Active man-machine cooperation method based on multi-modal information of human body

The invention discloses an active man-machine cooperation method based on multi-modal information of a human body. The method comprises the following steps: collecting signals of a human and a robot which work cooperatively; inputting the joint angle signal into a constructed prediction neural network for processing to obtain a motion prediction track; the electromyographic signals are input into a constructed estimation neural network to be processed, and three-dimensional arm strength is obtained; inputting the three-dimensional arm power into a refined sensing interface of a human cooperation state to obtain a cooperation comfort index and a stability quantitative index; weighting the collaborative comfort index and the stability quantitative index to obtain an optimization target, obtaining an optimal robot impedance parameter by using an exploration iterative algorithm, and setting an impedance model of the robot based on the impedance parameter to obtain an expected trajectory; and a PD position control