AUV (Autonomous Underwater Vehicle) autonomous obstacle avoidance method based on reinforcement learning

Abstract

The invention discloses an AUV (Autonomous Underwater Vehicle) autonomous obstacle avoidance method based on reinforcement learning. The method comprises the following steps: firstly, setting an AUV seabed navigation autonomous obstacle avoidance logic rule; if the foresight sonar cannot detect the obstacle, enabling the AUV to keep the existing navigation height to pass through the obstacle; if the foresight sonar detects the obstacle and the height of the obstacle can be determined, determining the height of the obstacle, and pulling up the AUV to enable the navigation height of the AUV to exceed the obstacle to pass through the obstacle; if the foresight sonar detects an obstacle and the height of the obstacle cannot be determined, enabling the AUV to start an autonomous yaw strategy and bypasses the obstacle in a yaw mode; and enabling the AUV autonomous yaw strategy to adopt a reinforcement learning model based on strategy gradient to determine an optimal yaw angle to bypass an obstacle. According to the method, various factors such as the task execution time, the angle deviating from the main route and the obstacle distance are comprehensively considered, and optimal strategy obstacle avoidance which is autonomously achieved under various constraints is achieved.

Description

technical field [0001] The invention belongs to the technical field of underwater vehicles, and in particular relates to an AUV autonomous obstacle avoidance method. Background technique [0002] For AUV, its navigation and obstacle avoidance capabilities are the most important part of its basic indicators. Developing its autonomous detection and avoidance capabilities is an essential step in improving its performance and potential compared to relying on humans for remote control. However, due to the complex underwater terrain, relying on forward-looking sonar and simple left and right fixed value yaw circumnavigation and obstacle avoidance logic will easily cause the AUV to deviate from the scheduled route, take more time, and be more likely to encounter obstacles. [0003] Traditional planning obstacle avoidance methods include visualization method, grid method, artificial potential field method and so on. The visualization method is to point-line the entire environment,...

AI Technical Summary

Problems solved by technology

This method has the advantages of simple algorithm and high execution efficiency. Its disadvantage is that when there are obstacles near the target point, the gravitational force tends to be 0 but the repulsive force is not 0, which makes it difficult for the AUV to approach the target point. Secondly, there may be At the position where the resultant force is 0, the AUV may move to the dead zone, then start to oscillate in place, and fall into a stagnant state; in addition, the path obtained by pushing the AUV forward through the virtual potential field has not been analyzed and evaluated, so most of the results are not global optimal untie

Images