一种基于深度强化学习的自适应巡航控制算法

An Adaptive Cruise Control Algorithm Based on Deep Reinforcement Learning

自适应巡航控制是智能驾驶领域的核心技术,可通过分层控制或参数可变控制算法实现,但这些算法无法有效应对突发的跟车路况。为此,将深度强化学习与自适应巡航控制相结合,提出基于确定性策略梯度算法的自适应巡航控制算法,使智能车辆可以在自学习过程中完成自适应巡航并不断改进。在开源平台上的测试结果表明,该算法可以使智能驾驶车辆在跟车时加速度保持在1.8 m/s^2以内的比例超过90%,达到人类驾驶员的巡航跟车水平。

Adaptive Cruise Control（ ACC） is one of the most core technologies in the field of smart driving.Researchers mostly use traditional hierarchical control methods or variable control algorithms to implement this technology. These algorithms can not respond effectively to unexpected follow-up road conditions. For this reason,this paper combines deep reinforcement learning with ACC, and proposes an ACC algorithm based on deterministic strategy gradient algorithm, so that the intelligent vehicle can complete adaptive cruise and continue to improve in the continuous self-learning process. The test results under the open source platform show that this algorithm can make the ratio of the acceleration of the smart driving vehicle within 1. 8 m/s^2 within 90% of the follow-up acceleration,which can reach the level of the cruise control of the human pilot.