Robot control method, device and equipment and computer storage medium

Abstract

An embodiment of the invention provides robot control method, device and equipment and a computer storage medium, and relates to the technical field of artificial intelligence. The method comprises the steps that environment interaction data and an actual target value actually completed after an action corresponding to action data in the environment interaction data is executed are obtained; a reward value after the action is executed is determined according to the state parameter of the first moment in two adjacent moments, the action data and the actual target value; a reward value in the environment interaction data is updated by the aid of the reward value after the action is executed; an intelligent agent corresponding to a robot control network is trained by the aid of the updated environment interaction data; and the action of the target robot is controlled by the aid of the trained intelligent agent. Through the robot control method, device and equipment and the computer storage medium, the data utilization rate can be increased, training of the intelligent agent can be accelerated, a large number of targets can be trained at the same time, and all tasks in a certain targetspace can be completed through one model.

Description

technical field [0001] The embodiment of the present application relates to the technical field of the Internet, and relates to but not limited to a robot control method, device, equipment and computer storage medium. Background technique [0002] At present, when controlling the robot, one implementation method is the deep reinforcement learning control algorithm based on the priority experience playback mechanism, which uses the state information of the object operated by the robot to calculate the priority, and uses the deep reinforcement learning method to complete the end-to-end The robot control model allows deep reinforcement learning agents to learn independently in the environment and complete specified tasks; another implementation method is a kinematic self-grasping learning method based on simulated industrial robots, which belongs to the field of computer-aided manufacturing and is based on simulation environment and use reinforcement learning theory to carry ou...

AI Technical Summary

Problems solved by technology

[0003] However, the implementation methods in related technologies usually need to train a model to complete a feature task, the generalization is not strong, and the training process of robot tasks is slow

Images