基于深度强化学习的汽车自动紧急制动策略  

Vehicle Automatic Emergency Braking Strategy Based on Deep Reinforcement Learning

作　　者：黄舒伟 何少炜 金智林[1] Huang Shuwei;He Shaowei;Jin Zhilin(State Key Laboratory of Mechanics and Control of Mechanical Structures,Nanjing University of Aeronautics and Astronautics,Nanjing 210016)

机构地区：[1]南京航空航天大学,机械结构力学及控制国家重点实验室,南京210016

出　　处：《汽车技术》2021年第5期9-15,共7页Automobile Technology

基　　金：国家自然科学基金项目(51775269)。

摘　　要：针对传统自动紧急制动策略制动减速度波动大、制动过程乘坐舒适性及弯道制动安全难以保障的问题,提出一种基于深度强化学习的汽车自动紧急制动策略。建立了包括纵向、横向及横摆运动的3自由度车辆模型,根据碰撞预警时间设计奖励函数,应用深度确定性策略梯度算法设计了基于深度强化学习的自动紧急制动策略,开展了直道行驶工况与弯道行驶工况仿真测试。结果表明,所提出的策略具有很好的收敛性,在满足中国新车评价规程(C-NCAP)的直道行驶安全性要求的同时,提高了紧急制动时的乘坐舒适性,且实现了汽车弯道行驶的自动紧急制动,提高了弯道行驶安全性。To deal with the large fluctuation of the deceleration speed of the traditional automatic emergency braking strategy,the difficulty to ensure ride comfort during the braking process and the braking safety at curves,this paper proposes an automatic emergency braking strategy with deep reinforcement learning.A 3-degree-of-freedom vehicle model including longitudinal,transverse and yaw motion is established,a reward function is designed according to the collision warning time,and an automatic emergency braking strategy based on deep reinforcement learning is designed by using the deep deterministic policy gradient algorithm.The driving condition on straight road and the curve is simulated.The results indicate that the proposed strategy has good convergence,meets the straight road safety requirements of the C-NCAP test regulations,and improves the ride comfort of the vehicle during the emergency braking on the straight road.Moreover,the automatic emergency brake is implemented on curves to improve driving safety on the curve.

关 键 词：高级驾驶辅助系统 自动紧急制动 深度强化学习 制动安全性 乘坐舒适性 

分 类 号：U461.1[机械工程—车辆工程]