Autonomous obstacle avoidance system and method for UAV based on deep reinforcement learning

Abstract

The invention discloses a UAV autonomous obstacle avoidance system and method based on deep reinforcement learning. Through a novel system architecture, training and decision-making are separated, which can greatly reduce training time consumption and improve the timeliness of aircraft decision-making; the autonomous obstacle avoidance method adopts a method based on The deep reinforcement learning model of strategy iteration uses the original RGB image taken by the drone monocular camera as the training data, without the need for complex point clouds and other 3D information, and obtains the depth by training the original RGB image through a fully convolutional neural network Image information, and then analyze and predict the image through the reinforcement learning method based on strategy iteration, predict the flight action of the drone at the next moment in advance, and realize autonomous obstacle avoidance. The training time consumption of the obstacle avoidance method proposed by the present invention is more efficient and less time-consuming than the existing typical value-based iteration method, and can realize flexible and autonomous obstacle avoidance, and is suitable for high-speed inspections such as substation automatic inspections and drone cruises. Required autonomous obstacle avoidance scenarios.

Description

technical field [0001] The invention relates to an unmanned aerial vehicle obstacle avoidance system and method, in particular to an unmanned aerial vehicle autonomous obstacle avoidance system and method based on deep reinforcement learning, and belongs to the technical field of unmanned aerial vehicle flight control. Background technique [0002] Obstacle avoidance is one of the core problems of UAVs, and its goal is to allow UAVs to autonomously explore unknown environments to avoid collisions with other objects in order to obtain a flight path that can avoid threats and reach the target safely. Traditional obstacle avoidance technology detects traversable spaces and obstacles, and then performs path planning. The data information used is captured by RGB-D cameras, light detection, ranging sensors (LIDAR), and even sonar. These traditional obstacle avoidance technologies can be well applied to the autonomous obstacle avoidance of ground robots, but it is difficult to appl...

AI Technical Summary

Problems solved by technology

Although these works can scale the trained networks to the real world, they still require a lot of computing resources to generate huge datasets and train them

Images