Human-Machine Coupling Control With PSO-Deep Q Network

In the lower limb rehabilitation exoskeleton, A pure trajectory control strategy may not be the most effective means of providing gait rehabilitation. By using the man-machine coupling force as a measure of patient participation during training, this article introduces an adaptive gait regulation control system utilizing PSO-DQN. The system aims to maintain a consistent man-machine coupling force throughout the rehabilitation training process, thereby implementing a strategy for assistance provided on control. The difference in torque obtained through torque sensors and parameter identification in the lower limb exoskeleton is utilized as an estimate for human-machine coupling force. Throughout each gait cycle, the controller adjusts the stride length and walking speed of the gait curve according to the established and target values of the human-machine coupling force. This ensures the stability of the human-machine coupling force throughout the training process. In the data from eight patients, a comparison was conducted using Q-learning, DQN, and the proposed PSO-DQN algorithm. The human-machine coupling force errors after 100 cycles of 200 epochs in the training set were 4.045 Nmi, 4.379 Nmi, and 2.340 N·m for Q-learning, DQN, and PSO-DQN, respectively. In the test set, the corresponding errors were 2.969 N·m, 4.667 N·m, and 2.473 N·m.