Keywords: Mobile Robots, Dynamics and Control
Abstract: Legged manipulators, such as quadrupeds equipped with robotic arms, require motion planning techniques that account for their complex kinematic constraints in order to perform manipulation tasks both safely and effectively. However, trajectory optimization methods often face challenges due to the hybrid dynamics introduced by contact discontinuities, and tend to neglect leg limitations during planning for computational reasons. In this work, we propose RAKOMO, a path optimization technique that integrates the strengths of K-Order Markov Optimization (KOMO) with a kinematically-aware criterion based on the reachable region defined as reachability margin. We leverage a neural-network to predict the margin and optimize it by incorporating it in the standard KOMO formulation. This approach enables rapid convergence of gradient-based motion planning -- commonly tailored for continuous systems -- while adapting it effectively to legged manipulators, successfully executing loco-manipulation tasks. We benchmark RAKOMO against a baseline KOMO approach through a set of simulations for pick-and-place tasks with the HyQReal quadruped robot equipped with a Kinova Gen3 robotic arm.

Keywords: Humanoid Robots, Learning and Adaptation, Dynamics and Control
Abstract: Balance control remains a fundamental challenge in humanoid robotics due to the underactuated, high-dimensional, and dynamically unstable nature of bipedal systems. Conventional control methods, typically relying on simplified dynamic models, offer limited adaptability and require extensive manual tuning, particularly under unpredictable external disturbances. In this work, we propose a data-driven framework for push recovery of the humanoid robot Talos using deep reinforcement learning (DRL). A key contribution of our approach lies in the reward function, which incorporates principles from capture point and divergent component of motion (DCM) theory to encourage stable and human-like balance strategies. By training under a broad distribution of perturbations, the learned policy autonomously discovers a spectrum of recovery behaviors, including ankle, hip, and stepping responses, without access to explicit model dynamics. We further evaluate the policy’s robustness under previously unseen perturbations to assess generalization. Results demonstrate that our method enables fast convergence, diverse strategy deployment, and strong resilience to unexpected disturbances, highlighting the efficacy of physically informed reward shaping in DRL-based humanoid control.

Keywords: Learning and Adaptation, Dynamics and Control
Abstract: Legged robots must exhibit robust and agile locomotion across diverse, unstructured terrains, a challenge exacerbated under blind locomotion settings where terrain information is unavailable. This work introduces a hierarchical reinforcement learning framework that leverages terrain-specialized policies and curriculum learning to enhance agility and tracking performance in complex environments. We validated our method on simulation, where our approach outperforms a generalist policy by up to 16% in success rate and achieves lower tracking errors as the velocity target increases, particularly on low-friction and discontinuous terrains, demonstrating superior adaptability and robustness across mixed-terrain scenarios.

Keywords: Rehabilitation Robotics
Abstract: Harmonic drive gears are compact and lightweight reducers, typically more complex and expensive than other traditional gearing. However, their complexity introduces a disadvantage on systems that might suffer from external forces causing overload, leading to a failure known as ratcheting. This phenomenon results in improper tooth meshing between its components and can ultimately lock the actuator. This paper presents the design and development of a semi-planar passive mechanism developed to prevent the locking of the harmonic drive reducer used in an existing active ankle-foot prosthesis. The prosthesis has previously experienced ratcheting events, and the proposed mechanism aims to protect the actuator from momentary torque peaks exceeding the reducer’s rated capacity. The design of the anti-lock mechanism weighs 253 g, is able to rigidly withhold 54.4 Nm and has a torsional stiffness of 217 Nm/rad. Finite Element Method analysis was used to evaluate and ensure the structural integrity of the custom machined parts.

Keywords: Human-Robot Interaction, Rehabilitation Robotics, Dynamics and Control
Abstract: Wearable robots assist humans in tasks while naturally following the user’s intended movements, a characteristic known as transparency. Achieving high transparency requires accurate control of interaction forces, which are challenging to measure directly using sensors. Hence, the literature suggests using estimation methods to enable an effective human-robot interaction interface. This study compares a Kalman Filter and a Monte Carlo based estimation method within a transparency control framework. We performed experiments on the Impedance Control in 2 Dimensions (IC2D) test bench, which simulates the coupling between the human body and the robot. We assess each method’s accuracy against our groundtruth data and discuss the results, contributing to the development of more transparent and responsive control strategies.

Keywords: Rehabilitation Robotics, Dynamics and Control
Abstract: For patients with lower limb disabilities, the primary goal of self-care rehabilitation is to restore their limb functions, where robotic knee prostheses act as the primary functional component. In this paper, it is presented the design of a mechanism suitable for the implementation of a series elastic actuator on a polycentric knee prosthesis. It was used a commercial polycentric knee joint composed by a four bar mechanism and a pneumatic damping system. For the test of the knee joint prosthesis, it was executed a flexion-extension cycle in order to obtain the response of the sensors. For the test of the Series Elastic Actuator output, the knee joint was blocked with a rigid bar to view the signal provided by the elastic component deformation. During the flexion-extension test, it had been found some differences between the displacement values for the two cycle phases. For the testing of the elastic element, it was observed that the deformation values relative to the pulses sent to the mechanism's drive were organized linearly. These results are essential for the implementation of a force control system on the mechanism.

Keywords: Multi-Robot Systems, Robotics Architectures, Cooperation and Competition
Abstract: Non-cooperative high-speed space debris has the potential to cause severe damage to functioning systems in space. A scalable, compact gripper could provide a solution for the removal of such inactive satellites from low Earth orbit. This paper proposes the design of a scalable, lightweight gripper called consTorq using a constant-torque underactuated mechanism based on the caging principle to secure debris much larger than itself, without breaking them into smaller pieces. ConsTorq achieves this by making more than two contact points with the debris per arm. To overcome scalability and versatility challenges, this research demonstrates the application of a Genetic Algorithm for geometry optimization reducing the mass and form factor of the proposed design. This leads to a mass reduction of 57.76% in simulations. Furthermore, inverse kinetics is analysed to validate the structural rigidity and synchronicity of the gripper arm movements. Ultimately, the proposed design is scaled down by 90% for rapid prototyping and real-world testing. The initial caging concept with multiple contact points is validated, achieving a reasonable power grasp of 3 N. This demonstrated the intended objective of avoiding the Kessler syndrome by grasping larger objects reliably.

Keywords: Dynamics and Control, Human-Robot Interaction
Abstract: Interaction control has become increasingly important with the rise of human-robot collaboration. Among various strategies, impedance and admittance control stand out by regulating the relationship between velocity and force, rather than individual system states. Impedance control offers stability in stiff environments but may lack precision in free space, while admittance control performs well in non-contact tasks but can be unstable with stiff environments. Since stiff environments exhibit high admittance, impedance control is generally preferred, and vice versa. To overcome the limitations of each method, hybrid systems combining both approaches have been developed, typically categorized as switch-based or non-switch-based. This study proposes a comparative framework for such hybrids using a novel visualization method: the impedance space. New metrics are introduced to assess how well these systems achieve the target impedance, enabling quantitative performance evaluation across varying environmental conditions.

Keywords: Dynamics and Control, Robot Operating Systems, Mobile Robots
Abstract: This work presents the implementation of a Sliding Mode Controller (SMC) for a mobile manipulator, considering both a conventional version with a linear PI sliding surface and a proposed nonlinear variant (SMC+NLn). The nonlinear approach introduces an error-dependent gain in the discontinuous component of the control law to mitigate chattering effects. The dynamic model of the mobile manipulator was obtained by combining the kinematics of the robotic arm and the dynamics of the mobile base. Both controllers were tuned using genetic optimization algorithms, minimizing standard performance indices. The evaluation was carried out through simulations involving a reference trajectory that excites all degrees of freedom, as well as external disturbances emulated by variations in the Jacobian matrix. The results show that the SMC+NLn outperforms the conventional SMC in all tested scenarios, achieving lower tracking errors, reduced control effort, and significantly attenuated oscillations in the discontinuous component. In disturbance rejection tests, the nonlinear controller exhibited faster recovery and smoother transient response while maintaining similar steady-state behavior to the linear version near the reference. The proposed method improves performance without increasing implementation complexity, making it suitable for future validation on real systems to evaluate practical limitations, actuator impact, and robustness against unmodeled dynamics.

Keywords: Learning and Adaptation, Biologically-Inspired Robots, Dynamics and Control
Abstract: Blind locomotion refers to the challenge of navigating varied terrains without prior knowledge or exteroceptive data. Although quadrupeds often use external sensors, these can be unreliable in low-light or resource-limited settings and cannot anticipate disturbances such as slippage. In such cases, robots must adapt using only proprioceptive feedback. While slip detection and terrain topography estimation methods exist, leveraging proprioceptive information offers multiple advantages across various applications. This work explores Meta-Reinforcement Learning (Meta-RL) to enhance policy robustness and fast adaptation for quadruped robots during blind locomotion in challenging terrain. We build on the RL^2 algorithm, integrating recurrent neural networks into Proximal Policy Optimization (PPO) to implicitly encode task-specific information from experience. Two novel RL^2-based architectures are proposed and evaluated in simulations with the quadruped robot Anymal C across diverse terrain conditions, focusing on flat surfaces with stochastic slip and highly unstructured terrains. Results show that recurrent policies significantly outperform standard PPO, improving both adaptability and robustness under unpredictable ground dynamics, thereby enhancing the performance of blind quadrupedal locomotion in complex real-world environments.

Keywords: Biologically-Inspired Robots, Dynamics and Control, Mobile Robots
Abstract: We present a biologically-inspired gait control system for quadruped robots based on spiking neural networks (SNNs) functioning as central pattern generators (CPGs). By modulating dopamine, the same neural structure can produce different gaits such as trot and bound. A reward-driven mechanism adjusts the neuromodulation to select the desired gait, achieving reliable transitions between attractor states. The neural dynamics are validated in simulation, and the controller is directly deployed on a physical quadruped robot. Our results show that dopamine modulation enables robust and adaptive gait switching through low-dimensional control, with stable performance on real hardware.

Keywords: Simulation and Visualization, Dynamics and Control, Mobile Robots
Abstract: This paper presents a Load-Based Variable Transmission (LBVT) mechanism designed to enhance robotic actuation by dynamically adjusting the transmission ratio in response to external torque demands. Unlike existing variable transmission systems that require additional actuators for active control, the proposed LBVT mechanism leverages a pre-tensioned spring and four-bar linkage to passively modify the transmission ratio, thereby reducing the complexity of the robot joint actuation system. To evaluate the effectiveness of the LBVT mechanism, we conducted simulation-based analyses. The results confirm that the system achieves a 40% increase in transmission ratio upon reaching a predefined torque threshold, effectively amplifying joint torque when needed, without additional actuation. Furthermore, the simulations demonstrate a torque amplification effect, triggered when the applied force exceeds 18N, highlighting the system’s ability to autonomously respond to varying load conditions. This research contributes to the development of lightweight, efficient, and adaptive transmission systems for robotic applications, particularly in legged robots, where dynamic torque adaptation is crucial.

Keywords: Human-Robot Interaction, Rehabilitation Robotics, Dynamics and Control
Abstract: Human exoskeletons require controllers that adapt to user intentions while maintaining stability and comfort. This paper presents the design and experimental validation of the "ExoLoLi" unilateral knee–ankle exoskeleton equipped with a suite of controllers, including torque tracking, admittance and central pattern generator (CPG) based control. Custom control electronics are described together with an auto–adaptive Hopf‑oscillator architecture for human‑on‑the‑loop control. Six healthy volunteers participated in treadmill walking experiments under five conditions. Results show that torque control preserves user initiative but introduces hysteresis, admittance control requires careful tuning to avoid oscillations, and the CPG based controller yields nearly periodic motion with reduced interaction force. A hybrid controller combining CPG and admittance provides a compromise between transparency and adaptability. These findings support the potential of an adaptive CPG based architectures for lower–limb exoskeletons.

Keywords: Rehabilitation Robotics, Dynamics and Control, Human-Robot Interaction
Abstract: Impedance control of rehabilitation exoskeletons provides an effective solution to improve and personalize patient rehabilitation sessions. However, impedance control requires sensor measurements of interaction torques between the user and the exoskeleton. These sensors are expensive, difficult to integrate into some devices, and can affect patient comfort. This paper presents a new control approach using interaction torques estimated in real time from an artificial neural network. A multilayer perceptron artificial neural network, trained on real exoskeleton data, is used. The inputs of the neural network are different physical measurements made on the exoskeleton such as position and speed measurements by an encoder, acceleration by an accelerometer placed at the end-effector as well as motor signals. Experimental results show good control stability and consistent adaptation with interaction torques applied to joints. This solution thus encourages the use of control strategies based on real-time estimates of exoskeleton physical signals using neural networks.

Keywords: Dynamics and Control, Human-Robot Interaction, Simulation and Visualization
Abstract: Lower limb rehabilitation exoskeletons have attracted significant research interests due to their immense potential to improve quality of life. Despite significant progress in their development, accurate dynamic modeling and effective control system design remain growing areas of concern. This paper presents a comprehensive dynamic modeling and control framework for the human-exoskeleton interaction. The overall model is developed to capture the coupled dynamics through three interconnected components: the exoskeleton, the human, and the physical interaction between them. In this case, the physical interaction is represented by a simplified mass-spring-damper model, and the dynamic equations are derived using Euler-Lagrange mechanics. To ensure compliant and adaptive assistance, an impedance control framework is designed to minimize the interaction torques between the human and the exoskeleton system. Therefore, the control framework integrates three core modules: the physical human-exoskeleton interaction dynamics, parameter adaptation, and the control module to supervise actuator behavior. The modeling and control framework are validated through simulations and analysis of joint positions under sinusoidal and real gait trajectory references using MATLAB, and supported by a dedicated graphical user interface for simulation and visualization. The simulation results validate the accuracy of the dynamic modeling approach and demonstrate the controller’s ability to achieve stability and minimize interaction effects in different scenarios.

Keywords: Simulation and Visualization, Learning and Adaptation, Dynamics and Control
Abstract: Modeling is an important process for robots to handle 3D deformable objects, and making precise predictions at contact points is essential for successful manipulation. In this context, a framework is proposed for the development of a deformation model with low data and computational requirements that adapts to different object shapes with minimal reconfiguration. This framework integrates spatial and force information of a robotic grasping task into a network architecture with incorporation of physical principles to simultaneously solve forward and inverse problems in terms of predicting the temporal evolution of shape deformation and estimating material parameters. The model is evaluated in simulation environments for 3D deformable objects of various geometries and under multiple grasping conditions.

Keywords: Learning and Adaptation, Robotics Vision, Simulation and Visualization
Abstract: This study presents AffordGen, an automated method for generating task-oriented affordance-based datasets for robot manipulation learning, addressing limitations such as data scarcity and embodiment gaps. We evaluate our method through three experiments: a proof-of-concept, a progressive complexity task involving mug manipulation with varied environmental conditions, and a generalization test with unseen objects. Results demonstrate that models trained on the generated datasets effectively learn to manipulate objects through appropriate affordance regions, achieving up to 82.6% success rates in controlled environments. Moreover, the models successfully generalize to novel objects not seen during training, correctly identifying and using appropriate affordance regions.

Keywords: Robotics Vision, Learning and Adaptation, Humanoid Robots
Abstract: This paper proposes a geometry-based method to extract graspable regions on convex objects by analyzing their CAD models in conjunction with two-finger gripper specifica- tions. Inspired by human strategies for stable pinching, our algorithm identifies parallel surface pairs and applies a series of filters - stroke limits, finger length, clearance, and edge-aware placement - to ensure physically feasible and collision-free grasps. Beyond generating grasp poses, our method provides high-quality, explainable ground-truth annotations for training learning-based grasp predictions. The proposed method is validated using convex cuboid shapes, demonstrating consistent extraction of stable graspable areas across various grippers. Moreover, its gripper-aware yet generalizable design allows extensions to non-convex objects via primitive decomposition, making it suitable both for grasp planning and scalable dataset generation in learning-based robotics.

Keywords: Adversarial Planning, Mobile Robots, Learning and Adaptation
Abstract: Efficient path planning and smooth trajectory gen- eration are crucial for mobile robots in complex environments. While Rapidly-exploring Random Tree (RRT) methods provide robust solutions, they often suffer from slow convergence, sub-optimal paths, and extensive post-processing. We introduce RL-BiRRT, a Reinforcement Learning-enhanced Bidirectional RRT that leverages a Dueling Deep Q-Network (Dueling DQN) with an LSTM-based architecture for intelligent node generation and inherent path smoothing. Our adaptive exploration mechanism optimizes tree growth and reduces computational overhead. Experiments demonstrate that RL-BiRRT reduces iterations, nodes, and path length by 8.91×, 5.33×, and 1.84× compared to Bi-RRT, and by 9.62×, 5.63×, and 1.47× compared to RRT*, respectively. The results confirm RL-BiRRT’s efficiency in complex environments.

Keywords: Mobile Robots, Dynamics and Control, Robot Operating Systems
Abstract: This paper addresses motion planning and control of an overactuated 4-wheel drive train with independent steering (4WIS) where mechanical constraints prevent the wheels from executing full 360-degree rotations (swerve). The configuration space of such a robot is constrained and contains discontinuities that affect the smoothness of the robot motion. We introduce a mathematical formulation of the steering constraints and derive discontinuity planes that partition the velocity space into regions of smooth and efficient motion. We further design the motion planner for path tracking and obstacle avoidance that explicitly accounts for swerve constraints and the velocity transition smoothness. The motion controller uses local feedback to generate actuation from the desired velocity, while properly handling the discontinuity crossing by temporarily stopping the motion and repositioning the wheels. We implement the proposed motion planner as an extension to ROS Navigation package and evaluate the system in simulation and on a physical robot.

Keywords: Mobile Robots, Robotics Vision
Abstract: This work has its origins in the design of autonomous stretchers and presents a vision-based docking system designed to autonomously connect a differential drive robot to a mobile platform. A visual marker is mounted on the mobile platform, allowing its detection through a forward-facing camera onboard the robot, and the robot is equipped with a gripper to connect to the two devices. The connection, similar to a docking problem, is performed by executing a sequence of maneuvers. To do so, a hybrid control strategy is employed, using position-based control for coarse maneuvering and image-based control for precise final alignment, overcoming the robot's non-holonomic constraints. Numerous experimental tests demonstrate the system's feasibility and highlight its potential for improving hospital logistics using low-cost, modular solutions.

Keywords: Multi-Robot Systems, Mobile Robots, Dynamics and Control
Abstract: This paper presents a novel distance-based formation control strategy for multi-agent systems (MAS), based on the definition of the Formation Matrix, which describes both the distances and relative velocities among the agents of the arrangement. Inspired by the control of redundant robotic manipulators, we formulate an inverse dynamics-based control law that operates in the edge space—analogous to the operational space in robotics. By drawing an analogy with the Jacobian of redundant manipulators, we show that when the graph characterizing the MAS is minimally rigid, the Formation Matrix has full rank, and fixed-time tracking of time-varying inter-agent distances can be ensured. Furthermore, the redundancy in the system is exploited via the null space of the Formation Matrix. This allows for the incorporation of secondary objectives, such as collision avoidance between non-connected agents, desired positioning and orientation of the formation, and kinetic energy minimization. Stability of the equilibrium point is demonstrated in the Lyapunov sense. The simulations, involving a group of omnidirectional mobile agents, validate the effectiveness of the proposed control framework in achieving accurate distance tracking while fulfilling multiple secondary tasks through null-space projections.

Keywords: Multi-Robot Systems, Robot Swarms, Learning and Adaptation
Abstract: Multi-agent exploration in unknown environments is a fundamental problem in robotics and autonomous systems, with applications in surveillance, search and rescue, and en- vironmental monitoring. Traditional Coverage Path Planning (CPP) approaches often suffer from inefficient coordination, high collision rates, and redundant movements, particularly as the number of agents increases. In this work, we propose MultiAgent-DeepQ, a novel reinforcement learning framework that leverages a multi-headed Deep Q-learning architecture to optimize agent coordination and exploration efficiency. Our approach enables agents to dynamically adapt their movement policies based on learned coverage patterns while minimizing redundant revisits and avoiding collisions. We evaluate our method across multiple environments of increasing complex- ity, comparing it against existing mult-agent reinforcement learning-based approaches. The proposed framework achieves 100% coverage in all tested environments, completely eliminates collisions, and significantly reduces the total number of steps taken—achieving an average 10.5× improvement over the existing approach. Additionally, our method demonstrates strong scalability from 2 to 12 agents without performance degradation. These results highlight the robustness and efficiency of MultiAgent-DeepQ, making it a promising solution for largescale multi-agent exploration tasks.

Keywords: Multi-Robot Systems, Robot Operating Systems, Search and Rescue Robots
Abstract: Multi-Robot Exploration (MRE) systems with communication constraints have proven efficient in accomplishing a variety of tasks, including search-and-rescue, stealth, and military operations. While some works focus on opportunistic approaches for efficiency, others concentrate on pre-planned trajectories or scheduling for increased interpretability. However, scheduling usually requires knowledge of the environment beforehand, which prevents its deployment in several domains due to related uncertainties (e.g., underwater exploration). In our previous work, we proposed an intermittent communications framework for MRE under communication constraints that uses scheduled rendezvous events to mitigate such limitations. However, the system was unable to generate optimal plans and had no mechanisms to follow the plan considering realistic trajectories, which is not suited for real-world deployments. In this work, we further investigate the problem by formulating the Multi-Robot Exploration with Communication Constraints and Intermittent Connectivity (MRE-CCIC) problem. We propose a Mixed-Integer Linear Program (MILP) formulation to generate rendezvous plans and a policy to follow them based on the Rendezvous Tracking for Unknown Scenarios (RTUS) mechanism. The RTUS is a simple rule to allow robots to follow the assigned plan, considering unknown conditions. Finally, we evaluated our method in a large-scale environment configured in Gazebo simulations. The results suggest that our method can follow the plan promptly and accomplish the task efficiently. We provide an open-source implementation of both the MILP plan generator and the large-scale MRE-CCIC.

Keywords: Adversarial Planning, Mobile Robots, Human-Robot Interaction
Abstract: Decision-making in automated driving must consider interactions with surrounding agents to be effective. However, traditional methods often neglect or oversimplify these interactions because they are difficult to model and solve, which can lead to overly conservative behavior of the ego vehicle. To address this gap, we propose two quantum game models, QG-U1 (Quantum Game - Unitary 1) and QG-G4 (Quantum Game - Gates 4), for interaction-aware decision-making. These models extend classical game theory by incorporating principles of quantum mechanics, such as superposition, interference, and entanglement. Specifically, QG-U1 and QG-G4 are designed for two-player games with two strategies per player and can be executed in real time on a standard computer without requiring quantum hardware. We evaluate both models in merging and roundabout scenarios and compare them with classical game-theoretic methods and baseline approaches (IDM, MOBIL, and a utility-based technique). Results show that QG-G4 achieves lower collision rates and higher success rates compared to baseline methods, while both quantum models yield higher expected payoffs than classical game approaches under certain parameter settings.

Keywords: Mobile Robots, Robotics Architectures, Robot Operating Systems
Abstract: Service robots are complex and specialized machines used in a variety of industries from different areas, such as oil and gas, agriculture and mining. A design requirement for such robots is ingress protection. For this reason, these robots are prone to overheating when not equipped with a cooling system. This article uses EspeleoRobô, a service robot developed by Instituto Tecnológico Vale, as a case study. While ROS provides sufficient tools for general troubleshooting and debugging, this task is largely dependant on user experience. A self-diagnostic system could provide the user with assistance in parsing the information retrieved with these tools, thus smoothing the learning curve and improving diagnostic effectiveness. We propose an architecture for the overheating detection component that such a system would require, as well as its subcomponents, and use it on a real use-case recorded telemetry. Predictions made by the proposed system, using a hybrid model, showed an RMSE of 0.89 °C, an improvement of 20% over a purely physics based model. The results are promising, and support locking in the proposed architecture and focusing on its subcomponents.

Keywords: Simulation and Visualization, Search and Rescue Robots, Multi-Robot Systems
Abstract: Due to the intricate tasks and environments, particular complexities arise in developing and testing algorithms for autonomous heterogeneous vehicle interactions. Additionally, there is a need for a large amount of high-fidelity annotated data in various conditions and environments. Focused on these needs, we introduce BiguaSim as a novel open-source multi-domain simulator built on Unreal Engine 5 (UE5). BiguaSim aims to generate synthetic, realistic, annotated data based on various sensor implementations, for both air and water vehicles. Our simulator includes multi-agent support with GPU-enabled dynamics and integration with robotics tools like ROS2 and ArduPilot. To demonstrate the simulator, we conducted a multi-domain data collection employing a sample coverage strategy.

Keywords: Mobile Robots, Self-Localization and Navigation
Abstract: This paper presents a semantics-aware path-planning framework that enables quadrupedal robots to navigate environments by walking on specific terrains while considering safety, energy, and time constraints. The proposed method identifies key traversable elements and classifies them on a preference scale in a cost function where the weights are obtained from real-world experiments. Walking systems treat floors, stairs, and low vegetation as traversable; however, each terrain type significantly impacts safety, time, and energy efficiency. Pavements provide the best time and energy efficiency, whereas stairs require higher power consumption and take longer to traverse. Low vegetation, such as tall grass, can conceal hazards like holes, posing a threat to the robotic agent. The proposed method integrates a probabilistic global path planner with an interface that generates mission files for the quadrupedal robot ANYmal. The planner takes as input a segmented point cloud, which enables the identification of key features such as traversable and avoidable areas. Real-world experiments with the ANYmal robot in an outdoor environment were conducted to validate the effectiveness of the proposed approach.

Keywords: Mobile Robots, Human-Robot Interaction, Learning and Adaptation
Abstract: Safe and socially aware navigation in human-populated environments remains a major challenge for autonomous mobile robots. This paper presents DIR-SAFE, a reinforcement learning–based local planner that integrates feasibility, safety, and social compliance for differential-drive robots. The method models the robot feasible velocity space using Dirichlet distributions and guarantees collision-free navigation via a lightweight action-space bounding algorithm informed by static obstacle maps. An actor–critic architecture with augmented state inputs enables efficient, real-time action inference without requiring online optimization or simulators. The policy is trained using Proximal Policy Optimization across diverse social interaction scenarios to promote robust generalization. Numerical results demonstrate that DIR-SAFE is an effective navigation algorithm, achieving high success rates and maintaining compliance with social-space constraints, even in dense and previously unseen environments, without requiring careful parameter tuning.

Keywords: Learning and Adaptation, Multi-Robot Systems, Unmanned Aerial Robots
Abstract: Path planning in unknown and partially observable environments remains a major challenge in autonomous unmanned aerial vehicles (UAVs) navigation. This paper presents a Deep Reinforcement Learning (DRL) approach for two-dimensional navigation with dynamic obstacles and limited sensory input. The proposed strategy relies on Proximal Policy Optimization (PPO) to train agents to reach randomly placed target positions while avoiding collisions with static and moving obstacles. The agent perceives the environment using a local laser-based map and a relative vector to the target, without access to the global map. To evaluate performance, the PPO approach is compared with classical A* and DRL algorithms, such as Deep Q-Network (DQN), Deep Deterministic Policy Gradient (DDPG), under the same simulation conditions. Metrics include trajectory length, goal-reaching success rate, number of collisions, and cumulative reward. Experimental results show that PPO can generalize to novel scenarios, outperforming other DRL-based methods in terms of safety and goal efficiency.

Keywords: Mobile Robots, Multi-Robot Systems, Robot Swarms
Abstract: Autonomous navigation in complex environments requires robots to plan safe and robust paths, and potential field techniques based on Boundary Value Problems (BVP) have shown promise for this task. However, when applied to robot swarms, many existing approaches fail to maintain group cohesion in cluttered environments, often allowing the swarm to split unintentionally. In this work, we propose a novel strategy that dynamically computes elastic swarm formations using BVP-based potential fields with local distortions, while coordinating velocity to preserve formation integrity. Our method enables the swarm to expand or contract according to environmental constraints, ensuring uniform progress toward the goal without leaving individual robots behind. Experimental results in both simulated and real-world settings demonstrate that our approach allows the swarm to navigate safely and effectively through complex spaces, consistently reaching the goal as a single, cohesive group. In addition, comparative analysis with recent swarm navigation methods shows that our approach achieves favorable results, particularly in terms of consistent inter-robot spacing and stable formation behavior.

Keywords: Robot Swarms, Multi-Robot Systems, Mobile Robots
Abstract: Swarm robotics offers a robust and flexible solution for complex tasks, but navigating through narrow passages often leads to congestion and coordination issues. To address this, we propose the Multiple Line Coordination (MLC) algorithm, which enhances a previous single-line approach by introducing a new behavioral state, two waiting sub-states for refined control, a line size limit, support for multiple simultaneous lines, and explicit failure management. We evaluated MLC through extensive simulations across various swarm sizes, comparing it with the baseline strategy. The results show that MLC significantly reduces the average distance traveled, shortens the duration of close encounters, lowers the frequency of line entries per robot, and increases the overall success rate, particularly in smaller swarms. These outcomes confirm that MLC improves coordination and robustness in spatially constrained environments.

Keywords: Multi-Robot Systems, Mobile Robots, Self-Localization and Navigation
Abstract: This paper presents a comparative analysis of three classical trajectory planning algorithms, Dijkstra, A* with visibility graph representation, and Wavefront, for differentialdrive mobile robots in simulated logistics environments. Implemented in CoppeliaSim with a centralized conflict management policy based on task priority, the algorithms were evaluated in five scenarios with varying complexity. Metrics such as path length, execution time, and conflict resolution were assessed. The A* algorithm achieved the shortest paths (up to 25% shorter than Dijkstra) and lowest execution times, while Dijkstra was the only algorithm to complete all scenarios without failure. These results offer practical trade-offs for algorithm selection in autonomous mobile robot (AMR) logistics systems.

Keywords: Learning and Adaptation, Mobile Robots, Robotics Vision
Abstract: Standard reinforcement learning aims to maximize rewards through experience of repetitive interactions with a given environment. This perspective however neglects the natural property of learning to be non-stationary and to manifest continual adaptation and memorization, with diverse tasks emerging and disappearing spontaneously. Continual reinforcement learning (CRL) aims to account for this characteristic by making the environment change over episodes; however, this typically implies the usage of enormous replay buffers to prevent catastrophic forgetting (CF). In order to deal with CF in a memory-efficient fashion, the present paper proposes adiabatic reinforcement learning (ARL), an approach that utilizes adiabatic replay (AR), a recent technique originally intended for continual supervised learning. AR uses selective, internal replay of samples that are likely to be affected by forgetting. Consequently, ARL ensures a low memory footprint and a constant time-complexity with respect to the number of tasks. The approach was empirically evaluated over several benchmarks involving a differential wheeled robot with visual feedback.

Keywords: Underwater Robotic Systems, Unmanned Aerial Robots, Simulation and Visualization
Abstract: Unmanned aerial vehicles (UAVs), especially those that carry significant payloads, generally have a short flight range. Having a mobile platform to support them in long-range missions might increase their effectiveness. Therefore, an unmanned surface vehicle (USV) was designed to increase the UAV's flight time. This paper presents the development of a mathematical dynamic model for a catamaran-type USV in collaboration with the UAV. The established modeling has 6 degrees of freedom to analyze the coupled dynamics during the landing and take-off operations of the UAV. This model integrates rigid-body mechanics, with parameters derived from the USV's three-dimensional CAD model, hydrodynamic effects, and external forces and moments generated by the UAV. Numerical simulations were conducted to analyze the USV's response to these disturbances. The results quantify the platform's response, including heave, roll, and pitch oscillations.

Keywords: Simulation and Visualization, Robotics Vision, Learning and Adaptation
Abstract: High-quality datasets are critical for robotics, yet creating them is costly and time-consuming, leading to a reliance on pre-trained models that often fail when applied to niche domains due to data mismatch. To address this 'data gap,' we introduce SAGE (Synthetic Automated Generation Environment), a novel, open-source framework that integrates procedural content generation (PCG) with a high-fidelity physics engine to create fully parameterized, physically-grounded, and reproducible synthetic datasets. We demonstrate SAGE's efficacy in a challenging 3D LiDAR semantic segmentation task for off-road environments, using the RELLIS-3D dataset as a niche target. Our experiments show that pre-training on a SAGE-generated dataset significantly outperforms both a baseline model trained on limited real data and a model fine-tuned from a large but out-of-distribution real-world dataset (Semantic KITTI), which suffered from negative transfer. Crucially, we showcase SAGE's ability for iterative refinement: after identifying underperforming classes, we dynamically generated a targeted dataset update, boosting the model's average Intersection over Union (IoU) from 0.296 to 0.386 and successfully mitigating specific class failures. By enabling the rapid and targeted generation of customized multi-modal sensor data, SAGE provides a powerful and flexible solution to accelerate research in specialized robotics applications. SAGE is open-source and available at: https://sites.google.com/view/sage-robotics

Keywords: Mobile Robots, Robotics Vision
Abstract: This paper presents the design, development, and validation of an autonomous hexapod robot capable of detecting and tracking ant trails. The system integrates a mechanical platform, computer vision algorithms, and a decision-making heuristic to navigate complex agricultural terrains. The robot's hardware underwent multiple design iterations to ensure stability and mobility on surfaces like grass, soil, and inclined planes. A vision model based on YOLOv8 was trained using a composite dataset of real-world captures, public datasets, and synthetic data generated in a simulator. The complete system was validated in a high-fidelity simulator, which incorporated real-world motion and detection errors. The results demonstrate the system's effectiveness, achieving a 94% success rate in following ant trails under realistic simulated conditions. This work lays the groundwork for an eco-friendly alternative to chemical pesticides in agricultural pest control.

Keywords: Biologically-Inspired Robots, Educational Robotics, Underwater Robotic Systems
Abstract: The term embodied intelligence (EI) conveys the notion that body morphology, material properties, interaction with the environment, and control strategies can be integrated into the process of robotic design to generate intelligent behav- ior; in particular, locomotion and navigation. In this paper, we discuss EI as a design principle for advanced microrobotics, with a particular focus on co-design—the simultaneous and in- terdependent development of physical structure and behavioral function. In contrast to traditional architectures that decouple sensing, computation, and actuation, we present a series of robots developed by the author and his team at the Autonomous Microrobotic Systems Laboratory (AMSL) in which intelligent behavior emerges from physical interaction and structural dynamics. Platforms such as the Bee++, RoBeetle, SMALLBug, SMARTI, WaterStrider, VLEIBot+, and FRISSHBot exemplify how feedback loops, decision logics, sensing mechanisms, and smart actuation strategies can be embedded into the physical properties of the robotic system itself. Along these lines, we contend that co-design is not only a method for empirical optimization under constraints, but also an enabler of EI, offering a scalable and robust alternative to classical control for robotics at the mm-to-cm–scale.

Keywords: Biologically-Inspired Robots, Robotics Architectures, Embedded and Mobile Hardware
Abstract: We present a six-arm rigid–radial peristaltic motor cell that translates annelid-inspired anchoring and thrust into a practical locomotion unit for lunar-relevant granular media and ISRU workflows. The core idea is an angle-defined, open-loop cycle where sensing is not required for coordination, paired with a concise kinematic formulation that exposes an axial thrust window—a configuration band in which joint motion is most effectively converted into forward leverage. This insight guides a design that is predictable to schedule (speed governed by timing), robust and serviceable (rigid transmission with shieldable joints), and modular (a reusable cell that stacks into multi-segment trains or couples to light drilling/emplacement tools). The scope of this work is to establish a clear, reproducible, and extensible geometry+timing baseline in the laboratory; waveform optimization and the use of conformable skins are identified as future lines. In addition to laboratory trials, we include simulations of net displacement for a two-cell train that open the way for physical demonstration: experimental verification of net advance requires at least two coupled cells. Together, the results support the role of the cell as a building block for peristaltic locomotion in confined settings and shallow emplacement within dusty, abrasive, low-maintenance lunar environments.

Keywords: Rehabilitation Robotics, Human-Robot Interaction, Humanoid Robots
Abstract: Understanding and measuring children’s attention is a key challenge in educational human–robot interaction (HRI). This paper presents a novel evaluation framework for detecting attention in school-age children during a storytelling activity with the NAO social robot. Unlike traditional scenarios where the robot narrates, here the child actively tells a story while the robot listens and adapts, fostering engagement through constructivist and sociocultural learning principles. We integrate multimodal features—including gaze behaviour (automatically labelled using Gaze360 and K-means clustering), task performance, and physiological signals (heart rate)—to classify attention levels via machine learning methods (SVM, KNN, RF). Seventy-four children (aged 7–9) participated in the study, with attention labels validated by expert observers. Results show that combining gaze, task, and physiological features improves classification accuracy (>0.70) compared to unimodal approaches, with SVM achieving the best performance. These findings highlight the potential of multimodal attention detection for enabling adaptive, context-aware robot behaviours in education. Our framework advances the development of socially intelligent robots that can dynamically respond to children’s attentional states, ultimately supporting more engaging and effective learning experiences.

Keywords: Human-Robot Interaction, Rehabilitation Robotics, Robotics Vision
Abstract: Individuals with upper limb mobility impairments often require assistive technologies to perform activities of daily living. One relevant technique is the use of robotic arms to help such patients. However, constraints originating from these conditions significantly reduce a person's motor capacity, leaving only a few features that can be used as input in a robotic system, such as voice or gaze. While gaze-tracking has emerged as a promising method for robotic assistance, existing solutions lack sufficient mobility within the workspace and redundancy in marker tracking, resulting in reduced adaptability and uncertainty in recognizing user intent. This paper presents a Mixed-Reality Interface for Human-Robot Interaction via Gaze-Oriented Control (MIHRAGe), an integrated system that combines gaze-tracking, robotic assistance, and a mixed-reality environment to create an immersive environment for controlling the robot using only eye movements. The system was evaluated through an experimental protocol involving four participants, assessing gaze accuracy, robotic positioning precision, and the overall success of a pick-and-place task. Results showed an average gaze fixation error of 1.46 cm, with individual variations ranging from 1.28 cm to 2.14 cm. The robotic arm demonstrated an average positioning error of less than 1.53 cm, with discrepancies attributed to interface resolution and calibration constraints. In a pick-and-place task, the system achieved a success rate of 80%, highlighting its potential for improving accessibility in human-robot interaction with visual feedback to the user.

Keywords: Robot Operating Systems, Mobile Robots, Learning and Adaptation
Abstract: The integration of Large Language Models (LLMs), such as GPT, in industrial robotics enhances operational efficiency and human-robot collaboration. However, the computational complexity and size of these models often provide latency problems in request and response times. This study explores the integration of the ChatGPT natural language model with the Robot Operating System 2 (ROS 2) to mitigate interaction latency and improve robotic system control within a simulated Gazebo environment. We present an architecture that integrates these technologies without requiring a middleware transport platform, detailing how a simulated mobile robot responds to text and voice commands. Experimental results demonstrate that this integration improves execution speed, usability, and accessibility of the human-robot interaction by decreasing the communication latency by 7.01% on average. Such improvements facilitate smoother, real-time robot operations, which are crucial for industrial automation and precision tasks.

Keywords: Robot Operating Systems, Robotics Vision, Self-Localization and Navigation
Abstract: This work presents a dynamic modeling and control strategy for the Parrot Anafi quadrotor, with application to autonomous navigation in outdoor environments. The proposed model is identified from flight data and validated through experiments using a PD controller with feedback linearization. A Kalman filter is employed for state estimation, enabling robust performance even without using motion capture systems. Real-world tests demonstrate accurate vertical and yaw tracking under wind disturbances. The proposed framework is open-source and ready for deployment in aerial robotics research.

Keywords: Unmanned Aerial Robots, Dynamics and Control, Mobile Robots
Abstract: This paper compares the performance of a linear model predictive controller against that of a feedback linearization controller to check the feasibility of using the former technique in controlling the navigation of an unmanned aerial vehicle with obstacle avoidance. The comparison is performed by running an experiment in which a quadrotor is controlled to follow a prescribed path with an obstacle, guided by two control systems that utilize inner-outer control loops. The first control system is designed using the feedback linearization technique, resulting in two control loops: the outer loop, which operates on the position-tracking error, and the inner loop, which operates on the velocity-tracking error. The second control system uses the same outer loop, but the inner one is a model-based predictive control framework designed to consider constraints on the control signal applied to the UAV. Approximating the dynamic model of the quadrotor using linear state equations, considering that it is in a near-hover condition, which means small pitch and roll angles, and thus low longitudinal and lateral linear velocities, the framework is a linear model-based predictive control system. The comparison results indicate that model-based predictive control has high potential for guiding a quadrotor, offering promising perspectives for advancing this research.

Keywords: Unmanned Aerial Robots, Dynamics and Control, Embedded and Mobile Hardware
Abstract: This work addresses the control problem of a foldable quadrotor whose in-flight arm reconfiguration changes its moment of inertia. A Linear Parameter Varying (LPV) controller is first designed to maintain consistent performance across configurations, but its computational cost limits its use on low-cost autopilots. Based on frequency domain analysis, we approximate the LPV with a low-order controller, akin to an adaptive PID, achieving similar response at lower complexity. Simulations and experimental tests on a Cortex M3-based platform, using both fixed-point and doubled-precision implementations, show similar behavior. The approach approximates LPV-like performance with PID-level efficiency, enabling real-time deployment on resource-constrained UAVs.