Unmanned ship real-time obstacle avoidance algorithm based on deep reinforcement learning

Abstract

The invention discloses an unmanned ship real-time obstacle avoidance algorithm based on deep reinforcement learning, and relates to the technical field of unmanned ships. According to the invention,a deep learning method is adopted to obtain image information, on the basis of an A3C algorithm, a network structure is optimized, obstacle avoidance information is enriched, and according to three requirements of path planning, obstacle avoidance and environment exploration and adaptation, the action space of an intelligent agent is re-standardized, and three kinds of environments are selected for training; and in combination with a GPU platform, the pre-training data is integrated into the deep neural network, so that the training efficiency is improved, and the real-time performance of thealgorithm is ensured. The result shows that the method shortens the training time by 59.3% and improves the efficiency by more than 71.7% while meeting the single processing speed requirement, and theperformance of the algorithm model in an unknown environment is effectively improved and is superior to the existing scheme.

Description

technical field [0001] The invention relates to the technical field of unmanned boats, in particular to a real-time obstacle avoidance algorithm for unmanned boats based on deep reinforcement learning. Background technique [0002] As an autonomous surface unmanned vehicle, the surface unmanned vehicle has broad application prospects in military and civilian fields due to its small size, high intelligence, and ability to complete tasks independently. As one of the important criteria to measure its intelligence, the local obstacle avoidance technology of unmanned boats requires the unmanned boats to perceive and judge the surrounding unknown and known environments within a certain range, and to be able to quickly avoid obstacles. Finally arrived at the designated location safely. Unmanned boats usually use infrared, camera, ultrasonic and laser sensors as information acquisition sources. With the advancement of hardware technology and manufacturing capabilities, the accuracy...

AI Technical Summary

Problems solved by technology

[0004] (1) Environmental information is complex and highly dynamic. Unmanned boats will be disturbed by natural environments such as wind, waves, and currents when navigating on the water surface. These disturbances are highly dynamic and difficult to predict; in addition, complex environmental information also includes Static obstacles such as buoys and signal lights and dynamic obstacles such as other ships may be encountered during the course of travel. Therefore, the local path planning of unmanned vehicles needs to consider multiple environmental factors, and quickly extract effective information from complex and changeable environments. make a response;

Images