Robot confrontation method based on evolutionary reinforcement learning

Abstract

The invention discloses a robot confrontation method based on evolutionary reinforcement learning. The robot confrontation method comprises the following steps: constructing a robot confrontation strategy based on deep reinforcement learning, wherein the confrontation strategy is a depth deterministic strategy gradient algorithm DDPG; forming an evolutionary depth deterministic strategy gradient algorithm EDDPG in combination with an evolutionary algorithm and the confrontation strategy; and the strategy network model trained by the algorithm DDPG and the algorithm EDDPG being used as a controller to control the robot to confront with the robot controlled by the confronting strategy based on the threat index. According to the method, an evolutionary reinforcement learning method is considered and used, the evolutionary reinforcement learning method is a combination of deep reinforcement learning and an evolutionary thought, the deep reinforcement learning method does not need modeling, end-to-end control over the robot can be achieved, certain generalization ability is achieved, and the problems existing in an existing method can be effectively solved; the evolutionary idea is to improve the exploration ability and stability of reinforcement learning by using the population.

Description

technical field [0001] The invention relates to the technical field of robot confrontation, in particular to a robot confrontation method based on evolutionary reinforcement learning. Background technique [0002] Robot confrontation is a kind of zero-sum game problem with the robot as the carrier and the goal of defeating the opponent under specific rules. Considering the kinematics and dynamics characteristics of the robot itself, robot confrontation is essentially a differential game problem, which is extremely difficult to solve. The current theoretical methods can only solve the differential game problems of some simple models, and the theoretical methods and technologies of robot confrontation still need new breakthroughs. [0003] For the problem of robot confrontation, the existing methods are mainly divided into two categories: differential game and artificial intelligence. The method of differential game can obtain the analytical solution of the robot confrontatio...

AI Technical Summary

Problems solved by technology

[0003] For the problem of robot confrontation, the existing methods are mainly divided into two categories: differential game and artificial intelligence. The method of differential game can obtain the analytical solution of the robot confrontation strategy by establishing and solving the differential equation. However, their modeling process is very complicated, and Without generalization ability, it can only be applied to specific simple environments; artificial intelligence methods are often difficult to directly deal with the continuous control problems of robots, and need to rely on prior knowledge or other underlying control methods, with poor exploration ability

Images