首页|基于博弈论的右转无信号交叉口行人行为模拟

基于博弈论的右转无信号交叉口行人行为模拟

扫码查看
原文链接
  • NETL
  • NSTL
  • 万方数据
为逼真模拟行人-车辆交互的真实交通场景,融合博弈理论和数据驱动的思想,在逆强化学习和博弈论的基础上提出博弈——深度最大熵逆强化学习算法(Game-deep max entropy inverse reinforcement learning,G-DMIRL).将行人建模为智能体,通过真实的行人-车辆交互轨迹获取不同博弈决策下的行人的奖励函数,并推断行人-车辆交互的博弈机制,利用获取的奖励函数和动作策略开发出行人行为模拟模型.仿真结果表明,开发的模型在有限状态下能够准确地模拟出不同决策下行人的行为动作,建立的行人-车辆交通场景能够为自动驾驶汽车的识别、预测与路径规划算法的开发验证提供支撑.
Game Theory-based Simulation of Pedestrian Behavior at Right-turn Unsignalized Intersections
In order to realistically simulate the real traffic scene of pedestrian-vehicle interaction,this study integrates game theory and data-driven ideas,and proposes a game-deep maximum entropy inverse reinforcement learning algorithm(G-DMIRL),modeling pedestrians as intelligent bodies,obtaining the reward functions of pedestrians under different game decisions through real pedestrian-vehicle interaction trajectories,and inferring the game mechanism of pedestrian-vehicle interaction,and developing a simulation model of pedestrian behavior by using the obtained reward functions and action strategies.The simulation results show that the developed model can accurately simulate the behavioral actions of pedestrians under different decisions in a finite state,and the established pedestrian-vehicle traffic scenario can provide support for the development and validation of recognition,prediction and path planning algorithms for self-driving cars.

inverse reinforcement learningreinforcement learninghuman-vehicle interactiongame theoryright-turn intersection

李文礼、张祎楠、石晓辉、王梦昕

展开 >

重庆理工大学汽车零部件先进制造技术教育部重点实验室 重庆 400054

重庆长安汽车股份有限公司 重庆 400020

逆强化学习 强化学习 行人-车辆交互 博弈论 右转交叉口

重庆市自然科学基金重庆市留学人员回国创业创新支持计划重庆市教委科学技术研究重庆市技术创新与应用发展专项重大资助项目

cstc2021jcyjmsxmX0183CX2021070KJQN202201170CSTB2022TIAD-STX0003

2024

10.3901/JME.2024.10.086

机械工程学报
中国机械工程学会

机械工程学报

CSTPCD北大核心
影响因子：1.362
ISSN：0577-6686
年,卷(期)：2024.60(10)