Skill Training for Space Teleoperation: DRL-Based Intelligent Agent for Training Strategy Adjustment with Innovative Evaluation

This paper introduces a skill training system for space teleoperation, featuring a self-adjusting strategy implemented by an intelligent agent and evaluated through Fr´ echet Distance. Operating within a dual-user shared control frame work, the system balances practical task engagement with essential task safety. Departing from traditional expert-dependent and manually-crafted training strategies, the study employs Deep Reinforcement Learning (DRL) controlled by intelligent agent. The DRL method transforms the trainee engagement and operational guidance calibration into a sophisticated multi objective optimization problem. This transformation facilitates the tailored refinement of training strategies in sync with each trainee's skill level, leading to the amplification of training effectiveness. Moreover, the system includes a task-independent skill assessment mechanism, employing Fr´ echet-Distance technique for the dynamic evaluation and adjustment of training strategies. Simulation results demonstrate the system's proficiency in optimizing trainee engagement and delivering skill-appropriate guidance across various competence levels.