Reinforced learning based robot joint motion control method and system

Abstract

The invention discloses a reinforced learning based robot joint motion control method and system. The method comprises the following steps: obtaining to-be-operated track of a robot terminal; calculating position increment within each interpolation period of the robot joint according to the to-be-operated track of the robot terminal and a robot inverse kinematic model; determining position increment compensation within each interpolation period of the robot joint according to a policy network; taking sum of given position increment within each interpolation period and position increment compensation as motion parameters of the robot joint, inputting the motion parameters into a robot to obtain practical motion amount within each interpolation period of the robot joint; performing real-timetraining update on a value network according to the given position increment and the practical motion amount; after operation of the to-be-operated track is accomplished, performing training update on the policy network according to parameters updated according to each interpolation period, of the value network; and controlling motion, in next to-be-operated track, of the robot joint by adoptingthe updated policy network. The reinforced learning based robot joint motion control method has the characteristics of being small in errors and high in efficiency.

Description

technical field [0001] The invention relates to the field of robot control, in particular to a robot joint motion control method and system based on reinforcement learning. Background technique [0002] There is a problem of trajectory deviation at the end of the robot during operation, and reinforcement learning can be used to compensate and reduce the deviation. There are currently two main types of reinforcement learning applications in this field. One is the method based on the value function, also known as critic-only, which is to derive the corresponding optimal strategy by observing and evaluating the performance of the system. The disadvantage of this method is that the errors will continue to accumulate, and slight changes in the strategy will have a great impact on the value function. The other is a method based on strategy search, also known as actor-only. This method directly improves the strategy. This method performs well in solving continuous state problems ...

AI Technical Summary

Problems solved by technology

The disadvantage of this method is that the errors will continue to accumulate, and slight changes in the strategy will have a great impact on the value function

The other is a method based on strategy search, also known as actor-only. This method directly improves the strategy. This method performs well in solving continuous state problems and is widely used in the field of robotics. However, only the current strategy is considered in the training process. Or several adjacent strategies, therefore, there is a local optimal problem

Images