A review of recent trend in motion planning of industrial robots

Motion planning is an integral part of each robotic system. It is critical to develop an effective motion in order to achieve a successful performance. The ability to generate a smooth, optimal, and precise trajectory is crucial for a robotic arm to accomplish a complex task. Classical approaches such as artificial potential fields, sampling-based, and bio-inspired heuristic methods, have been widely used to solve the motion planning problem. However, most of these methods are ineffective in highly dynamic and high-dimensional configuration space due to the high computations and low convergence rates impeding real-time implementations. Recently, learning-based methods have gained considerable attention in tackling the motion planning problem due to their generalization and high ability to deal with complex issues. This research presents a detailed overview of the most recent developments in solving the motion planning problem for manipulator robotics systems. Specifically, it focuses on how learning-based methods are developed to address the drawbacks of classical approaches. We examined current works on manipulator motion planning and outlined the gaps, limitations, and prospects for further research and analysis. Subsequently, this study investigates three main learning-based motion planning methods: deep learning-based motion planners, reinforcement learning, and learning by demonstration. This paper can help experts to benefit from concise version of advantages and disadvantages of different motion planning techniques to use them in their research. We anticipate that learning-based path planning methods will remain the subject of research in the foreseeable future because these solutions are typically dependent on problem-specific knowledge and datasets.