驾驶员与自动驾驶系统同时分享控制权的人机共驾技术为降低接管风险、缓解驾驶员工作负担提供了全新思路。然而共驾系统为双智能体冗余系统,人机同时分享控制权所带来的人机冲突问题不仅影响驾驶体验,更威胁车辆主动安全。因此本文针对共驾系统中的人机决策分歧与转向控制冲突问题,围绕人机决策分歧建模、人机决策融合及拟人化转向共享控制三大内容展开研究,建立了一套自恰的驾驶员意图预测-人机协同决策-转向共享控制理论以解决共驾系统中的人机协调性问题,并设计了高主动安全、低人机冲突的眼动伺服型人机混合智能共驾系统的工程方案。首先针对控制层的人机预瞄行为差异问题,提出了模仿驾驶员时变预瞄行为的拟人化轨迹跟踪控制模型,并在模型中进一步引入控制权的动态调度策略避免人机转向控制冲突。运用带根轨迹约束的LPV-H∞方法对该时变模型进行综合,所推导的控制策略在确保人机共享控制系统的渐进稳定性、轨迹跟踪精度的同时,降低了人机协同式车道保持工况下的人机转向力矩冲突。其次针对人机决策建模问题,提出了基于转向力矩冲突的人机决策分歧估计方法。运用非合作型博弈理论建立了人机决策分歧与转向力矩冲突之间的映射关系,提出了基于纳什、斯塔克博格均衡策略的人机决策分歧建模方法,证明了人机转向冲突在博弈均衡条件下人机双闭环系统的李雅普诺夫稳定性,解决了人机转向力矩冲突的量化描述问题。基于人机决策分歧模型,设计了基于纳什均衡策略的驾驶员目标轨迹的后向估计算法,克服了双智能体系统人机决策混淆问题。 第三,针对人机决策融合问题,提出了基于眼动行为的驾驶员目标轨迹前向估计算法,该算法基于卷积神经网络和编码-解码网络的分层结构解决驾驶员意图的早期、长时域预测问题。随后设计了同时包含驾驶员目标轨迹前向、后向估计结果的时变评估函数,提出了“前向预测-后向矫正”的人机协同式路径规划算法,降低了人机决策分歧,增加了驾驶员的决策自由度。最后,构建了眼动伺服型人机共驾试验平台,对人机决策融合算法、转向共享控制策略进行了试验验证。主客观评价结果表明,所提出的人机协同决策-控制方案能显著提升共驾系统人机协调性及车辆主动安全性。
By simutaneously sharing the control authority between the dirver and the autonomous driving system, the human-machine co-piloting scheme provides a new thinking for reducing the driver’s takeover risk and lightening his/her workload. However, the co-piloting system is a double-agent redundant system. Thus The human-machine decision-making and control conflicts deteriorate the driving experience of the human driver and threaten the vehicle active safety. Focusing on three research contents of driver-machine decision divergence modeling, driver machine decision fusion, and human-like steering control strategy designing, this paper proposes a systematical driver intention prediction, driver-machine cooperative decision, and driver-machine shared control theories to handle the coordination issues in co-piloting system. Furthermore, an engineering scheme of eye-gazing-servo co-piloting system is designed for experiment validation. Firstly, to deal with the driver-machine behavior difference in steering control, a path-tracking model that mimics the driver’s preview behavior is established. and the steering torque conflicts are further mitigated by dynamically regulating the driver-machine control authorities. Then the control system is synthesized by a novel linear parameter varying H-infinity control strategy with root locus constraint, which not only ensures the robustness and asymptotical stability of the track tracking system, but achieves the expected path tracking control performance. Therefore, the steering conflicts in driver-machine cooperative lane keeping is reduced.Secondly, concentrating on the driver-machine decision modeling issue, we porpose a novel method for estimating the driver-machine decision divergence. The mathematical relationship between human-machine decision-making divergence and control conflicts is established by non-cooperative game theory. The Nash and Stackelberg equilibrium-based strategies are derived to model the driver-machine decision divergence, and the Lyapunov stability of the co-piloting system under the Nash equilibrium is proved. Based on the non-cooperative game theory, the reasonable quantification of driver-machine conflicts is exhibited. Based on game theoretical decision divergence model, the backward estimation algorithm of the driver’s target trajectory is designed, which overcomes the driver-machine decision confusion problem in the double-agent system.Thirdly, for handling the driver-machine decision fusion issue, the forward estimation algorithm of the driver’s target trajectory is given based on the driver’s eye-gazing information. Based on the hierarchical structure of the convolutional neural network and the encoder-decoder network, the forward eastimation algorithm realizes the early and long-term prediction of the driver's lane change intention. Then the forward and backward estimation results of the driver’s target trajectory are simultaneously merged into a time-varying path-planning cost function, thus creating a “forward prediction”-“backward correction”-based driver-machine cooperative path planning algorithm, which avoids the decision divergence and increase the decision freedom for the human driver.Finally, the driver-machine co-piloting simulation platform is established to verify the proposed driver-machine decision fusion algorithm and shared steering control strategy. The subjective and objective evaluation results imply that the proposed driver-machine cooperative decision-control scheme can significantly improve the driver-machine cooperative performance and the vehicle active safety.