Preliminary Results of Lightweight Reinforcement Learning Using FEMC for Human-Robot Collaboration

This paper introduces a lightweight Reinforcement Learning (RL) algorithm based on the Fuzzy Encoded Markov Chain (FEMC) for robot manipulation. This model-free, FEMC-based RL algorithm compresses the state and action spaces, enhancing decision-making efficiency. Two case studies are presented to demonstrate the algorithm's effectiveness. In one study, a double-integrator system used in robot motor control is trained with our algorithm, resulting in smooth motor control achieving the desired position and velocity. Additionally, the efficiency of our algorithm is validated in the reacher environment, where it rapidly converges with limited exploration. The algorithm's efficiency can significantly benefit applications that require rapid online decision-making, such as industrial production involving robots or human-robot collaboration.