Deep reinforcement learning obstacle avoidance navigation method fusing global training

Abstract

The invention discloses a deep reinforcement learning obstacle avoidance navigation method fusing global training. The method comprises the steps that: an initial path is planned according to known information; a robot moves from an initial point to a target point along the initial path; a temporary target is selected, and the robot moves towards the temporary target; an accumulated expected reward and punishment value is set, navigation is conducted continuously for multiple times, and a reward and punishment value is given to each frame in the robot navigation process; an interaction tuple is obtained from a sensor; the interaction tuple is input into a deep learning network for training; and a to-be-navigated interaction tuple is input into the trained deep learning network, an optimal path and an accumulated expected reward and punishment value are output, and the robot moves according to the optimal path. According to the method, the convergence speed of deep reinforcement learning can be effectively improved, the navigation efficiency of the robot is improved in the navigation process, meanwhile, the motion of the robot is environment-friendly, and the influence on the surrounding environment is reduced to the minimum.

Description

technical field [0001] The invention relates to a dynamic environment obstacle avoidance navigation method for a robot, in particular to a deep reinforcement learning obstacle avoidance navigation method integrated with global training. Background technique [0002] For automatic navigation robots working in dynamic environments such as delivery robots and indoor service robots, obstacle avoidance navigation is a very important function. The robot must reach the target point safely and quickly while avoiding various obstacles. The research on obstacle avoidance for static obstacles is relatively mature, and the problem of obstacle avoidance navigation for dynamic obstacles is more complicated, because it needs to predict the unknown surrounding dynamics (such as pedestrians, vehicles or other robots). As the complexity of the surrounding environment increases, the passable area of ​​traditional navigation methods will become smaller and smaller until the robot cannot plan a...

AI Technical Summary

Problems solved by technology

However, problems such as poor convergence of deep reinforcement learning and weak environmental adaptability also restrict the popularization and application of this method.

Images