Deep reinforcement learning has been applied to the car-following control of autonomous driving in recent years. However, the safety of car-following through reinforcement learning remains a problem. To address this problem, a feedback forecasting based deep reinforcement learning algorithm is proposed. A differential equation model is used to predict the collision of adjacent vehicles. The output action of reinforcement learning is supervised by the feedback single to achieve a safe distance. As a result, the performance of autonomous driving safety is improved while others are slightly affected. The simulation results, based on NGSIM data, show that the predict collision avoidance algorithm is feasible and effective in the car-following control of autonomous driving.
Feedback Forecasting based Deep Deterministic Policy Gradient Algorithm for Car-Following of Autonomous Vehicle
2021-10-15
4803035 byte
Conference paper
Electronic Resource
English
DETC2019-97884 Autonomous Vehicle Driving via Deep Deterministic Policy Gradient
British Library Conference Proceedings | 2019
|Stability Analysis for Autonomous Vehicle Navigation Trained over Deep Deterministic Policy Gradient
BASE | 2023
|