Motion Control for Robots and Automation

1. Introduction

Motion control constitutes one of the most fundamental aspects of robotics and automation engineering. The ability to generate, execute, and regulate the motion of robotic mechanisms determines the performance of systems, ranging from precision industrial manipulators to autonomous mobile platforms operating in unstructured environments [1,2]. Manufacturing processes demand higher throughput with tighter tolerances, construction operations move toward automation for improved safety, and rehabilitation systems require precise actuation of human joints; the field of motion control continues to evolve in both breadth and technical depth to meet these needs [3,4].

The classical foundations of robotic motion control rest on well-established principles of kinematics, dynamics, and feedback control theory. Interpolation-based trajectory planning, proportional-integral-derivative (PID) regulation, and model-based optimal control have served the robotics community for decades and remain indispensable in many practical applications [5]. However, the growing complexity of modern robotic tasks has motivated the development of more advanced approaches. Sliding mode control and its higher-order variants offer robustness against model uncertainties and external disturbances [6]. Optimization-based methods, including metaheuristic algorithms such as particle swarm optimization (PSO) and genetic algorithms, provide powerful tools for addressing trajectory planning problems with complex constraints [7]. Event-triggered and adaptive control architectures reduce communication and computation overhead in networked multi-robot systems [8].

In parallel, the rapid progress of artificial intelligence has introduced new capabilities into the motion control pipeline. Deep reinforcement learning enables robots to acquire manipulation and navigation skills through trial-and-error interaction with their environments, bypassing the need for explicit analytical models [9]. Convolutional neural networks and their variants have become standard tools for visual perception tasks that inform motion planning decisions, including object detection, defect classification, and scene understanding [10]. Graph neural networks and attention mechanisms are finding application in combinatorial optimization problems such as multi-robot task allocation [11]. The convergence of these data-driven techniques with established control-theoretic methods represents a defining trend in contemporary robotics research [12].

This Special Issue, entitled “Motion Control for Robots and Automation,” was established to capture recent advances across this evolving landscape. The fifteen research contributions collected here address topics spanning trajectory planning and vibration suppression, advanced control strategies for diverse robotic platforms, path planning and safety-oriented automation, environmental perception and quality inspection, and data-driven methods for visual intelligence. The contributing works involve a broad range of application domains, including industrial manufacturing, construction machinery, rehabilitation engineering, multi-robot coordination, and autonomous excavation. The following sections provide an overview of the published articles and conclude with a discussion of emerging themes and future research directions.

2. An Overview of Published Articles

Trajectory planning and vibration suppression in industrial robotic applications form the first thematic cluster. The challenge of optimizing S-shaped velocity profiles for robotic arms was taken up by Wu et al. [13], who presented a robot trajectory planning particle swarm optimization (RTPPSO) technique incorporating adaptive weight strategies and random perturbation terms to overcome premature convergence. Simulation results confirmed improved acceleration characteristics and reduced execution times relative to standard particle swarm optimization and genetic algorithm methods. Residual vibration during high-speed motion of lightweight industrial components motivated the analytical study of Cui et al. [14], in which two vibration suppression conditions for trapezoidal velocity profiles were derived from the relationship between acceleration switching times and system natural periods. Experiments on two robot manipulators demonstrated vibration reductions exceeding 90% without requiring controller modifications. A rather different motion paradigm was explored in [15], where Lee et al. developed non-prehensile dynamic manipulation techniques for rapid object transfer. Two complementary methods, a two-fingered scoop-and-flick and a one-fingered topple-and-flick, were validated on a custom platform, showing that impulsive contact interactions can achieve accurate high-arc transfers as efficient alternatives to conventional grasp-based operations.

Advanced control strategies for diverse robotic platforms constitute the second group of contributions. To address the challenge of servo control for finger joint rehabilitation through functional electrical stimulation, Chen et al. [16] developed the proxy-based super-twisting algorithm, which effectively prevents integrator windup during actuator saturation. The algorithm was implemented on Arduino hardware with an implicit Euler method to suppress numerical chattering, and experiments with human volunteers confirmed superior tracking accuracy over conventional methods. Formation control of multiple nonholonomic mobile robots under full-state constraints was the focus of [17], where Wang et al. combined fuzzy logic approximation of unknown dynamics with barrier Lyapunov functions and a backstepping framework. Their event-triggered communication mechanism achieved an 88.17% resource-saving rate for a four-robot group while maintaining bounded formation tracking errors. Reinforcement learning provides the methodological foundation for the next two contributions. Yan et al. [18] tackled obstacle avoidance for robotic arms by integrating inverse-kinematics-guided exploration with an adaptive reward function, achieving faster convergence and higher success rates than standard soft actor–critic (SAC) and deep deterministic policy gradient (DDPG) in dynamic environments. The problem of controlling a manipulator to direct water jets at target positions was addressed in [19] through Goal-Priority Hindsight Experience Replay (GPHER), a method that dynamically adjusts sampling priorities based on goal-space distance to improve performance in difficult boundary regions of the workspace.

Path planning and safety-oriented automation represent another important thread running through the collection. Recognizing the risk that dropped loads pose to ground workers during tower crane operations, Cai et al. [20] developed a dynamic path planning model that combines You Only Look Once version 8 (YOLOv8) instance segmentation with an improved dynamic window approach, validated on actual construction sites. The concept of line-following was extended into three-dimensional space by Wawrzyniak et al. [21], who equipped an articulated 6-Degrees-of-Freedom (DoF) robot with six low-cost Time-of-Flight (ToF) sensors and achieved reliable centimeter-level accuracy with minimal computational overhead. In the domain of multi-robot coordination, Hu et al. [22] formulated the task allocation problem under spatiotemporal and spatial constraints as an asynchronous Markov Decision Process (MDP) over a directed heterogeneous graph and proposed the Edge-Enhanced Attention Network (E2AN), whose encoder incorporates edge attributes reflecting actual path costs under obstacles.

Environmental perception and quality inspection also received significant attention. A notable contribution to terrain sensing for autonomous excavation came from Zhao et al. [23], who introduced a framework that processes Light Detection and Ranging (LiDAR) point clouds through a custom GridNet to extract terrain features informing a reinforcement learning agent. Surface anomaly detection on spot welds in automotive production was the subject of [24], where Liu et al. proposed improved YOLOv8 incorporating task-aligned detection heads and cross-scale feature fusion alongside a balanced generative adversarial network (GAN) for data augmentation. Efficient point cloud registration under varying data density conditions was investigated by Liu et al. [25], who presented a method based on dynamic neighborhood features that adapts the neighborhood radius to local geometry, achieving competitive accuracy across standard benchmark datasets. Finally, two contributions addressed the optimization of visual intelligence methods from complementary angles: Zou and Wang [26] systematically optimized convolutional neural network (CNN) parameters for imbalanced image classification through analysis of variance with fractional factorial design, while Guo et al. [27] integrated multi-scale feature extraction with adversarial domain adaptation in their multi-scale, multi-channel Taylor adversarial domain adaptation network (MMTADAN) algorithm for cross-domain image classification.

3. Conclusions

The fifteen articles collected in this Special Issue reflect both the breadth and the depth of contemporary research in robotic motion control. Several themes cut across the individual contributions. Classical control and analytical optimization methods continue to provide reliability and formal guarantees, while reinforcement learning and neural networks contribute adaptability to complex, unmodeled dynamics. Computational efficiency is a recurring concern, evidenced by embedded platform implementations, lightweight network architectures, and event-triggered communication strategies. Safety has emerged as a prominent design objective, particularly in construction and human–robot coexistence scenarios where vision-guided dynamic path planning shows considerable promise. Multi-robot coordination is advancing toward richer environmental representations that go beyond Euclidean distance assumptions to capture the true spatial structure of operating environments.

Several open questions merit further investigation. Reliable sim-to-real transfer of learned policies, extension to scenarios involving deformable objects and unstructured outdoor environments, and the potential integration of foundation models and large language models into robotic planning pipelines all represent promising research frontiers. Equally important, as robotic systems increasingly share workspaces with humans, research on safety verification, ethical governance, and user trust will be essential for translating laboratory achievements into dependable real-world deployments. We hope this Special Issue offers useful perspectives for researchers and engineers advancing the science and practice of motion control in robotics and automation.