Priori knowledge guided robot operation skill learning system and method

Abstract

The invention provides a priori knowledge guided robot operation skill learning system and method. The system comprises a physical environment module, an evaluation module and a robot operation skillpolicy learning module. The physical environment module collects robot operation environment information and a current operation state. The evaluation module evaluates a robot search operation policyby means of priori knowledge of an operation process and feeds back rewards and punishment values. The robot operation skill policy learning module learns an operation policy based on collected environment information, operation state and fed back rewards and punishment values to establish an expert knowledge base where operation contact states and actions of a mechanical arm are mapped one by oneso as to output a robot action decision, so that the learning efficiency is improved and the learning difficulty is reduced effectively.

Description

technical field [0001] The disclosure belongs to the technical field of machine learning, and in particular relates to a robot operation skill learning system and method guided by prior knowledge. Background technique [0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art. [0003] Reinforcement learning has become the main method for learning robot operation skills. It imitates human learning methods, uses sensor feedback information, and continuously interacts with the environment to obtain external reward or punishment signals without explicit teaching. Adapt to changing circumstances and take appropriate actions. [0004] However, the inventor found in the research that most of the current learning methods are completely explored from scratch, and the training time is relatively long, which reduces the efficiency of robot skill learning. Contents of the invention [0...

AI Technical Summary

Problems solved by technology

[0004] However, the inventor found in the research that most of the current learning methods are completely explored from scratch, and the training time is longer, which reduces the efficiency of robot skill learning

Images