A manipulator model learning method, device, electronic equipment and storage medium

Abstract

The present invention relates to the technical field of intelligent manipulators, and specifically discloses a manipulator model learning method, device, electronic equipment, and storage medium, wherein the learning method includes the following steps: acquiring multiple sets of information about the same execution task for the manipulator model to learn expert demonstration data; generating an expert policy associated with a learning cost function based on the expert demonstration data, the learning cost function being established based on the imitation cost required to imitate the expert demonstration data and the density of the expert demonstration data; minimizing The learning cost function is used to obtain the optimal expert strategy; the manipulator model is trained according to the optimal expert strategy; the optimal expert strategy finally obtained by this method pushes the manipulator model to the range where the expert demonstration data is densely distributed to imitate the expert demonstration Behavior, so that the manipulator model can accurately imitate and complete the expert demonstration behavior at the lowest possible imitation cost.

Description

technical field [0001] The present application relates to the technical field of intelligent manipulators, and in particular, to a manipulator model learning method, device, electronic device and storage medium. Background technique [0002] At present, in production applications, manipulators can enhance the versatility of autonomous interaction of manipulators and efficiently complete complex tasks through reinforcement learning. Existing reinforcement learning models generally speed up model convergence by learning optimal expert behavior strategies combined with demonstration data, but It is easy for the final reinforcement learning model to fail to accurately imitate and complete the expert demonstration behavior due to the deviation of the expert behavior strategy or only learning at the lowest imitation cost. [0003] There is currently no effective technical solution for the above-mentioned problems. SUMMARY OF THE INVENTION [0004] The purpose of the present app...

AI Technical Summary

Problems solved by technology

[0002] At present, in production applications, manipulators can enhance the versatility of autonomous interaction of manipulators through reinforcement learning and efficiently complete complex tasks; existing reinforcement learning models can generally accelerate model convergence by learning optimal expert behavior strategies combined with demonstration data, but It is easy to cause the final reinforcement learning model to fail to accurately imitate the expert's demonstration behavior due to the deviation of the expert's behavior strategy or learning only at the lowest imitation cost

[0003] For the above problems, there is no effective technical solution

Images