Reinforcement learning-based motion imitation for physiologically plausible musculoskeletal motor control