Speed control multi-target optimized car following algorithm of automatic driving vehicle

Abstract

The invention discloses a speed control multi-target optimized car following algorithm of an automatic driving vehicle. According to the algorithm, a car following speed control model is provided on the basis of deep reinforced learning, and the model not only simulates human driving but also optimizes the driving safety, efficiency and comfortableness. Collision time, heat time-distance experience distribution and jerk are combined to construct a reward function reflecting the driving safety, efficiency and comfortableness, a practical driving data training model in an NGSIM (Next GenerationSimulation) project is used, a car following behavior simulated by the model is compared with a behavior observed in NGSIM experience data, test and error test of a learning intelligent agent in the simulated environment is reinforced, and safe, comfortable and efficient vehicle speed control is learned in a maximally reward accumulation way. It is proved that, the car following speed control algorithm provides a safer, more efficient and more comfortable driving capability compared with a human driver in the real world.

Description

technical field [0001] The invention relates to the field of car-following control for automatic driving, in particular to a car-following algorithm for multi-objective optimization of speed control of an automatic driving vehicle. Background technique [0002] Car-following control is an important part of intelligent decision-making for autonomous driving, including speed selection under free driving, maintaining the distance between vehicles when following, and braking in emergency situations. In the case of the coexistence of autonomous driving and human driving, automatic driving vehicles make car-following control decisions similar to human drivers (referred to as anthropomorphic), which will improve the comfort and trust of passengers, and at the same time facilitate other traffic participants. Understand and predict the behavior of autonomous vehicles to enable safe interactions between autonomous and human drivers. However, the traditional car-following model has ma...

AI Technical Summary

Problems solved by technology

However, the traditional car-following model has many limitations when applied to automatic car-following control, such as limiting the flexibility and accuracy of the model, it is difficult to extend to driving scenarios and drivers other than the calibration data, and it cannot reflect the actual situation of the vehicle when applied to automatic driving. Driver's driving style and driving scene, etc.

Images