A Reinforcement Learning-Based Computational Model of Human Elbow Joint Operation for Effective Human-Machine Interface

Designing effective human-machine interfaces requires understanding complex interactions between humans and machines. However, even bi-directional communication that relies on sensor data is often insufficient due to the unpredictability of human behaviors. To better anticipate these behaviors, a computational model that reflects the operating principles of the human nervous system is essential. This study presents a computational neuromechanical model that focuses on human sensorimotor operations, particularly in the context of elbow joint rotation. The model leverages reinforcement learning (RL) to simulate the brain's reward mechanism in motor adjustment, with the reward function adjusted by a multiplication constant (M) that reflects individual variability in sensorimotor processing. Measurement and simulation data were evaluated based on their overlap ratios. The test results demonstrate that the RL-based model, when calibrated with an optimal M value, closely matches with measured motor outputs, indicating its potential for improving the effectiveness of the human-machine interface.