Learning method, device, equipment and storage medium of behavior control strategy

Abstract

The present application discloses a behavior control strategy learning method, device, computer equipment, and storage medium. The method includes: sampling a demonstration behavior data segment including at least two demonstration behavior data from a demonstration behavior data sequence; Fragment, set the initial state information of each joint of the target object simulated in the physical simulator, and use the neural network model to be trained to determine the force data of each joint of the target object; control each joint of the target object simulated in the physical simulator Joint movement, so that the physical simulator simulates the simulation behavior data sequence of the target object based on the set action behavior limitation characteristics; according to the demonstration behavior data and simulation behavior data, the action behavior difference is determined; based on the action behavior difference, Optimize the neural network model until the optimization goal is reached. The solution of the present application is beneficial for the object of demonstration learning to generate extended action behaviors based on demonstration actions.

Description

technical field [0001] The present application relates to the field of computer technology, and in particular to a learning method, device, equipment and storage medium of a behavior control strategy. Background technique [0002] Demonstration learning is an autonomous learning technology that takes demonstration behavior as the goal. In demonstration learning, the object to learn skills is required to imitate the behavior of the demonstration, so that the object can acquire the motor skills corresponding to the demonstration behavior. Among them, in different application fields, the objects of skills to be learned will also be different. For example, in the field of games, the objects to learn skills can be characters, animals, etc. in the game; and for example, in the field of robot control, the objects to learn skills can be robots. [0003] At present, in the demonstration learning process, behavior control strategies can be learned from several groups of demonstration...

AI Technical Summary

Problems solved by technology

[0004] However, in the existing demonstration learning process, if you want the object to learn the skill to have a certain motor skill, you need to obtain the action demonstration data corresponding to the motor skill in advance; if you lack the corresponding action demonstration data, you cannot make the object Possess corresponding motor skills, resulting in high complexity for the object to learn the skill to generate a certain skill

Images