基于类人行为表征的场景可迁移决策控制方法

Scenario Transferable Decision-Making and Control Based on Human-Like Behavior Representation

为提升智能车对不同驾驶场景的适应能力和在复杂场景下的决策控制性能,提出了一种基于类人行为表征的智能车场景可迁移决策控制方法.该方法在人类驾驶数据采集的基础上进行类人行为表征与决策基元提取,采用强化学习方法构建决策控制模型,完成在复杂驾驶场景下的决策基元选取与场景通行.进一步从决策基元迁移和决策基元组合优化迁移两个维度构建决策控制迁移模型,并在仿真环境下对算法和模型进行了试验验证.结果表明,所提出的智能车场景可迁移决策控制方法能够实现在同类场景下的通行效率提升,提升百分比达到21.9%;在异类场景之间迁移的任务完成率达到97.5%.

In order to improve the adaptability of intelligent vehicles to different driving scenarios and their de-cision-making and control performance in complex scenarios,a scenario transferable decision-making and con-trol method was proposed based on human-like behavior representation.Collecting human driving data,a de-cision-making and control model was constructed with the reinforcement learning method to carry out human-like behavior representation and decision primitive extraction,making the decision primitives chosen and scen-ario transfer functions suitable for complex scenarios.Furthermore,the transferable decision-making and control model was constructed with two dimensions,decision primitive transfer and decision primitive combination strategy transfer,and was verified in simulation.The simulation results show that the proposed scenario transfer-able decision-making and control method for intelligent vehicles can improve traffic efficiency up to 21.9%in similar scenarios,and the driving task completion rate can reach up to 97.5%in transferring between heterogen-eous scenarios.