The study presents a novel approach to tackle the issues of stability and safety in robotic manipulators. The researchers propose an obstacle-free deep reinforcement learning (DRL) trajectory planner integrated with an auto-tuning low- and joint-level control strategy. This model-free DRL agent plans velocity-bounded and obstacle-free motion for a manipulator in task space through joint-level reasoning. The plan is then input into a robust subsystem-based adaptive controller, which uses the Cuckoo Search Optimization (CSO) algorithm to enhance control gains. The approach ensures that position and velocity errors exponentially converge to zero, accounting for variations, modeling errors, and disturbances. Theoretical assertions are validated through simulation outcomes.
Publication date: 4 Feb 2024
Project Page: https://arxiv.org/abs/2402.02551v1
Paper: https://arxiv.org/pdf/2402.02551
