基于驾驶员决策行为表征的宜人性自动驾驶决策规划与鲁棒控制研究

Automatic driving vehicle is a new technology under the background of technological revolution, which represents the future strategic high vehicle technology. Investigating the vehicle path planning and robust control based on the drivers’ decision behavior has important significance that promoting the development of advanced autonomous vehicle to comfort, safety and artificial intelligence. However, the existing autonomous decision making method is lack of driving comfort and path tracking control is lack of robustness. According to the typical scenarios of highway driving, the driver decision behavior characteristics are deeply studied. The coupling mechanism among drivers' decision-making behavior characteristics, working conditions and characteristic parameters are revealed, and the decision-making behaviors of different personality drivers are accurately depicted in this proposal. Then the hybrid dynamic system hierarchical decision making model based on semantic logic and dynamic trajectory optimization is established. The pleasant driving decision and trajectory planning method is proposed that satisfies driver's driving style demand. Finally, the cellular predictive tracking model is established based on the quantization and analysis of perceptual error and decision error. The robust predictive control algorithm with anti-interference ability is proposed to realize the robust tracking control. This proposal provides theoretical and technical support for breaking through and mastering the key technologies of highly autonomous driving.

自动驾驶汽车是新一轮科技革命背景下的新兴技术,代表着未来汽车技术的战略制高点，研究基于驾驶员决策行为特性的宜人性自动驾驶决策规划与鲁棒控制对推动高级自动驾驶汽车向舒适、安全及人工智能方向发展具有重要意义。本项目针对现有自动驾驶决策规划缺乏驾驶舒适性以及运动跟踪控制缺乏鲁棒性等问题，从高速公路典型场景出发，深入研究驾驶员决策行为特性，揭示驾驶员决策行为特性、工况及特征参数之间的耦合机理，准确刻画不同个性驾驶员的决策行为；建立融合语义逻辑和动态轨迹优化的混合逻辑系统分层决策模型，提出满足驾驶员驾驶风格需求的宜人性驾驶决策与轨迹规划方法；以感知误差和决策误差的特征分析及量化为突破口，建立多胞预测控制模型，设计具有抗干扰能力的鲁棒预测控制器，实现对宜人性轨迹规划的稳健、鲁棒跟踪控制。为突破和掌握高度自动驾驶的核心关键技术提供理论和技术支撑。

本项目围绕不同驾驶风格的驾驶员决策行为建模、基于驾驶员决策行为特性的宜人性驾驶决策与轨迹规划方法和基于感知、决策与控制误差的路径跟踪鲁棒控制方法三个关键科学问题，研究了驾驶员决策行为特征参数提取及耦合机理、准确刻画了不同个性驾驶员的决策行为，建立了驾驶员决策行为模型。探索了基于纵横向差异化碰撞风险评估的路径规划方法，建立了融合语义逻辑和动态轨迹优化的混合逻辑系统分层决策模型，提出了满足驾驶员驾驶风格需求的宜人性驾驶决策与轨迹规划方法；量化了感知误差和决策误差，建立了多胞预测控制模型，设计了具有抗干扰能力的鲁棒预测控制器，提出了宜人性轨迹规划的稳健、鲁棒跟踪控制方法，通过驾驶员在环仿真试验系统，验证了驾驶员行为决策及宜人性驾驶规划与控制方法的有效性。项目在基于驾驶语义平面的驾驶风格量化、基于高斯混合模型的驾驶员行为建模、基于KL的驾驶员风格差异性分析方法、融合语义逻辑和动态轨迹优化的混合逻辑决策模型、考虑驾驶员驾驶风格的宜人性驾驶决策与轨迹规划以及宜人性轨迹规划的稳健、鲁棒跟踪控制方法方面取得重要结果，为自动驾驶拟人化决策与控制技术的研发和应用奠定了坚实的理论基础和技术支撑。项目共发表论文22篇，其中SCI、EI检索论文20篇，获权发明专利5项，申请发明专利1项，组建了拟人化自动驾驶研发团队，参加高水平学术会议9人次，培养博士研究生3人，硕士研究生5人。