Robot path navigation method and system based on deep reinforcement learning

Abstract

The invention discloses a robot path navigation method and system based on deep reinforcement learning. The method comprises the following steps: constructing a double Actor-Critic neural network based on deep reinforcement learning, and outputting an initial movement action of a robot and an evaluation value of the initial movement action by adopting a first Actor-Critic neural network accordingto an obtained current movement state of the robot; training a second Actor-Critic neural network by taking the evaluation values of the current movement state and the initial movement action of the robot as a training set; and updating the first Actor-Critic neural network according to the trained second Actor-Critic neural network, and outputting an optimal moving action by using the updated first Actor-Critic neural network according to the current movement state of the robot, thereby performing optimal path navigation on the robot. The optimal action strategy under the current movement state of the robot is found in combination with the perception capability of the deep learning method and the strategy capability of the reinforcement learning method, and the limitation that traditionalrobot navigation depends on an obstacle map is eliminated in a highly complex scene.

Description

technical field [0001] The invention relates to the technical field of path planning, in particular to a robot path navigation method and system based on deep reinforcement learning. Background technique [0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art. [0003] Motion planning consists of path planning and trajectory planning. The sequence of points or curves connecting the starting point and the end point is called a path, and the strategy that forms the path is called path planning. The non-collision movement of the robot, that is, the robot navigation is also a kind of path planning. Traditional path planning algorithms include: simulated annealing algorithm, artificial potential field method, fuzzy logic algorithm, tabu search algorithm, etc. [0004] Deep reinforcement learning combines the perception ability of deep learning and the decision-making ability of re...

AI Technical Summary

Problems solved by technology

[0005] In addition, the inventor believes that the traditional robot navigation method mainly depends on the obstacle map of the navigation environment, and the navigation process requires all or part of the prior environmental knowledge, which has poor flexibility and large limitations, and is not suitable for complex and dynamic unknowns. environment; and due to the huge difference between the simulated environment and the highly complex real environment, it is difficult to convert the trained model into real robot navigation

Images