direct and active cooperation framework for the product agent. The directly and actively cooperating product agent can identify and actively negotiate scheduling constraints with other agents in the system. A new modeling formalism, based on priced timed automata, and an optimization-based decision making strategy are proposed as part of the framework. Two simulation case studies showcase how direct and active cooperation can be used to improve the flexibility and performance of manufacturing systems.

Keywords: Intelligent and Flexible Manufacturing, Factory Automation, Assembly
Abstract: With increasing demand for productivity, flexibility, and sustainability, there is the need for a flexible manufacturing system that is auto-reconfigurable for variations in product types and assembly processes. However, the repeatability of reconfigurable components needs to be controlled and quantified in order to achieve the critical product tolerances required. High levels of repeatability for reconfigurable components is often achieved by a lengthy calibration. Besides, automated processes would rely on the precise tool and part positioning or an adaptive process approach. In this paper, an adaptive, highly repeatable and rapid auto-reconfiguration process in a smart manufacturing environment is proposed for small box product assembly, such as rudders, elevators and winglets. The process involves a reconfigurable tooling system for physically supporting different products, robots and end effectors to perform automated processes, programmable logic controllers to orchestrate cell safety and robotic tasks, an autonomous guided vehicle (AGV) to provide jig mobility, and a metrology system to realise cell-level positional layout. The rapid reconfigurable tooling system was tested and quantified for repeatability and configuration time, and the adaptive auto-reconfiguration process was validated by moving the jig frame in a lab environment simulating inaccurate AGV parking. The repeatability of profile board positioning can achieve a value smaller than +/-0.04mm, with an estimated between-product changeover time less than 10 minutes. With an external metrology system, the positional layout of the cell was captured and used to adapt robot programs. Successful engagement was observed, proving the feasibility of the adaptive process.

Keywords: Intelligent and Flexible Manufacturing, Factory Automation, Reinforcement
Abstract: The automated guided vehicle (AGV), a typical form of automated material handling system, generally utilizes electric power from an internally mounted battery pack. AGVs need to occasionally visit a battery station and swap the battery to manage their state of charge. An AGV system therefore needs a textit{swapping policy}, which determines when a vehicle should proceed to a battery station for battery replacement. In real industrial practice, most swapping policies are conservative and are based heuristically on the experiences of decision makers, which results in production inefficiency. The objective of this research is to develop a swapping strategy to improve the AGV system production efficiency. The proposed swapping policy is based on sequential decisions that consider current and future situations, and utilizes a Markov decision process framework and deep reinforcement learning. We present the results of numerical experiments to demonstrate the superior performance of the proposed swapping policy compared with heuristic policies. We also analyze the properties of the proposed swapping policy, and the results demonstrate its application potential for AGV systems.

Keywords: Intelligent and Flexible Manufacturing, Learning and Adaptive Systems, Reinforcement
Abstract: Robot skills systems are meant to reduce robot setup time for new manufacturing tasks. Yet, for dexterous, contact-rich tasks, it is often difficult to find the right skill parameters. One strategy is to learn these parameters by allowing the robot system to learn directly on the task. For a learning problem, a robot operator can typically specify the type and range of values of the parameters. Nevertheless, given their prior experience, robot operators should be able to help the learning process further by providing educated guesses about where in the parameter space potential optimal solutions could be found. Interestingly, such prior knowledge is not exploited in current robot learning frameworks. We introduce an approach that combines user priors and Bayesian optimization to allow fast optimization of robot industrial tasks at robot deployment time. We evaluate our method on three tasks that are learned in simulation as well as on two tasks that are learned directly on a real robot system. Additionally, we transfer knowledge from the corresponding simulation tasks by automatically constructing priors from well-performing configurations for learning on the real system. To handle potentially contradicting task objectives, the tasks are modeled as multi-objective problems. Our results show that operator priors, both user-specified and transferred, vastly accelerate the discovery of rich Pareto fronts, and typically produce final performance far superior to proposed baselines.

Keywords: Intelligent and Flexible Manufacturing, Manipulation Planning, Compliance and Impedance Control
Abstract: We propose a new approach to programming by demonstration of finishing operations. Such operations can be carried out by industrial robots in multiple ways because an industrial robot is typically functionally redundant with respect to a finishing task. In the proposed system, a human expert demonstrates a finishing operation and the demonstrated motion is recorded in Cartesian space. The robot's kinematic model is augmented with a virtual mechanism, defined according to the applied finishing tool. This way the kinematic model is expanded with additional degrees of freedom that can be exploited to compute the optimal joint space motion of the robot without altering the essential aspects of Cartesian space task execution as demonstrated by the human expert. Additionally, we propose a novel approach for the accurate estimation of the contact points between the machining tool and the workpiece using the measured forces and torques. We applied iterative learning control to refine the demonstrated operations and compensate for inaccurate calibration and different dynamics of the robot and human demonstrator. The proposed method was verified on real robots and real polishing and grinding tasks.

Keywords: Intelligent and Flexible Manufacturing
Abstract: There are many industrial robotic applications which require a manipulator's end-effector to fully cover a 3D surface region in a constrained motion, such as painting, spray coating, abrasive blasting, polishing, shotcreting, etc. The manipulator must satisfy surface task constraints imposed on the end-effector while maintaining manipulator joint constraints. Coverage path planning (CPP) in this context generally involves placing commonly used coverage patterns (such as raster, spiral, or dual-spiral) onto the surface. There is substantial research for CPP on 2D surfaces, however, the problem of generating surface task constraints and evenly spaced coverage paths becomes particularly difficult when considering 3D freeform surfaces. Previous research concerning CPP on 3D surfaces consider parametric surfaces with limited surface curvature or produce unevenly spaced coverage paths. In this paper, we introduce a novel method to generate a uv grid on a 3D freeform surface (represented as a 3D polygon mesh) to facilitate feasibility checking for constrained coverage motion under task and manipulator constraints and to significantly ease the creation of more evenly spaced coverage paths for optimal application of task requirements. We applied our method to example 3D freeform surfaces to demonstrate its effectiveness.

Keywords: Machine learning, AI and Machine Learning in Healthcare
Abstract: Classification models have been widely applied to detect anomalies in many real-world fields, such as manufacturing process monitoring and disease early detection. However, the classification models may suffer from the data imbalance issue, as the abnormal/unhealthy states are usually rare events in regular data collection. Imbalanced data may result in significant training bias, leading to unsatisfactory classification accuracy. Incorporating data augmentation techniques, such as the popular generative adversarial networks (GAN), is a common strategy to eliminate the data imbalanced issue in classification. However, the performance of most GAN-based approaches may be unsatisfactory when the size of available training samples is small. To address this issue in GAN, the paper develops a novel collaborative discrimination-enabled GAN (CoD-GAN) to enhance its discrimination robustness. With the proposed collaborative discrimination framework, CoD-GAN is able to perform discrimination more effectively when the available real data is limited. Thus, the synthesized samples will be more effective, and the classification accuracy can be improved. The effectiveness of the proposed CoD-GAN has been validated by both numerical simulation data and real-world dataset. The results have demonstrated that the proposed method can further improve the data augmentation capability of GAN for imbalanced data classification.

Keywords: Deep Learning in Robotics and Automation
Abstract: This paper proposes a new control framework for manipulating soft objects. A Deep Reinforcement Learning (DRL) approach is used to make the shape of a deformable object reach a set of desired points by controlling a robotic arm which manipulates it. Our framework is more easily generalizable than existing ones: it can work directly with different initial and desired final shapes without need for relearning. We achieve this by using learning parallelization, i.e., executing multiple agents in parallel on various environment instances. We focus our study on deformable linear objects. These objects are interesting in industrial and agricultural domains, yet their manipulation with robots, especially in 3D workspaces, remains challenging. We simulate the entire environment, i.e., the soft object and the robot, for the training and the testing using PyBullet and OpenAI Gym. We use a combination of state-of-the-art DRL techniques, the main ingredient being a training approach for the learning agent (i.e., the robot) based on Deep Deterministic Policy Gradient (DDPG). Our simulation results support the usefulness and enhanced generality of the proposed approach.

Keywords: Machine learning, Probability and Statistical Methods, Semiconductor Manufacturing
Abstract: In semiconductor manufacturing, virtual metrology (VM) is a method of predicting physical measurements of wafer qualities using in-process information from sensors on production equipment. The relevance vector machine (RVM) is a sparse Bayesian kernel machine that has been widely used for VM modeling in semiconductor manufacturing. Missing values from equipment sensors, however, preclude training an RVM model due to missing kernels from incomplete instances. Moreover, imputation for such kernels can lead to a loss of model sparsity. In this work, we propose a restricted RVM (RRVM) that selects its basis functions from only complete instances to handle incomplete data for VM. We conduct the experiments using toy data and real-life data from an etching process for wafer fabrication. The results indicate the model’s competitive prediction accuracy with massive missing data while maintaining model sparsity.

Keywords: Probability and Statistical Methods, Big-Data and Data Mining, Modelling, Simulation and Optimization in Healthcare
Abstract: Understanding the underlying component structure is crucial for multivariate signal analysis. Among all the techniques that try to learn the latent structure, independent component analysis (ICA) is one of the most important and popular methods, which aims to extract independent components from multivariate signals and enables further analysis. For example, in electroencephalogram (EEG) analysis, artifacts filtering and disease detection are conducted based on the independent components of the signals. One critical challenge in existing ICA approaches is that the component extraction accuracy may degrade when the available data of a unit are limited. To address this issue, this paper proposes a transfer learning-based ICA method by innovatively transferring component distribution from a source domain, so that accurate component extraction results can be achieved even when only limited data are available in the target domain. To the best of our knowledge, this is the first work that leverages transfer learning to improve ICA accuracy with limited available data. In particular, we first extract all the independent components from the source domain by maximizing the log-likelihood function with a Newton-like method on a smooth manifold. Then for the target domain, the component with the largest negentropy is extracted in each round. To effectively leverage the knowledge from the source domain and to prevent the negative transfer, we try to find a component in the source domain that matches the component we are extracting. The probability density function of the matched component will then be used to improve the component extraction accuracy if such matched component can be found; otherwise, no knowledge will be transferred. Numerical simulations and a case study with elect

Keywords: AI and Machine Learning in Healthcare, Health Care Management, Probability and Statistical Methods
Abstract: Patient-reported outcome (PRO) measures are increasingly collected as a means of measuring healthcare quality and value. The capability to predict such measures would enable patient-provider shared decision making and the delivery of patient-centered care. However, PRO measures recorded in the electronic health record often suffer from a high missing rate, and the missingness may depend on many covariates. Under such a complex missing mechanism, statistical inference of the parameters in a model for predicting PRO measures is challenging. In this work, we propose to use an informative surrogate to efficiently infer the parameters of interest without estimating the complex missing mechanism. It is expected that an informative surrogate can lead to a semiparametric imputation model lying in a low-dimensional subspace. Although such subspace can be inferred using dimension reduction methods, the initial parameter estimation might still suffer from the bias due to the nonparametric component in the imputation estimation. To remove the bias, we further identify and estimate a low-dimensional weighting function as an alternative to the traditional propensity score, which is difficult to obtain due to the complex missing mechanism. Based on the imputation model and the weighting function, we construct a one-step debiased estimator without using any information of the true missing propensity. We establish the asymptotic normality of the one-step debiased estimator, which is valid even when the marginal missing rate approaches 1. We provide real data analysis to demonstrate the superiority of the proposed method.

Keywords: AI and Machine Learning in Healthcare, Data fusion, Diagnosis and Prognostics
Abstract: Technology advancements have enabled a wealth of health information to be remotely collected, resulting in increasing use of telemonitoring for patients with chronic diseases. In particular, multi-channel bio-signals such as ECG and EEG, gold-standard diagnostic approaches for many diseases, are able to be collected at home. To utilize multi-channel bio-signals for telemonitoring and telemedicine, it is critical to develop a rigorous prediction model that “translates” the monitored symptomatic signals into a clinical indicator of disease severity to facilitate disease monitoring and diagnosis. However, multi-level and multi-channel data pose major challenges for most statistical prediction methods. To address these challenges, this paper proposes a multi-level multi-channel framework to integrate multi-channel epoch-level bio-signals and other patient-level health covariates for precise prediction of disease severity in health telemonitoring. The proposed framework was applied in a real-world dataset consisting of 409 patients and achieved the highest prediction accuracy in comparison with benchmark methods that do not take account of the multi-level structure of features into prediction. The findings of this study demonstrate the efficacy and performance of the proposed method in predicting disease severity from multi-level multi-channel features, which contributes a novel predictive analytical tool to facilitate bio-signal-based health telemonitoring.

Keywords: Learning and Adaptive Systems, Reinforcement, Machine learning
Abstract: Intention inference of autonomous vehicles is crucial to guarantee safety and to mitigate risk. This paper reports a performance objective extraction from expert's data trajectories for experience transference and to uncover the hidden cost associated to the intent. The algorithm is inspired in the hippocampus learning system for experience exploitation that exhibits the human brain. The hippocampus is responsible of memory and to store past experiences to enable transfer learning and fast convergence.

The proposed algorithm extracts, from expert's data, the performance matrices associated to a hidden utility function using a complementary approach based on an off-policy policy iteration and a matrix extraction inverse reinforcement learning algorithms. Exact performance extraction is obtained by adding a constraint in terms of the measurements of the utility function in a batch-least squares algorithm. Convergence of the proposed approach is verified using Lyapunov recursions. Simulation studies are carried out to demonstrate the effectiveness of the proposed approach.

Keywords: Failure Detection and Recovery, Deep Learning Methods, Manufacturing, Maintenance and Supply Chains
Abstract: Intelligent machine condition monitoring (CM) for automatic fault diagnosis relies on data-driven algorithms to characterize machine health for predictive maintenance activities on smart factory floors. Since data collection can be expensive, CM data sets may not cover all the possible fault conditions, necessitating that CM algorithms continually learn new conditions. State-of-the-art CM research has focused on detecting unknown conditions rather than integrating unknown conditions into future predictions. Therefore, CM-ready Continual Learning (CL) solutions should learn to classify new conditions and use improved representations that minimize the need for future fine-tuning. Meta-learning approaches like Few-Shot Prototypical Networks (FSPN) regularize base-task learning to find these more generalizable representations. Experiments on a motor data set demonstrate that FSPN with only 5 or 10 examples of the novel fault consistently outperforms static, fine-tuning, and Elastic Weight Consolidation (EWC) approaches for CL, increasing the overall accuracy by up to 19 points (53% to 72%). Compared to recent FSPN work for image classification, these results show that FSPN may be advantageous for CM due to the limited class diversity of CM data sets. Future work should extend the FSPN architecture to include open set recognition and quantitatively analyze varying numbers of base-task classes.

Keywords: Reinforcement Learning, Control Architectures and Programming, Agent-Based Systems
Abstract: Multi-agent system control is a research topic that has broad applications ranging from multi-robot cooperation to distributed sensor networks. Reinforcement learning is shown to be promising as a control strategy in cases where the dynamics of the agents are non-linear, complex, and highly uncertain since it can learn policies from samples without using much model information. The presented manuscript proposes a multi-agent decentralized control approach based on a new multi-agent reinforcement learning setting in which two virtual agents, sharing the same environment, control a single avatar but have access to complementary details necessary to finish the task. Each of them is responsible for solving a portion of the problem, and in order to efficiently solve it, a collaboration should emerge among the virtual agents not to compete but to focus on the final goal. Each virtual agent, performing individually, is not fully autonomous since it does not have a complete vision of the scene and needs the other one to properly command the avatar. The proposed approach proved to be able to solve efficiently constrained navigation problems in two different simulated setups. An actor-critic architecture with a Proximal Policy Optimization (PPO) algorithm has been employed in continuous action and state spaces. The training and the testing have been done in a maze-like environment designed using the StarCraft II Learning Environment.

Keywords: Model Learning for Control, Sensor-based Control, Deep Learning in Robotics and Automation
Abstract: Recent advances in latent space dynamics model from pixels show promising progress in vision-based model predictive control (MPC). However, executing MPC in real time can be challenging due to its intensive computational cost in each timestep. We propose to introduce additional learning objectives to enforce that the learned latent space is proportional derivative controllable. In execution time, the simple PD-controller can be applied directly to the latent space encoded from pixels, to produce simple and effective control to systems with visual observations. We show that our method outperforms baseline methods to produce robust goal reaching and trajectory tracking in various environments.

Keywords: Intelligent Transportation Systems, Machine learning, Big-Data and Data Mining
Abstract: Automated detection of anomalous trajectories is an important problem with considerable applications in intelligent transportation systems. Many existing studies have focused on distinguishing anomalous trajectories from normal trajectories, ignoring the large differences between anomalous trajectories. A recent study has made great progress in identifying abnormal trajectory patterns and proposed a two-stage algorithm for anomalous trajectory detection and classification (ATDC). This algorithm has excellent performance but suffers from a few limitations, such as high time complexity and poor interpretation. Here, we present a careful theoretical and empirical analysis of the ATDC algorithm, showing that the calculation of anomaly scores in both stages can be simplified, and that the second stage of the algorithm is much more important than the first stage. Hence, we develop a FastATDC algorithm that introduces a random sampling strategy in both stages. Experimental results show that FastATDC is 10 to 20 times faster than ATDC on real datasets. Moreover, FastATDC surpasses the baseline algorithms and is comparable to the ATDC algorithm.

Keywords: Human-Centered Automation, Agent-Based Systems, AI-Based Methods
Abstract: As higher education grows increasingly complex, educators will inevitably run into students who require more in-depth support to succeed in their learning. However, time constraints and strained resources limit the ability of educators to provide personalized tutoring, especially when dealing with large classes. Adaptive serious games offer a solution to this issue. By creating complex student models, these educational games can model and predict student knowledge and behavior. Then, building on those models, adaptive serious games can offer tailored support to students that address their specific needs. In adaptive serious games, there is a lack of generalized systems that model and direct student progress through an educational game. In this area, we demonstrate a generalized approach to student modelling and directing through the use of Petri nets. We also extend the Petri net model with automated, adaptive student support using reinforcement learning agents.

Keywords: Reinforcement, AI-Based Methods, Machine learning
Abstract: Motivated by a real problem of service mechanism in the output port of routers, this paper studies the dynamic bandwidth allocation in a G=G=1=K parallel queueing system, where weighted fair queueing (WFQ) scheduling discipline is applied to support differentiated services for different packet queues. The bursty and complicated characteristics of Internet traffic pose a challenge on the analytic solution for dynamic bandwidth allocation, which requires distributional information of traffic patterns. Since the distributional information of Internet traffic is always unavailable and varied with time, we propose a deep reinforcement learning (DRL) framework to train a bandwidth controller by adaptively interacting with the environment. The controller dynamically allocates bandwidth weights among multiple queues according to the instant queue lengths observed. We train the controller with two advanced DRL algorithms, DDPG and SAC, respectively. With real traffic data, experiment results show that our trained controllers achieve a lower average delay and packet loss rate than a rule-based policy. Our proposed WFQ-DRL algorithm is a first attempt to apply RL algorithms in real scenarios of routers, where the system has eight or more queues and a diversity of real traffic without Poisson assumption is applicable.

Keywords: Logistics, Simulation and Animation, Planning, Scheduling and Coordination
Abstract: With the ever-increasing international container trade and e-commerce, container terminals face the biggest challenge in terms of container management and sustainability. For this reason, they have turned to the latest technology to upgrade their facilities and achieve higher levels of efficiency. The BoxBay project, an innovative high-bay storage system, is one of the most recent breakthroughs in container terminals. In this paper, the BoxBay project is being used as an example to investigate the different control policies that can be implemented to improve the efficiency of similar systems. Policies for task sequence, storage, and dwell points are being reviewed. Then, to estimate the minimum possible energy consumption for the system to be efficient, a mathematical model for the energy consumption of an automated unit load vehicle is being developed. Finally, a simulation study takes place to better analyze the system's performance. Both the simulation results and the implementation of the proposed policies demonstrate that the performance of high-bay storage systems in terminals could be promising for the future and capable of achieving high levels of efficiency.

Keywords: Deep Learning in Robotics and Automation, AI-Based Methods, Computer Vision in Automation
Abstract: Artificial neural network (ANN) has been widely used in automation. However, the vulnerability of ANN under certain attacks poses security threat for critical automation systems. Adding noises to artificial neural network has been shown to be able to improve robustness in previous work. In this work, we propose a new technique to compute the pathwise stochastic gradient estimate with respect to the standard deviation of the Gaussian noise added to each neuron of the ANN. By our proposed technique, the gradient estimation with respect to noise levels is a byproduct of the back propagation algorithm for estimating gradient with respect to synaptic weights in ANN. Thus, the noise level for each neuron can be optimized simultaneously in the processing of training the synaptic weights at nearly no extra computational cost. In numerical experiments, our proposed method can achieve significant performance improvement on robustness of several popular ANN structures under both black box and white box attacks.

Keywords: AI-Based Methods, Inventory Management, Logistics
Abstract: In this work, we investigate an integrated optimization problem of inventory placement and transportation vehicle selection in a logistics system with multiple central distribution centers and multiple regional distribution centers. The main decision in our problem refers to the selection of transportation vehicles, concerning the trade-offs among different types of costs in the system, such as the vehicle selection cost, commodity transportation cost and inventory holding cost. We formulate the problem as a non-convex mixed-integer quadratically constrained program. Due to the non-convexity of the objective function which makes the model difficult to solve, we establish a convex approximation on the proposed formulation using Cauthy inequalities. An efficient two-phase solution framework, combining neural network prediction and branch-and-bound search, is developed to solve the approximate model. Computational results demonstrate that using a neural network is effective in predicting values of a subset of integer variables in solution, which can be subsequently extended to form a high-quality solution to the integrated optimization. Moreover, the two-phase method has a significant advantage in solving speed over the pure implementation of branch-and-bound method, which suggests its strength in solving larger mixed-integer programs.

Keywords: Big-Data and Data Mining, AI-Based Methods, Machine learning
Abstract: When designing a wrapper-based feature selection algorithm for regression problems, the choice dilemma between computational cost and accuracy is one focal and difficult point. In this paper, we analyze the characteristics of wrapper-based feature selection algorithm for regression problems. We find that a rough estimate method could be used when designing the selection operator of genetic algorithm (GA). So, we can effectively reduce the computation cost while keeping GA’s global search capability and wrapper methods’ high accuracy to the greatest extent. The method is designed under the guidance of ordinal optimization. A wrapper-based feature selection algorithm named ordinal genetic algorithm-based feature selection (OGAFS) is designed, and convergence analysis of OGAFS is given. Finally, simulations on three regression problems show that the algorithm is effective.

Keywords: AI-Based Methods, Big-Data and Data Mining, Failure Detection and Recovery
Abstract: A CNC machine may contain various types of faults and may harm production safety in multiple ways. It is difficult for managers to determine which faults are more severe due to the complex structure of the machine. This further confuses managers to generate efficient preventions and responses. Consequently, there is a need to develop a method to assist production managers to identify the safety-critical components (SCCs) in a CNC machine. The current work developed a solid framework for SCCs examination using a risk-based knowledge graph method. Firstly, we summarize the structure of the CNC machine using ontology and examine CNC faults and corresponding consequences. The reasoning between faults and consequences is via literature examination using a web-crawler. This is followed by using BiLSTM-CRF for knowledge processing and using TransE for entity alignment. Then, a risk-based knowledge graph is developed via developing a weight scale for the severity of consequence; and using historical fault data to access frequency. The knowledge graph visualization is developed using Gephi, and SCCs can be virtually determined with deeper color in nodes as the darker color represents higher risk. This has the potential to help industry practitioners better understand their current operations situation and generate responses to avoid fault occurrence and improve production safety.

Keywords: Discrete Event Dynamic Automation Systems, Petri Nets for Automation Control, Optimization and Optimal Control
Abstract: This paper proposes a method to design an optimal supervisor for a Petri net model with two different control specifications. First, according to the two control specifications, an integer linear programming problem is formulated to compute two control places, i.e., one control place is designed for each specification. They have the same structure and data inhibitor arcs but different initial markings. Then, a specification switcher is introduced to change the number of tokens in the control place. By the specification switcher and the two control places, an optimal supervisor is obtained that can satisfy the two control specifications. Finally, by using the proposed method, the net model is optimally controlled under the two specifications.

Keywords: Semiconductor Manufacturing, Planning, Scheduling and Coordination, Discrete Event Dynamic Automation Systems
Abstract: In semiconductor wafer fabrication, it is incredibly challenging to operate a multi-cluster tool that is composed of several single-cluster tools. This paper deals with the scheduling problem of a time-constrained single-arm-robot multi-cluster tool. We first propose a characteristic single-arm-robot cluster tool and investigate its cyclic scheduling problem. Then, we devise a feature transformation method. By using such a novel approach, a single-arm-robot multi-cluster tool with arbitrary topology can be easily reduced into a characteristic single-arm-robot cluster tool. Thus, the robot tasks are reassigned and then a feasible cyclic schedule can be obtained if the system is schedulable. Finally, we provide illustrative experiments to validate the practicability and availability of the proposed approach.

Keywords: Discrete Event Dynamic Automation Systems, Petri Nets for Automation Control, Optimization and Optimal Control
Abstract: Nowadays, high throughput screening (HTS) systems are widely used in pharmaceutical industries and laboratories for discovery of new drugs and biomedical substances. It is important to efficiently schedule them so as to reduce the cost. In the operation of an HTS system, there are complex microplate flows and also time window constraints are imposed on some activities and activity sequences. Moreover, with the consistency requirement, a one-microplate cyclic schedule is necessary. Thus, its scheduling problem is very challenging, and the existing studies apply mathematical programming methods to tackle this problem. This paper studies the scheduling problem of an HTS system for an enzymatic assay, a typical application of HTSs, from the perspective of control theory. It is modeled by a Petri net model. With the model, a closed-from solution method is proposed to find a feasible and optimal cyclic schedule, which shows that polynomial algorithms for an optimal schedule exist.

Keywords: Discrete Event Dynamic Automation Systems, Petri Nets for Automation Control, Robust/Adaptive Control
Abstract: This paper develops a robust deadlock control method for automated manufacturing systems (AMSs) with unreliable resources. The considered AMS is modeled by a generalized system of simple sequential processes with resources (GS^3PR). The notion of improved recovery subnets is developed to model a resource failure scenario where if a failure occurs,the damaged resource is removed for repair and the part is returned to the system for the subsequent production. Based on a reachability graph partition technique that can compute all forbidden markings and robust legal ones, an invariant-based robust supervisory control strategy is developed to guarantee the former reachable and the latter unreachable. It is verified that the designed robust controller can assure the liveness of the controlled system regardless of whether resource failures occur. Finally, we present two examples to show the validity of the proposed method.

Keywords: Planning, Scheduling and Coordination, Intelligent and Flexible Manufacturing, Petri Nets for Automation Control
Abstract: This work deals with a scheduling problem of flexible manufacturing systems (FMSs) with a Petri net. We propose a branch and bound (B&B) algorithm for scheduling of FMSs. The experimental results show that the proposed B&B algorithm can solve various FMS instances faster than the mathematical model solved with CPLEX and previous B&B algorithms.

Keywords: Intelligent and Flexible Manufacturing, Petri Nets for Automation Control, Reinforcement
Abstract: We consider a dynamic job shop scheduling problem where jobs arrive dynamically and are transported by automated guided vehicles (AGVs). The objective is to minimize the makespan. We model the problem with a timed Petri net (TPN) and modify the transition enabling rule to consider jobs that have not arrived yet. We propose a Monte Carlo Tree Search (MCTS)-based integrated scheduling algorithm with the TPN to use future state information in making a decision. The proposed method shows better performance than other dispatching rule combinations in various scenarios.

Keywords: Sensor Fusion, Deep Learning in Robotics and Automation, Robotics and Automation in Life Sciences
Abstract: Mobile robots need reliable maps for autonomous operation. Traditional SLAM systems, which are mainly developed for static scenes, often fail in dynamic environments with moving objects present in the scene. Learning based dynamic SLAM systems often suffer from insufficient or inaccurate identification of feature points. This paper proposes a novel real-time RGB-D SLAM system, which is targeted for dynamic environments, can further enhance feature detection and dynamic removal. This is done by fusing panoptic segmentation and geometry information. The system includes four components: dynamic segmentation that reduces the impact of moving objects, pose estimation with dynamic object removal, panoptic mapping, and loop closing. The pose estimation uses coarse-to-fine dynamic/static classification to further reduce the impact of unknown moving objects. Extensive evaluations demonstrate that our system can achieve robust performance in complex dynamic environments.

Keywords: Learning and Adaptive Systems, Reinforcement, Probability and Statistical Methods
Abstract: We study a dynamic pricing and capacity sizing problem in a M/GI/1 queue, where the service provider’s objective is to obtain the optimal service fee p and service capacity mu so as to maximize cumulative expected profit (the service revenue minus the staffing cost and delay penalty) in long term. Motivative by real application, we consider the setting in which the demand curve, or customers' sensitivity to price, is unknown to the system manager.

In this work we propose an online learning algorithm to solve this problem. In particular, our approach is model-free in the sense that we don't require any knowledge on the form of the demand function except for some regularity condition. We show that our algorithm is near optimal in the sense that it obtains a regret upper bound which meets the worst-case lower bound except for some logarithmic term. In the numerical experiments, we show that our algorithm is efficient for a variety fo M/GI/1 models and outperforms the traditional heavy-traffic approach and some model-based learning algorithms.

Keywords: Additive Manufacturing, AI-Based Methods, Deep Learning in Robotics and Automation
Abstract: In additive manufacturing (AM), accurate prediction for the deformation of printed objects contributes to compensation in advance, which is crucial to improving the accuracy of products. Many factors affect the deformation, such as the shape of the object, the properties of the material, and parameters in the printing process. Existing methods suffer from difficulties in modeling and generalizing between different shapes. In this paper, we formulate the error prediction in AM as a point-wise deviation prediction task and propose a point-based deep neural network to learn the complex deformation patterns by local and global contextual feature extraction. Furthermore, a data processing flow is proposed for automatically handling the real-scenario data. As an application case, we collect a dataset of dental crowns fabricated by the digital light processing 3D printing and validate the proposed method on the dataset. The results show that our network has a promising ability to predict nonlinear deformation. The proposed method can also be applied to other AM techniques directly.

Keywords: AI-Based Methods, Deep Learning in Robotics and Automation, Computer Vision for Manufacturing
Abstract: Particle defects on the cathodic copper plate surface always happen due to the immaturity of electrolytic copper processing. The removal of defects mainly depends on their height exceeding the plate and current removal requires manual measurement and operation, which is time-consuming and laborious. To automate the removal process, machine vision-based defect detection methods need to be developed. However, copper defects are of very small size, which increases the difficulty of feature extraction and prediction. Therefore, this paper proposes a novel Anchor-based Detection and Height Estimation (ADHE) framework, to locate the defect out and estimate the height of the defect in an end-to-end way. Large-scale raw images are transformed into several image blocks as input. Defect features are obtained by Defect Region Extraction Network and then sent into Height-RCNN for defect detection and height prediction. Dataset of cathodic copper plate surface defects has been collected from a real-world manufacturing factory. Experimental results show that the proposed ADHE method can effectively address the small size problem of copper defects and achieve excellent results in detection and height estimation.

Keywords: AI-Based Methods, Sensor Fusion, Human Factors and Human-in-the-Loop
Abstract: Lacking a certain level of activity is associated with multiple health issues, and activity management is significant for office workers who sit 77% of working hours. Recent studies on the Internet of Things spur the advent of applications for daily activity management. However, there is no activity management system collecting data unobtrusively and continuously, which provides activity recognition and assessment for office workers. Hence, this study develops a multimodal activity management system based on an infrared array sensor placed on the desk, a sensing chair, and a mobile phone. This system contains data collection, activity recognition, and activity assessment. A deep learning algorithm based on the feature-level fusion strategy is leveraged to fuse the multimodal activity data and achieve recognition. The activity assessment considers energy expenditure and sedentary bout to reflect office workers’ activity characteristics. Finally, an experiment is conducted to verify the feasibility of the proposed system. The results show that recognition accuracy can reach 99.9% and 84.5% by using the validation set approach and leave-one-subject-out cross-validation approach, respectively.

Keywords: AI and Machine Learning in Healthcare
Abstract: In recent years, online medical platforms have developed rapidly. Due to the differences in cognition of medical knowledge, some suitable doctors being eliminated in the recall stage. In order to expand the doctor retrieval to reduce search costs for users, it is necessary to get disease words embeddings. In this paper, we propose a new algorithm to get rid of the limitations of existing medical word representation learning, and successfully apply it to Baidu Health, one of the largest online medical platforms in China.

Keywords: AI and Machine Learning in Healthcare, Health Care Management, Machine learning
Abstract: Smart healthcare is changing our lives. As an emerging medical pattern, online medical platform is arising from the combination of traditional medical resources and Internet platform, which largely resolve the disequilibrium of offline medical resources in China. Compared with offline healthcare, online platforms shorten the distance between patients and medical resources and give patients more options to seek medical treatment during the COVID-19 epidemic. In order to better help and guide patients in making decisions, the platform provides physicians' treatment information for patients' reference. This information describes the physician's diagnostic capability and service level from different dimensions, such as the physician's specialty, the number of gifts received from patients, etc., which are important basis for patients to choose a physician. For the platform and physicians, it is crucial to understand patients' preferences for different characteristics of physicians in the consultation process, in order to manage data more targeted. This paper use machine learning methods to build a prediction model of physician's characteristics data on incremental volume of consultation to study patients' preferences in medical consultation. Most existing studies use linear models, but given the complexity of patient preferences, they may have greater limitations in reflecting patients' choice logic. Therefore, this paper turn to more complex models on the training data. For the lack of interpretation of complex models, this paper uses a Shapley Value-based approach to parse the model's feature contributions to obtain patients' preferences for physician information. From the perspectives of local interpretation, global interpretation and interaction effect, this paper

Keywords: Scheduling in Healthcare, Modelling, Simulation and Optimization in Healthcare, Health Care Management
Abstract: Our paper addresses a weekly physician scheduling problem in COVID-19. This problem has arisen in fever clinics in two collaborative hospitals located in Shanghai, China. Because of the coronavirus pandemic, the hospitals must consider some specific constraints in the scheduling problem. For example, due to social distance limitation, the patient queue lengths are much longer in the coronavirus pandemic, even with the same waiting patients. Thus, the hospitals must consider the maximum queue length in the physician scheduling problem. Moreover, the fever clinic’s scheduling rules are different from those in the common clinic, and some specific regulatory constraints have to be considered in the epidemic. We first build a mathematical model for this problem, in which a pointwise stationary fluid flow approximation method is used to compute the queue length. Some linearization techniques are designed to make the problem can be solved by commercial solvers such as Gurobi. We find solving this model from practical applications of the hospital within an acceptable computation time is challenging. Consequently, we develop an efficient two-phase approach to solve the problem. A staffing model and a branch-and-price algorithm are proposed in this approach. The performances of our models and approaches are discussed. The effectiveness of the proposed algorithms for real-life data from collaborative hospitals is validated.

Keywords: Scheduling in Healthcare
Abstract: This paper considers a setting where the OT manages the ORs and conducts OR planning and scheduling under stochastic surgery durations in a fixed planning horizon. We propose a stochastic mixed-integer program (SMIP) to determine the number of opening ORs, the date of surgeries, the allocation of ORs to surgeries, and the scheduling of appointment time of corresponding patients and surgeons simultaneously. The objective we consider is to minimize the total cost of OR operations, which includes (1) the fixed cost of opening ORs, (2) the cost of surgeon and patient's waiting, (3) cost due to OR idling, and (4) cost of OR's overtime using. Due to the stochastic nature of the problem, to achieve the tradeoff between modeling accuracy and computational tractability, we use sample average approximation (SAA) to reformulate our model and to derive a mixed-integer program (MIP) reformulation. To tackle the resulting large-scale MIP, we develop a problem-tailored branch-and-price (BP) algorithm and a fast heuristic within the column generation framework for generating suboptimal solutions. Furthermore, we design two speedup strategies for the BP algorithm. Finally, we conduct numerical experiments to explore the effect of different sequences of surgeries on ORs operational performance.

Keywords: Optimization and Optimal Control, Modelling, Simulation and Optimization in Healthcare
Abstract: Many decision problems in manufacturing involves the performance optimization of discrete-event dynamic systems. Due to the stochasticity of these systems, the performances are usually evaluated via simulation and the optimization problem can be modeled as a ranking and selection problem. However, a large design space will make the computation too expensive. This research considers the problem of selecting the best design from a large domain. The true mean of each design is unknown in practice, but it can be estimated via simulation. We divide the entire design domain into several adjacent partitions, and the performances of designs in each partition are estimated via a regression metamodel. The underlying function in each partition is assumed to be quadratic or approximately quadratic. The utilization of the regression metamodel can help us to incorporate the information from across the entire design domain, thus enhancing the simulation efficiency dramatically. In this research, we use the expected opportunity cost (EOC) to measure the quality of the selection. The EOC penalizes a particularly bad selection more than a mildly bad selection, and is thus preferred by risk-neutral decision makers. We formulate an optimization problem to determine the optimal simulation budget allocation rule among all partitions and within each partition so that the EOC of incorrect selection can be minimized when the limited simulation budgets are all consumed. Using the large deviation theory, we derive the asymptotical optimality conditions of the optimization problem, and develop an asymptotically optimal budget allocation rule to allocate the simulation budget intelligently. A sequential allocation algorithm is proposed to implement the allocation rule. The EOC measure is compared w

Keywords: Health Care Management, Logistics, Planning, Scheduling and Coordination
Abstract: Operating room (OR) is one of the most critical units in a hospital. Managing surgical processes in ORs for better utilization of medical resources, safe delivery of surgical cases, improving patient outcome, and reducing cost, is of significant importance. In this paper, a discrete-event simulation model of OR workflow in a Tertiary A hospital in Beijing, China, is developed. The model is validated by comparing with three-month data collected in the hospital. Using this model, the impacts of OR capacity, resource level, design and operation policies are investigated.

Keywords: Computer Vision for Manufacturing, Assembly, Failure Detection and Recovery
Abstract: To improve assembly quality and efficiency, a method based on deep learning and object matching is proposed to detect missing and wrong parts. An improved YOLO V3 neural network is designed to solve the problem of missing assembly. A small target detection scale and attention module is added to the neural network. The size of prior anchor box is optimized by K-means++ clustering algorithm. For the problem of wrong assembly, the standard assembly state detection template is constructed according to the virtual assembly scene in CAD software, and the 2D detection box of the current assembly object in the scene image is matched with the 2D box in the standard state template based on IoU (Intersection over Union) calculation. The assembly model MONA (a 3D model for the evaluation of manual Assembly tasks), is used to test the proposed method. Experimental results show that this method can accurately locate and identify assembly parts, and effectively detect the missing and wrong parts in the assembly process.

Keywords: Cyber-physical Production Systems and Industry 4.0
Abstract: Batch processing is common in mass manufacturing industries due to its high productivity, such as aerospace, semiconductor, and automotive. Based on the analysis of a Bernoulli production line consisting of a discrete machine and a batch machine, this paper investigates a general scenario of batch-batch lines. Utilizing a systems approach with theoretic and experimental analysis, the system properties, such as the monotonicity and reversibility, are analyzed, and the impact of batch size on the performance is studied. Based on them, continuous improvements of the batch-batch lines are explored which include constrained improvement and bottleneck analysis. This paper provides an effective method to analyze batch-batch manufacturing systems, which is also a building block for studying multi-machine systems.

Keywords: Discrete Event Dynamic Automation Systems, Modelling, Simulation and Validation of Cyber-physical Energy Systems
Abstract: Performance evaluation of telecommunication systems is challenging due to the complexity and large-scale of this problem. However, research on obtaining network measures is still scarce and difficult. Majority of previous research mainly applies the traditional simulation method to solve this problem. But in fact, the result of such traditional simulation methods may suffer from many shortcomings. To solve this problem, we adopt the max-plus method to the telecommunication system and establish simulation processes, inspired by the success of using max-plus algebra on the simulation of the production line. In this paper, we focus on the simulation of an origin-destination flow telecommunication system. The efficiency and accuracy of the proposed method are verified by comparative simulation with other simulation methods. On the premise of ensuring the accuracy of simulation, our method performs 20 times faster than traditional event scheduling method. Moreover, we analyze simulation results under different scheduling strategies to illustrate the accuracy of the simulation.

Keywords: Logistics, Plug-in Electric Vehicles, Smart Home and City
Abstract: In recent years, many countries have introduced policies to reduce carbon emissions, and people also pay more attention to the reduction of carbon emissions. At present, a large number of gas-based motor vehicles emit a lot of carbon dioxide. Thus, replacing them with electric vehicles (EVs) is regarded as an important measure to reduce carbon emissions, and electric vehicles are favored by increasing number of customers. However, there are still many problems in the use of EVs, such as short mileage, slow charging speed, few charging facilities and so on. All these bring anxiety and inconvenience to people when driving EVs. In order to alleviate these problems, many countries are building more charging facilities for EVs, and enterprises are also developing battery with larger capacity and longer mileage, and faster charging technology. At the same time, the valet charging service is a promising way for the successful application of EVs. The new service mode makes it possible for the customers to hire a valet to pick up the EV, drive the EV to the charging station, and return the vehicle after charging. To analyze the valet charging service, a mixed integer programming model is formulated. This problem is similar to the multi-depot multi-trip vehicle routing problem with time windows. Then we transform it into a set covering problem, and a branch-and-price with label-setting algorithm is designed to tackle this problem. In addition, we design a heuristic to provide better initial solutions for the restricted subproblem to speed up the solution speed.

Keywords: Motion and Path Planning, Intelligent Transportation Systems
Abstract: The capacitated vehicle routing problem with two-dimensional loading and time windows (2L-CVRPTW) is an extension of the vehicle routing problem, in which customers have the requirements of time windows and two-dimensional loading constraints. It aims at planning a shortest trip that can visit all customers and satisfy customers' demands, using a series of homogeneous vehicles. In this paper, we propose a multi-stage algorithm to solve the 2L-CVRPTW of customers whose geographical data is clustered (C-type). To accelerate the solving process, we merge the multiple customers into a few customers. Then, an improved ant colony algorithm incorporated with the skyline heuristic algorithm is used to solve the shortest route of merged customers. The results show that the algorithm can reduce the size of the points well for the C-type dataset, and obtain a solution closing to the optimal solution.

Keywords: Sustainable Production and Service Automation, Intelligent and Flexible Manufacturing, Factory Automation
Abstract: Green manufacturing is of significant importance to meet the strategic goals of carbon peak and carbon neutral. Reducing energy consumption in manufacturing plays a critical role to achieve green objectives, particularly for production lines with energy-intensive setup processes. In this paper, a two-machine line model with setups is introduced. In addition to deriving analytical method to evaluate productivity and energy performance in such lines, an optimization model to minimize energy consumption while still maintaining desired production rate is introduced. Using this model, system properties and design policies to optimize line performance are presented. Such a model provides a quantitative tool to analyze and optimize energy and productivity performance in manufacturing systems.