Driving collision avoidance optimization method based on deep reinforcement learning in car-following state

Abstract

The invention relates to the technical field of automatic driving, in particular to a driving collision avoidance optimization method based on deep reinforcement learning in a car-following state. According to the method, a model is constructed by using deep reinforcement learning, automatic driving is evaluated by setting a set of independent evaluation index system instead of carrying out evaluation by comparing the difference between the model and a human driver, and the driving risk avoiding behavior can be learned more intelligently. A driving collision avoidance strategy is provided, NGSIM empirical data can be trained, driving simulation data and the like can also be trained, and the obtained result has universality. Although the driving collision avoidance optimization method is a driving collision avoidance strategy in a car-following scene, the change of model training in a new scene is not large, and high feasibility is still kept. The driving collision avoidance optimization method adopts a relatively complex reward function, so that score evaluation is more objective and accurate.

Description

technical field [0001] The invention relates to the technical field of automatic driving, in particular to a driving collision avoidance optimization method based on deep reinforcement learning in a car-following state. Background technique [0002] In recent years, with the help of the development of current artificial intelligence technology, intelligent control systems can not only make decisions according to the current environment, but also continuously learn and adapt to the environment. Machine learning is one of the key technologies of artificial intelligence. On the basis of historical data and the continuous learning and feedback properties of current intelligent control systems, autonomous decision-making strategies based on reinforcement learning (RL) perform quite well. [0003] With the development of the automobile industry, driverless cars have gradually entered the stage of history. The most important step to achieve unmanned driving is the research and de...

AI Technical Summary

Problems solved by technology

Therefore, there are relatively many dangerous situations in the car-following process, but there are not many researches on the optimization method of collision avoidance in the state of driverless car-following. It is very meaningful to realize

Images