Deep Reinforcement Learning for Autonomous Driving: A Survey
作者：互联网
郑重声明：原文参见标题，如有侵权，请联系作者，将会撤销发布！
IEEE Trans. Intell. Transp. Syst. 23(6): 4909-4926 (2022)
Abstract
随着深度表示学习的发展，强化学习(RL)领域已经成为一个强大的学习框架，现在能够在高维环境中学习复杂的策略。本综述总结了深度强化学习(DRL)算法，并提供了已采用(D)RL方法的自动驾驶任务的分类，同时解决了自动驾驶智能体在现实世界部署中的关键计算挑战。它还描绘了相关但不是经典RL算法的相邻领域，例如行为克隆、模仿学习、逆强化学习。讨论了模拟器在训练智能体中的作用，以及验证、测试和鲁棒化RL中现有解决方案的方法。
Index Terms——Deep reinforcement learning, Autonomous driving, Imitation learning, Inverse reinforcement learning, Controller learning, Trajectory optimisation, Motion planning, Safe reinforcement learning.
I. INTRODUCTION
II. COMPONENTS OF AD SYSTEM
A. Scene Understanding
B. Localization and Mapping
C. Planning and Driving policy
D. Control
III. REINFORCEMENT LEARNING
A. Value-based methods
B. Policy-based methods
C. Actor-critic methods
D. Model-based (vs. Model-free) & On/Off Policy methods
E. Deep reinforcement learning (DRL)
IV. EXTENSIONS TO REINFORCEMENT LEARNING
A. Reward shaping
B. Multi-agent reinforcement learning (MARL)
C. Multi-objective reinforcement learning
V. REINFORCEMENT LEARNING FOR AUTONOMOUS DRIVING TASKS
A. State Spaces, Action Spaces and Rewards
B. Motion Planning & Trajectory optimization
C. Simulator & Scenario generation tools
D. LfD and IRL for AD applications
VI. REAL WORLD CHALLENGES AND FUTURE PERSPECTIVES
A. Validating RL systems
B. Bridging the simulation-reality gap
C. Sample efficiency
D. Exploration issues with Imitation
E. Intrinsic Reward functions
F. Incorporating safety in DRL
G. Multi-agent reinforcement learning
VII. CONCLUSION
标签：methods,Driving,reinforcement,学习,Survey,Reinforcement,learning,RL,REINFORCEMENT 来源： https://www.cnblogs.com/lucifer1997/p/16484135.html