Unmanned ship hybrid sensing autonomous obstacle avoidance method and system based on reinforcement learning

Abstract

The invention relates to an unmanned ship hybrid sensing autonomous obstacle avoidance method and system based on reinforcement learning. The method comprises the following steps: 1) building a marineenvironment; 2) setting an action space according to the condition of an unmanned ship propeller, and performing learning according to global planning information provided by a static chart and obstacle information in a radar system detection radius range to obtain a reinforcement learning state code; 3) setting a reward target weight to obtain a comprehensive reward function; 4) establishing andtraining an evaluation network and a strategy network; and 5) respectively inputting the reinforcement learning state codes into the evaluation network and the strategy network, inputting the comprehensive reward function into the evaluation network, and determining the output of the controller according to the action corresponding to the learned average value of the strategy network. Compared with the prior art, the method and sysem have high self-learning capability, can adapt to different large-scale complex environments through simple deployment training, and further realize autonomous perception, autonomous navigation and autonomous obstacle avoidance.

Description

technical field [0001] The invention relates to an autonomous obstacle avoidance method and system for an unmanned boat, in particular to a mixed-sensing autonomous obstacle avoidance method and system for an unmanned boat based on reinforcement learning. Background technique [0002] The unmanned boat is a surface unmanned vehicle that can realize autonomous navigation, autonomous obstacle avoidance, and autonomous surface operations. It has the advantages of small size, high speed, good stealth, and no risk of casualties. Unmanned boats are very suitable for carrying out surface operations tasks in dangerous sea areas that have a greater risk of casualties or simple surface operations tasks that require low personnel participation, and have a good cost-effectiveness ratio, so they are widely and effectively used in ocean monitoring , marine survey, maritime search and rescue, unmanned cargo and other fields. [0003] At present, the more mainstream way to realize autonomo...

AI Technical Summary

Problems solved by technology

Although these methods have relatively good performance in their respective application backgrounds, they need to be carefully designed for different functional modules, and the comprehensive algorithm needs to be configured and adjusted as a whole, which makes the realization of the intelligent algorithm of the unmanned vehicle complex and cumbersome.

Furthermore, due to the lack of self-learning ability of these methods, it is difficult to adapt to large-scale and complex environments, and it is necessary to redesign, reorganize and cooperate with different algorithm modules for different environments.

Images