基于博弈论的多车智能驾驶交互决策综述

智能驾驶是交通和汽车领域未来发展的重要方向,决策规划作为智能驾驶系统中的关键模块,一直是其重点研究领域之一.当前的研究热点正在从单车智能驾驶决策向混行交通场景下的多车智能驾驶决策进行拓展,因此,需要在复杂动态场景和多并行任务下生成符合车辆动力学且不与道路边界和其他交通参与者发生碰撞的高质量轨迹.多车混行驾驶是对道路时空资源的竞争性使用,博弈论可为多车交互决策提供重要的理论与技术手段.对此,应用博弈论方法进行智能驾驶决策研究的综述,基于滚动时域、微分博弈和马尔科夫博弈这3类常用的博弈模型,对现有相关研究进行归类总结和分析.首先简要介绍博弈论基础知识;其次,总结常见的智能驾驶场景并分析各场景下交互决策的核心问题;然后,通过3种不同的博弈模型对多车交互决策进行建模,分别介绍它们的求解算法和思路及相关的研究工作;最后,介绍相关的仿真实验和测试方法,同时也对未来的技术发展和挑战给出见解.

A review on interactive decision-making of multi-vehicle autonomous driving with a game theoretical perspective

Autonomous driving is an important direction for the future development of the transportation and the automotive field. Planning and decision-making as a key module has always been an important area of autonomous driving research. The research interest of the community is expanding from single-vehicle driving to multi-vehicle driving in hybrid traffic scenarios. The main forthcoming challenge is to generate high-quality trajectories that conform to vehicle dynamics and do not collide with road boundaries and other traffic participants in complicated and dynamical scenarios with multiple tasks. Multi-vehicle driving can be treated as competitive utilization of the spatial-temporal resources of the road, hence game theory can provide an important theoretical and technical tool for multi-vehicle interactive decision-making. This paper provides a review on applying game theory to the decision-making of intelligent driving. We summary and classify the literatures on the interactive decision-making of intelligent driving based on game theory into three commonly used game methods, including the receding horizon games, the differential games, and the Markovian games. At first, the basic knowledge of game theory is introduced, and then the common intelligent driving scenarios are introduced and the key problems of interactive decision-making are summarized. Then, the above three game methods are used to model the multi-vehicle interactive decision-making, and the solution algorithms in literatures are introduced respectively. Finally, the relevant simulation experiments and test methods are summarized. The remaining challenges for the future development of areas are discussed with a personal perspective.