Robot arm inverse model learning method based on self-supervised learning

The invention discloses a robot arm inverse model learning method based on self-supervised learning. The method comprises the following steps: acquiring a predicted position p (t) of the tail end of a robot arm; reasoning a joint angle q (t) required by the tail end of the robot arm to reach the predicted position p (t) based on an inverse model; predicting a tail end position p (t + 1) corresponding to the joint angle q (t) based on a positive model; reasoning a joint angle q '(t) required by the tail end of the robot arm to reach the predicted position p (t + 1) based on the inverse model, taking the joint angle q (t) as supervision information of the joint angle q' (t), and updating parameters of the inverse model by using a gradient descent method; and repeating the steps until the difference between the joint angle q (t) and the joint angle q '(t) is smaller than a set value to obtain a trained inverse model. According to the method, the self configuration and state of the robot arm are used for coordina