A Feasibility Enhancing Hybrid Learning Approach for Motion Planning for Manipulators in Manufacturing Setups

Abstract Industrial robots are widely used in diverse manufacturing environments. Nonetheless, how to enable robots to automatically plan trajectories for changing tasks presents a considerable challenge. Further complexities arise when robots operate within work cells alongside machines, humans, or other robots. This article presents a practical multilevel hybrid motion planning strategy combining operator demonstration-driven task-space planning with joint-space optimization that enforces key constraints such as reachability, joint limits, manipulability, and collision avoidance. The framework uses a supervisory agent to automatically switch between the planning modules, ensuring all generated robot trajectories are both feasible and suited for real manufacturing environments. Therefore, the derived hybrid motion planning policy generates a feasible trajectory that adheres to task constraints with simplistic demonstrations. Experimental validation in simulated and physical setups shows the approach reduces reconfiguration time and increases task success rates compared to conventional motion planning solutions.