Aiming at the problem of human-machine collaboration in the use of exoskeleton robots

sensors are mounted on the exoskeleton robot to collect data of five human behaviors:walking

squatting

ascending stairs

descending stairs and standing.Based on different data features of sensor data

a two-stream network-based human activity recognition(HAR) algorithm is proposed.The two-stream network is adopted to extract and learn temporal features of sensors

and the gated recurrent unit (GRU) is used for feature fusion to recognize the above five behavioral states.This enables the assistive function of the exoskeleton robot.Experimental results show that the proposed method achieves a recognition accuracy of 96.11%

which can effectively solve the human-machine collaboration issue in the use of exoskeleton robots.