Collision avoidance planning method for mobile robots based on deep reinforcement learning in static environment

Abstract

The invention belongs to the technical field of mobile robot navigation, in particular to a collision avoidance planning method for mobile robots based on deep reinforcement learning in a static environment. The present invention uses a laser rangefinder to collect raw data, takes the processed data as the state S of the A3C algorithm, constructs an A3C-LSTM neural network, takes the state S as the network input, and uses the A3C algorithm to output the corresponding parameters from the neural network. The actions performed by the mobile robot at each step are selected by the normal distribution. The invention does not need to model the environment, and finally realizes the successful obstacle avoidance of the mobile robot in the complex static obstacle environment through the deep reinforcement learning algorithm. The present invention designs a continuous action space model with turning constraints, and adopts multi-threaded asynchronous learning. Compared with the general deep reinforcement learning method, the learning and training time is greatly improved, the sample correlation is reduced, and the high utilization of the exploration space and the exploration strategy are guaranteed. The diversity of the algorithm improves the convergence, stability and obstacle avoidance success rate of the algorithm.

Description

technical field [0001] The invention belongs to the technical field of mobile robot navigation, in particular to a collision avoidance planning method for mobile robots based on deep reinforcement learning in a static environment. Background technique [0002] The application of mobile robots has penetrated into all fields of life. In industrial production lines, robots can replace humans to perform heavy and repetitive work, improve production efficiency and liberate human labor. [0003] Obstacle avoidance is a key issue in robotics research during multi-robot task execution. It is a dynamic, real-time, uncertain typical environment model. Furthermore, it requires robots to adapt quickly and independently to changes in the environment. For obstacle avoidance problems, behavior-based response control methods and rule-based control strategies are often used. Both methods are simple, deterministic, and the robot responds better to the environment. But when tasks and envir...

AI Technical Summary

Problems solved by technology

Generally speaking, traditional obstacle avoidance methods have relatively large limitations, especially for complex and dynamic unknown environments; while intelligent obstacle avoidance algorithms, especially the popular obstacle avoidance algorithms combined with deep learning and reinforcement learning in recent years, are suitable for continuous High-dimensional complex dynamic unknown environment has great advantages

Images