The Research on Robotic Arm Motion Planning Based on Deep Learning and Hybrid Path Planning

This paper proposes a robotic arm motion planning method based on deep learning and hybrid path planning. Environmental point cloud data is obtained using a Realsense depth camera, and the GSNet network is employed to extract global environmental features, generating key information for path planning. The motion planning module utilizes LSTM and MDN networks to predict the next motion pose of the robotic arm, generating a Gaussian mixture distribution for the next joint state. The specific next joint state is determined through Gaussian sampling. A collision detection module is integrated into the path planning process to validate the feasibility of the generated joint states. If collision detection fails beyond a certain threshold, a strategy combining the Artificial Potential Field (APF) method and the Rapidly-Exploring Random Tree (RRT) algorithm is employed. This approach overcomes local minima issues and achieves local obstacle avoidance, ensuring the feasibility and smoothness of the path. Experimental results demonstrate that this method effectively integrates the advantages of deep learning and traditional hybrid path planning algorithms. It not only improves planning efficiency but also adapts to the motion planning requirements of complex environments, providing an efficient solution for robotic motion planning.