Method for establishing autonomous mobile robot navigation system through virtual environment

Abstract

The invention discloses a method of establishing navigation for navigation robot based on deep reinforcement learning in an unknown environment. The method is realized through the following mode: first, expressing a virtual environment, the virtual state of a mobile robot agent, starting and ending point information and behaviors, then constructing a deep reinforcement learning structure based onthe DQN algorithm, training the mobile robot agent in the virtual environment, updating DQN network parameters, and finally migrating a training strategy through a progressive neural network, and constructing the mobile robot intelligent navigation system. The method for establishing the autonomous mobile robot navigation system through the virtual environment not only has relatively good flexibility and universality, but also establishes a complete one-stop solution for establishing the navigation system for the navigation robot.

Description

technical field [0001] The invention relates to a method for establishing navigation of a navigation robot in an unknown environment based on deep reinforcement learning, and belongs to the field of robot navigation and obstacle avoidance. Background technique [0002] Robot intelligent navigation is a crucial research content in robot research, which means that a mobile robot can independently plan a safe and fast destination without collision in an unknown environment through autonomous learning. [0003] At present, the navigation of mobile robots based on unknown environments generally adopts the method of local path planning. Most of these methods require all or part of prior environmental knowledge and corresponding training data, so they are not flexible enough to deal with dynamic scenes well. [0004] The reinforcement learning method based on partially observable Markov decision process modeling adopts the method of trial and error learning to iterate the optimal s...

AI Technical Summary

Problems solved by technology

[0003] At present, the navigation of mobile robots based on unknown environments generally adopts the method of local path planning, most of which require all or part of the prior environment knowledge and corresponding training data, so the flexibility is not enough to deal with dynamic scenes well.

[0004] The reinforcement learning method based on partially observable Markov decision process modeling adopts the method of trial and error learning to iterate the optimal strategy. Therefore, in the scenario of consuming more resources, the utilization rate of data is very low and the loss is large.

In the real environment, it is often difficult to effectively obtain such a large amount of training data online.

Images