Unmanned vehicle obstacle avoidance method and device based on reinforcement learning

Abstract

The invention discloses an unmanned vehicle obstacle avoidance method and device based on reinforcement learning, and the method comprises the steps of training a dynamic window obstacle avoidance model under a pre-obtained environmental constraint through a reinforcement learning algorithm, and obtaining a prediction window; sampling multiple sets of velocities based on the prediction window and velocity sampling constraints, and forming a set of trajectories for each set of sampled velocities; scoring the multiple groups of tracks by using a pre-constructed obstacle avoidance penalty function and an evaluation function; selecting the track with the highest score from each group of tracks as a target track; outputting the sampling speed corresponding to the target track to a control system of the unmanned vehicle, so that the unmanned vehicle is driven according to the sampling speed corresponding to the target track; and training the dynamic window obstacle avoidance model by utilizing reinforcement learning, and selecting a track with relatively high obstacle avoidance performance from each group of tracks by utilizing an obstacle avoidance penalty function. Therefore, by means of the method, the obstacle avoidance performance of the unmanned vehicle can be improved, and active obstacle avoidance of the unmanned vehicle is achieved.

Description

technical field [0001] The present application relates to the technical field of automatic driving, in particular to a reinforcement learning-based method and device for obstacle avoidance of unmanned vehicles. Background technique [0002] With the rise of artificial intelligence technology, autonomous driving technology has become more and more mature, and unmanned vehicles have also received extensive attention from academia and industry. Unmanned vehicles involve various fields, such as information and sensor technology, trajectory tracking technology, and obstacle avoidance technology. Among them, obstacle avoidance technology, as the basis of unmanned vehicles, has become the focus of attention of those skilled in the art. [0003] At present, the commonly used obstacle avoidance technology is the dynamic window method. However, the traditional dynamic window method lacks the ability to adapt to complex driving environments. When encountering obstacle clusters, there...

AI Technical Summary

Problems solved by technology

However, the traditional dynamic window method lacks the ability to adapt to complex driving environments. When encountering obstacle clusters, there will be problems such as oscillation and other traps, making it impossible for unmanned vehicles to avoid obstacle clusters, and its obstacle avoidance performance is relatively low. , it is impossible to realize the active obstacle avoidance of unmanned vehicles in the true sense

Images