Anti-collision control method based on depth reinforcement learning

Abstract

The invention discloses an anti-collision control method based on depth reinforcement learning. A depth deterministic policy gradient method (DDPG algorithm) is adopted for depth reinforcement learning, and the anti-collision control method includes the steps that first, vehicle parameters and environmental vehicle parameters are extracted; second, a virtual environment model is constructed through the vehicle parameters and the environmental vehicle parameters; third, according to the vehicle parameters, the environmental vehicle parameters and the virtual environment model, basic parametersof the depth deterministic policy gradient method are defined; fourth, according to the basic parameters defined in the third step, a neural network in depth reinforcement learning is used for constructing an anti-collision control decision making system, and the anti-collision control decision making system comprises a strategy network and an evaluation network; and fifth, the strategy network and the evaluation network are trained, and the anti-collision control decision making system is obtained. According to the anti-collision control method based on depth reinforcement learning, the anti-collision control decision making system based on the depth neural network and the anti-collision control decision making system for constantly optimizing network control results based on a time difference reinforcement learning method are constructed, and the control performance of the anti-collision control decision making system is improved effectively.

Description

technical field [0001] The invention relates to the technical fields of assisted driving and automatic driving, and in particular to an anti-collision control method based on deep reinforcement learning. Background technique [0002] When the vehicle is running, keeping a stable safe distance from the vehicle in front can effectively prevent the occurrence of collision accidents. With the increase of the number of cars, the density of cars on the road is increasing, so the forward collision warning of vehicles is particularly important. The anti-collision control system can judge the vertical and horizontal distances, vertical and horizontal relative speeds and orientations between the vehicle 1 and the vehicle in front 2 and the vehicle in front of the side lane. When there is a potential collision risk, the vehicle can be controlled to a certain extent, which can effectively reduce the collision between the vehicle and the vehicle. The collision accident of the vehicle in...

AI Technical Summary

Problems solved by technology

Although this patent can accurately realize collision warning, the cost is too high, which is not conducive to the general promotion and use

[0005] To sum up, in the vehicle anti-collision control system, the image of the front situation is generally obtained through the camera, the feature value is extracted to identify vehicles, pedestrians, etc., and the distance and speed information are obtained, and then the danger is judged. This system not only needs to design complex The anti-collision control decision-making system model has weak self-adaptive ability, does not have self-learning ability, and cannot adapt to the warning requirements in different environments; at the same time, the current forward collision warning system does not consider the impact of vehicles in the two lanes next to the vehicle to make forward collision warning decisions

Images