Method and system for adaptive control of robot motion parameters based on deep reinforcement learning

Abstract

The present application provides a method and system for adaptive control of robot motion parameters based on deep reinforcement learning. The method includes: constructing an agent in a simulation environment, the agent comprising: a strategy neural network, a value neural network and a task planning module; based on guided reinforcement learning, according to sample parameters, the strategy neural network in the agent is Carry out training; based on hierarchical reinforcement learning, according to multiple subtasks and their corresponding reward functions, the strategy neural network and the value neural network in the agent are alternately carried out strategy promotion and strategy evaluation, and the trained strategy neural network is obtained. Network model; based on the trained strategy neural network model, output control parameter optimization values ​​to the controller according to the target task, so that the controller can control the robot according to the control parameter optimization values .

Description

technical field [0001] The present application relates to the technical field of robot control, in particular to a method and system for adaptive control of robot motion parameters based on deep reinforcement learning. Background technique [0002] Control parameters play an important role in the kinematic performance of quadruped robot systems, while the parameter selection of traditional controllers depends on professional domain knowledge and engineering experience. At present, some control methods based on deep reinforcement learning expect to achieve end-to-end optimization from sensor data to motor control signals, but this technical route has a long training period and difficult convergence. Stability and robustness, if the performance of the training model is not good, it can only be redesigned and trained, which greatly limits the engineering application of deep reinforcement learning technology in robot motion control. [0003] Therefore, it is necessary to provid...

AI Technical Summary

Problems solved by technology

[0004] The purpose of this application is to provide a method and system for adaptive control of robot motion parameters based on deep reinforcement learning to solve or alleviate the problems in the above-mentioned prior art

Images