Résumé
Understanding human motor control remains a central challenge in biology and neuroscience, with significant implications for medicine and rehabilitation. Computational approaches, including optimal control and sequential decision-making models, have long been employed to study motor behavior. Recently, Reinforcement Learning (RL) has emerged as a promising framework for modeling motor control, supported by physics-based musculoskeletal simulations. However, the relevance of RL-derived motor policies to human motor strategies remains an open question. In this paper, we explore the potential of RL-based simulations for studying human motor control and muscular synergy dynamics under fatigue. Using MyoSuite and MuJoCo, we train Deep Reinforcement Learning (DRL) policies to perform an arm-reaching task and analyze their emergent muscle activation patterns. Our results reproduce expected behaviors in muscle synergy adaptation during fatigue, aligning with experimental findings in human studies. These findings support the validity of RL-driven simulations as a tool for investigating neuromuscular control, offering new possibilities for in-silico experimentation when human studies are limited or infeasible. The code to reproduce our experiments is available at: https://github.com/AsmaeOUHSSAIN/FEMS_RL.