智能汽车人机交互机理与人机共驾技术

Aiming at the two key scientific problems in human-machine intelligence fusion for intelligent vehicles – “Knowledge representation and inference for human-machine intelligence fusion in driving vehicles”, “Mechanisms of adaptation and learning for human-machine Intelligence in complex environments”, this project will investigate the theory of human-machine interaction for intelligent vehicles and the breakthrough on human-machine shared driving techniques will be made. Based on the innovation of theories and key technologies, as well as the cooperation with the automobile enterprise, an experimental system for human-machine intelligence fusion and shared driving for intelligent vehicles will be established for complex city roads. This experimental system will have multiple channels for natural human-machine interaction such as graphics and images, voices, etc. It will also have cooperative driving functions in multiple layers including sensing and prediction, decision and planning, control and execution. Under multiple road conditions, drivers can interact with the autonomous driving system for understanding driving intention and planning strategies via natural ways. The autonomous driving system can not only provide human drivers with vehicle states and road traffic conditions in an obvious manner but also generate personalized and intelligent decision-making and cooperative control outputs. Moreover, the shared driving system has adaptive and learning abilities to deal with complex environments.

围绕智能汽车人机智能融合的两个关键科学问题--“面向汽车驾驶人机智能融合的知识表示与推理”、“人机智能对复杂环境的适应性与学习机理”，深入研究智能汽车人机交互理论，突破人机共驾的核心关键技术。结合理论方法与关键技术创新，与汽车企业合作研制面向城市复杂道路环境的智能汽车人机交互与共驾技术实验演示系统。该系统将实现包括图形图像、语音在内的多种通道自然方式的人机交互界面，具有覆盖感知与预测、决策与规划、控制与执行等多层次的人机协同驾驶功能。在多种典型路况下，驾驶员能以自然方式与自主驾驶系统进行驾驶意图和规划策略的交互；自主驾驶系统能将汽车状态及道路交通态势以直观方式提供给人类驾驶员，并向驾驶员提供个性化、智能化的辅助决策和协同控制支持，具有对复杂环境的适应性和学习能力。

围绕“面向汽车驾驶人机智能融合的知识表示与推理”、“人机智能对复杂环境的适应性与学习机理”两个关键科学问题，开展了智能汽车人机交互理论与人机共驾的核心关键技术研究。在基于模型预测控制的“感知-决策-操控”一体化驾驶员行为模型与驾驶意图预测方法、基于动态贝叶斯网络和Grid-LSTM的车辆与行人轨迹预测、基于拓扑空间的一致性决策与规划算法、基于事件触发的智能车辆人机控制权切换与共享控制方法、智能车辆人机协同控制的强化学习方法与智能发育机制等方面取得了重要研究进展，在多种仿真与实车实验场景下验证了项目提出的驾驶员行为分析、轨迹预测、驾驶意图和规划策略交互机制、应急避障下的人机协同控制理论与方法的有效性。发表学术论文43篇，其中SCI收录论文26篇，IEEE Transactions汇刊长文10篇，申请/授权专利13项。.同时结合理论方法与关键技术创新，采用红旗EV电动汽车平台，与第一汽车集团合作研制了面向城市复杂道路环境的智能汽车人机交互与共驾技术实验演示系统，实现了包括图形图像、语音、脑—机、手机在内的多种通道自然方式的人机交互界面，具有覆盖感知与预测、决策与规划、控制与执行等多层次的人机协同驾驶功能。在典型场景下针对脑—机协同决策、人机智能切换以及人机协同共驾等功能进行了试验验证，实现了人机智能协同交互和智能辅助决策与控制，能够在人类驾驶员出现操控失误条件下实现实时控制权切换或者人机共享辅助控制，为新一代智能汽车主动安全产品奠定了技术基础。在此基础上构建了基于脑-机接口和智能驾驶系统的人机融合智能决策和人机共驾方法，建立了人脑决策输出信号与智能驾驶系统之间的物理连接，实现了人类上层决策与车辆智能驾驶控制的协同驾驶，并在实车试验场进行了验证，为基于脑-机接口的新型智能汽车人机交互和人机共驾机制奠定了基础。