Intelligent parking method based on model reinforcement learning

Abstract

The invention relates to an intelligent parking method based on model reinforcement learning, which uses Monte Carlo tree search, a vehicle kinematics model, an action classification network and a state value fitting network, and specifically comprises the following steps: S1, combining the Monte Carlo tree search with the action classification network and the vehicle kinematics model to obtain aparking data pre-training model; S2, training a state value fitting network according to parking data generated by the parking pre-training model; S3, combining the trained state value fitting networkwith a Monte Carlo tree search and action classification network to form an online driving strategy model; and S4, enabling the parking on-line driving strategy model to receive the parking space andthe vehicle motion information in a rolling time domain mode, generate a control instruction at each time interval, and send the control instruction to a vehicle motion control module to control thetarget vehicle to complete parking. Compared with the prior art, the method has the advantages that the final parking course angle and success rate are better, the influence of the accuracy of the vehicle model on the final parking effect is reduced, and the like.

Description

technical field [0001] The invention relates to the technical field of automatic parking, in particular to an intelligent parking method based on model reinforcement learning. Background technique [0002] Automated Parking Systems (APS) are important intelligent driver assistance systems because they have great potential to reduce accidents in cramped cities and increase the utilization of parking spaces. For all APS platforms, the smart vehicle must have a parking space that detects the location and generates its movement by on-board sensor systems, such as around view cameras (AVM) and LiDAR (LIDAR). The conventional motion planning method of APS is the path velocity decomposition method, which decomposes the parking task into a kinematics sub-problem and a dynamics sub-problem, which are solved by path planning and path tracking methods respectively, but cannot flexibly process real-time perception information, and cannot use historical parking Vehicle data to improve i...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to provide a smart parking method based on model reinforcement learning in order to overcome the defects of the above-mentioned prior art that are overly dependent on the accuracy of the vehicle model, the information processing is not flexible enough, and the existing data cannot be effectively used

Images