In this article, we propose a navigation approach based on a flocking algorithm for multiple agents that discover the environment with their sensors. The proposed method is able to move the agents in a flexible formation that preserves a pattern that minimizes a measure of energy, but is also able to avoid collision with obstacles while maintaining communication and visibility between agents. Control laws are combined with a new version of the planner called the Nearness Diagram (ND), extended to multiple agents. Two central ideas in the approach are: (1) the global leader of the formation can change, selecting as global leader one that has favorable conditions to reach the goal, (2) each agent locally follows another agent, the agent that is followed minimizes a cost related to the shortest path to the global leader in a graph. The method is intensively validated with several computer simulations, implemented in differential-drive robots. The results showed that the approach maintains communication formation and flexible enough to allow agents to line up when narrow passages appear in the environment.

This paper presents a comprehensive navigation strategy aimed at enabling autonomous and collision-free motion of automated guided vehicles (AGVs) in indoor environments. The work addresses limitations of conventional AGV systems which are often restricted by static routing and inadequate adaptability to dynamic obstacles. A framework on global–local motion planning is developed, optimising (i) a global path planner based on the Probabilistic Roadmap (PRM) algorithm, (ii) a local trajectory tracking using the Pure Pursuit (PP) controller, and (iii) a real-time obstacle avoidance method leveraging the vector field histogram (VFH) algorithm. The decision-making layer oversees the coordination between global routing and local manoeuvring, dynamically switching between planning and avoidance modes in response to environmental changes. The proposed framework is validated through real-world experiments on a physical AGV platform operating in a structured environment. Experimental results show that the proposed framework generates feasible and smooth paths, while PP-based trajectory tracking achieves high final positioning performance when supported by fused odometry. In free-path navigation, fused odometry reduced the average final percentage error to 0.61% and increased the average final positioning accuracy to 99.40%. In obstacle-path navigation, the integrated PP–VFH framework maintained high final positioning accuracy at 99.10% under varying obstacle configurations. The coordination of PRM, PP, and VFH enables smooth, adaptive, and safe navigation of the AGV within a structured workspace. The proposed framework contributes a unified, modular architecture that strategies to bridges global path planning, local motion control, and real-time reactive avoidance within a single system.

Blind search methods are widely used in the search phase of robotic peg-in-hole assembly. However, these methods are sensitive to the initial contact position and are time-consuming. Additionally, in scenarios with a limited search space, such as thin-walled holes, the hole search process is prone to failure due to loss of contact between the peg and the hole. To address this issue, this paper proposes a hole search strategy that uses visual detection of the contact state between the peg and the hole and peg position adjustment based on the detection results. First, a contact-state classification model under fixed single-view conditions is constructed, converting the disordered relative positional relationships between the peg and the hole into discrete contact states, laying a theoretical foundation for the vision-guided hole search process. Second, search trajectories are generated based on the contact states, and trajectory optimization is further performed through visual error analysis, thereby overcoming hole search failures caused by visual occlusion and camera mounting deviations. Finally, convergence analysis shows that the peg slides into the hole when the classification accuracy is greater than 50%. Robotic assembly experiments further demonstrate the effectiveness of the proposed contact-state classification model and search strategy in tasks such as thin-walled hole insertion and USB Type-A(USB-A) connector insertion.

Multi-robot coverage path planning (MCPP) is a key technology in the field of precision agriculture, providing fundamental support for autonomous agricultural vehicles and unmanned aerial vehicles (UAVs) to perform tasks such as crop monitoring, pesticide spraying, and seeding. Although existing MCPP methods can generate feasible coverage paths, they usually do not consider the nonholonomic kinematic constraints on robots during the planning stage but instead rely on post-processing methods to smooth the paths. This makes it difficult to evaluate the true costs of paths and easily causes an imbalance in task allocation, thereby prolonging coverage time. Therefore, this article studies the coverage path planning (CPP) problem of multiple curvature-constrained Dubins robots in agricultural scenarios. Firstly, inspired by the Generalized Traveling Salesman Problem, we propose an exact mathematical model to solve the optimal coverage paths. This model is more concise than existing ones as it does not use additional endpoint continuity constraints. Based on this, we further propose a game-based Dubins multi-robot coverage path planning algorithm, which achieves load balancing through task negotiation among robots and reduces the maximum coverage time. Experiments conducted in various agricultural scenarios show that the proposed algorithm significantly improves the system load balancing and effectively reduces the overall coverage time. Specifically, compared with existing algorithms, the algorithm in this article reduces the coverage time by an average of $10.8\%$ and improves the load balance degree by approximately $47.4\%$. The feasibility tests also indicate that the generated paths are suitable for fixed-wing UAVs.

This paper proposes an energy-efficient walking generation method utilizing limit cycles generated by nonlinear model predictive control (NMPC). Conventional limit cycle walking methods rely on strong feedback, such as output zeroing control, to attract the robot’s state toward a predefined periodic trajectory. However, we hypothesize that employing feedback control that better leverages the robot’s dynamics can improve energy efficiency during walking. Our previous work confirmed that using limit cycles generated by NMPC can produce energy-efficient walking patterns. This study builds upon this foundation and proposes a new method for generating walking in a general five-link bipedal robot. Through extensive numerical simulations, we demonstrate that the proposed method achieves highly energy-efficient walking while exhibiting excellent convergence to periodic trajectories.

Direct collocation (DC) methods are utilized for addressing trajectory optimization challenges in robotics due to their ability to generate dynamically consistent solutions. However, in the cable-driven robotic systems, where tension constraints impose kinodynamic restrictions, maintaining accuracy becomes significantly complex. This article addresses robot tensionability and proposes a method to overcome the limitations. A DC method is proposed to minimize the actuator force rate in a trajectory planning problem for a designed cable-driven parallel robot. The system comprises a 3-cable parallel mechanism with a central spine to counteract the end-effector’s weight and enhance tensionability. Integrating a pneumatic cylinder into the system that supports trajectory planning implementation is essential to minimize jerky motions. The DC method is applied through the proposed quadratic programming approach and benchmarked against existing packages to achieve and compare the resulting smoother trajectory. The numerical results demonstrate that the proposed method significantly reduces computation cost and enhances accuracy. Experimental data corroborate the simulation results, validating the method’s efficacy.

Reliable extraction of line segments from complex scenes has been enabled by the line segment detector (LSD) algorithm, which provides accurate geometric information for various visual perception applications. However, LSD often extracts line segments that are either excessively short or overly long, increasing computational complexity and degrading feature matching accuracy. Although many methods focus on eliminating redundant short lines, the challenges of overly long line segments remain largely unexplored. To address these issues, this paper proposes an enhanced LSD-based method that integrates short-segment elimination and long-segment truncation strategies. The resulting line segments are constrained within a reasonable length range, effectively balancing geometric information with feature matching robustness. In addition, a collinearity-based matching method is introduced to maintain the geometric continuity of truncated segments and improve inter-frame feature consistency. Furthermore, the proposed method is integrated into a point-line visual-inertial odometry system by fusing data from an inertial measurement unit, significantly improving system robustness and accuracy, especially in low-texture environments. Extensive experiments on public datasets and real-world scenarios show that the proposed method achieves competitive overall performance, particularly in structured and low-texture environments.

The state-of-charge (SOC) estimation of lithium-ion batteries is essential for the performance, safety, and longevity of electric vehicles. While traditional physics-based models face difficulties in predicting SOC under complex conditions, data-driven methods demand large volumes of labeled data, limiting their practical applicability. This paper introduces a Transformer-based SOC estimation method that incorporates a self-supervised learning framework. By combining a masking recovery mechanism with a self-attention mechanism, the approach effectively captures SOC’s long-term dependencies and identifies global patterns under complex operating conditions. Experimental results on the Nissan Leaf and University of Maryland datasets demonstrate significant performance improvements. Compared to traditional deep learning methods, this approach delivers higher prediction accuracy and robustness, offering an innovative solution for efficient SOC estimation in complicated environments.

Odor source localization (OSL) in complex industrial environments remains a significant challenge due to the coupled effects of building occlusion and turbulent advection, which often lead to pseudo-source stagnation. This paper proposes an adaptive collaborative Genghis Khan shark optimizer (ACGKSO), a unified search framework designed for robust multi-robot OSL. Specifically, an adaptive perturbation strategy is introduced to balance global exploration and local exploitation. This is achieved through a dynamic step size mechanism driven by two factors: an exponentially decaying convergence rate and a linearly amplified term based on swarm dispersion. Concurrently, to counter premature convergence in complex environments, we introduce a dual-modal experience sharing mechanism. This approach enhances collective learning by integrating both individual historical best solutions and neighborhood optima. Experimental results validate the effectiveness and robustness of the proposed approach compared with other swarm intelligence methods. In single-interference scenarios, ACGKSO achieves a localization success rate of 94%, significantly outperforming the baseline algorithms. In more complex multi-interference environments, it maintains an 83% success rate and reduces the average number of search steps by 21.1% relative to the standard GKSO. These results demonstrate that our framework significantly outperforms other SI methods in terms of accuracy and efficiency across different environmental conditions.

This paper presents a sensorless human-intention-based control framework for a differential-drive power-assist mobile robot for indoor cooperative transportation. The proposed method estimates interaction-consistent motion cues in the motion domain using wheel-encoder measurements and motor-side actuation information, without relying on dedicated force or torque sensors. An interaction observer is first used to extract an acceleration-like interaction cue from the discrepancy between the commanded robot motion and the measured robot response. This signal is then processed by a human-intent analysis module, in which encoder-derived motion features and statistical class modeling are used to distinguish representative human interaction patterns from disturbance-related motion variations. The resulting interaction-related signal is separated into estimated human-induced and disturbance-induced acceleration components. The human-induced component is converted into a compliant velocity command through a virtual impedance model, whereas the disturbance-induced component is used to construct a disturbance-compensation term in the power-assist controller. Experimental results obtained in indoor cooperative transport scenarios suggest that the proposed framework provides effective assistive control behavior while reducing sensitivity to environmental disturbances.

This study presents the control of an omnidirectional automated guided vehicle (AGV) with mecanum wheels using a hybrid optimization algorithm that combines a modified A* algorithm and the dynamic window approach (ADWA-HO). The method ensures efficient and precise navigation in both static and dynamic environments. The modified A* algorithm generates global paths, removes redundant nodes, and refines trajectories to improve efficiency and smoothness. At the same time, the dynamic window approach (DWA) enables real-time local path planning and obstacle avoidance. By evaluating the AGV’s motion commands in real time, ADWA-HO selects optimal velocity commands within a dynamically updated window, thereby reducing route conflicts and ensuring stable movement. Compared with benchmark methods including dynamic A* (D*), artificial potential field (APF), DWA, probabilistic roadmap (PRM) & rapidly exploring random tree (RRT) fusion, and PRM & DWA fusion, the proposed ADWA-HO achieves improvements in average path length of 28.10%, 22.95%, 21.16%, 17.35%, and 10.71% and in average motion time of 23.48%, 17.85%, 15.47%, 11.86%, and 7.53% on both Map 1 and Map 2, respectively. The difference between simulation and real-world experiments is limited to 5.35% in path length and 4.38% in motion time, confirming the method’s practical reliability. Furthermore, the algorithm achieves lower standard deviation in both metrics, indicating higher consistency of performance. This work also introduces a novel map-building strategy based on geometric and semantic data modules, which enhances the adaptability of real-world AGV deployment.

The integration of large language models (LLMs) into KUKA industrial robots offers a promising pathway toward intuitive natural language control and a higher level of intelligence. However, existing methods face two primary challenges: their reliance on predefined motion primitives and the syntactic fragility imposed by the strict constraints of proprietary KUKA Robot Language (KRL). To address these challenges, we propose MIKI (multi-agent integrated KUKA interface), a decoupled dual-agent framework that bridges high-level reasoning and low-level syntactic verification for industrial robot programming. The controller agent incorporates a chain-of-thought (CoT) reasoning mechanism to systematically decompose abstract human instructions into logical intermediate steps. By leveraging a visual toolchain, it translates these reasoned plans into a sequence of executable actions, each embedded with precise world coordinates. Subsequently, the code generation agent employs a syntax-guided iterative refinement process, underpinned by an ANTLR4 parser, to transform these actions into verified and executable KRL code. Experimental results demonstrate that this closed-loop approach substantially improves the syntactic success rate of code generation from 0.61 to 0.89. Furthermore, the framework achieves an average success rate of 0.6 across five desktop manipulation tasks of varying complexity using a physical KUKA robotic arm, validating its effectiveness in bridging high-level cognitive reasoning with deterministic industrial execution.

This paper presents an innovative hybrid approach that integrates traditional control strategies with deep reinforcement learning for robotic assembly. By fusing multimodal information from visual and force feedback, the method leverages admittance control to ensure safe force feedback while using deep reinforcement learning to process visual input, enabling precise control and real-time correction of assembly actions. This multi-sensor feedback mechanism not only enhances the stability and accuracy of the assembly process but also improves the robot’s robustness and adaptability in uncertain environments. Additionally, a twin-delay deep deterministic policy gradient algorithm based on residual reinforcement learning is proposed. The design of a task-specific reward function, which simultaneously considers visual goals, force compliance, and contact stability, effectively addresses challenges such as difficult state information acquisition and sparse rewards in assembly tasks. This improves the robot’s interaction capabilities and task execution efficiency in real-world environments. Experimental results demonstrate that the method designed in this paper effectively reduces the training time for reinforcement learning from 400 epochs to 100 epochs, significantly decreases the magnitude of contact forces during the assembly process, and shortens the contact time.

Reliable path planning is a significant requirement for humanoid robots to move around in an environment filled with obstacles. In this paper, a novel method for humanoid robot navigation is proposed based on the Adaptive Dragonfly Algorithm (aDA), which is an improvement over the traditional Dragonfly Algorithm. The aDA method provides a better trade-off between exploration and exploitation of the search space, thereby avoiding local optima and increasing diversity in the search process. The aDA approach has been implemented and evaluated in a range of Webots-based simulation and experimental environments. Its performance is assessed using both single and multiple humanoid robots to analyze navigation capabilities under static as well as dynamic conditions. In multi-robot scenarios, conflict situations arise since each robot can act as a moving obstacle for the others, to overcome such situation, the dining philosopher controller is integrated with the proposed technique. The results of the aDA method are compared with traditional heuristics and metaheuristics for different scenarios of obstacle densities. The results obtained from the virtual and real environments show a variation within 5%, which proves that the aDA method is a robust and intelligent solution for humanoid robot navigation in a real-time environment. Furthermore, the proposed controller is benchmarked against an existing navigation method, demonstrating its improved performance and effectiveness.

The 4-DOF (four degrees of freedom) parallel manipulator, namely 2PPaRR-2PPaRU, offers three translations and one rotation capabilities, boasting an extensive rotation range, and a regular workspace. Moreover, its workspace can infinitely extend in conjunction with the extension of the driving joints. Nonetheless, the symmetric structure relying on Pa joint design presents performance challenges in the initial configuration and complicates the analytical solution of the forward kinematics, thereby increasing the complexity of dynamics and control. To address these limitations, the structure of the moving platform is modified by coaxially connecting the three joints at the ends of the three limbs to the moving platform. This improvement is followed by analyses of mobility, motion/force transmissibility, and stiffness. The overall reachable workspace area, good transmission workspace ratio, and global stiffness index are then chosen as key objectives for dimensional optimization, thereby achieving satisfactory performance for the new 3T1R PM. Finally, the CAD model with the selected parameters is presented, together with the distributions of motion/force transmission index and virtual stiffness index within the reachable workspace. The results validate the effectiveness of the optimization design.

The use of passive exoskeletons in industrial settings has gained growing interest as a means to reduce muscle fatigue and prevent work-related musculoskeletal disorders. However, translating laboratory methods into realistic occupational environments remains a challenge. This study presents a modular and wearable-sensor-based experimental protocol designed to bridge this gap by enabling the evaluation of exoskeletons in both static (STC) and dynamic (DYN) tasks while preserving natural movement variability. A total of 52 participants, including both men and women, completed tasks with and without two different passive exoskeletons, while their motor activity was assessed using surface electromyography (sEMG) and inertial motion sensors. The protocol incorporates key EMG-based metrics – Root Mean Square (RMS) and Hilbert Median Frequency (MDF) – that effectively quantify muscle activation and fatigue, along with subjective Perceived Fatigue Scores (PFS) and a task performance metric (Screwing Velocity, SV). The results confirm that the exoskeletons significantly reduce muscle activation and perceived fatigue without impairing task performance. The proposed methodology, combining rigorous metrics with wearable and non-invasive instrumentation, offers a robust framework for evaluating fatigue in both STC and DYN tasks and usability in both laboratory and field settings. This protocol represents a valuable tool for both research and industrial evaluation, facilitating the evidence-based integration of exoskeletons into real-world industrial workflows.