Automatic parking trajectory planning method based on deep enhanced learning

Abstract

The invention relates to an automatic parking trajectory planning method based on deep enhanced learning. The method comprises the steps that a two-dimensional virtual parking environment is established according to information obtained by a visual system; parking movement of a vehicle is simulated to obtain the steering angles of a steering wheel; a series of corresponding steering wheel steeringangles in the automatic parking process are obtained by establishing an automatic parking model based on deep enhanced learning and adopting a deep enhanced learning method, and an automatic parkingtrajectory is generated; whether or not collision of a vehicle takes place in the virtual parking environment is judged, if yes, the two-dimensional virtual parking environment is initialized, the next parking training is started, otherwise the next step is started; the automatic parking trajectory is planned, and the length and time consumption of the trajectory are comprehensively considered toscreen out the optimal automatic parking trajectory. The method has the advantages that the method can be applied to different parking environments, the planned trajectory is more reasonable and optimized, a controlling strategy is combined, the vehicle can follow the trajectory to rapidly enter a parking space automatically without being operated by a driver, and the accuracy and the reliabilityare improved.

Description

technical field [0001] The present invention relates to the technical field of automobile unmanned driving, in particular to an automatic parking trajectory planning method based on deep reinforcement learning. Background technique [0002] The automatic parking of the car is an important part of the unmanned driving system of the car. The vehicle can automatically park into the parking space without the driver's operation. Existing automatic parking methods are mainly divided into two categories: one based on millimeter-wave radar, and the other based on vision systems. Among them, the method based on the millimeter-wave radar has a relatively large ranging error for objects in the middle and short distances, and the relative cost is relatively high. The above two types of methods mainly use the traditional PID control principle to determine the steering wheel angle, vehicle position and control error accumulation for the path planning part, which is easily affected by sur...

AI Technical Summary

Problems solved by technology

The above two types of methods mainly use the traditional PID control principle to determine the steering wheel angle, vehicle position and control error accumulation for the path planning part, which is easily affected by surrounding environmental factors, and it is difficult to achieve precise control, and the simulation experiment data is relatively small. Therefore, the accuracy and reliability are not ideal

Images