Driver-automation mutual adaptation: modeling, design, and evaluation of haptic interface for cooperative driving tasks

在2022财年，我们的研究成果总结如下：1.利用深度学习网络建立了驾驶员模型，并利用9人参与的驾驶模拟器实验数据进行了训练，结果表明该模型能够有效地预测个体驾驶员的目标轨迹。将驱动程序模型应用于共享控制的进一步设计。提出了一种间接共享控制框架下的车道保持自适应控制分配算法。该算法采用基于信任的数据驱动共享控制策略，融合了驾驶员的控制输入和自动化。自动控制代理采用库普曼模型预测控制进行计算。基于人机交互和驾驶员输入意图的信任机制，采用混合式人机信任模型实现自适应控制权限分配。在一个五人参与的交互式仿真环境中，该算法被证明是有效和有益的。开发了一种采用共享控制策略的转向辅助系统，用于自动驾驶车辆的驾驶员超驰。当车辆启动故障安全操作时，系统会考虑驾驶员对超驰的潜在需求。在车辆脱离危险时，采用基于驾驶员可控性的共享控制策略，实现驾驶权限的平稳移交。在多车道高速公路场景中验证了所提出的系统的有效性。

Our research focuses on driver-automation mutual adaptation for haptic shared control.In FY 2022, our research achievements are summarized as follows:1. A driver model was developed using a deep learning network and was trained on data collected from a driving simulator experiment with nine participants, and is demonstrated to be effective in predicting individual driver's target trajectory. The driver model is applied for further design of shared control.2. An adaptive control allocation algorithm was developed for lane keeping under an indirect shared control framework. The algorithm uses a trust-based data-driven shared control strategy and blends control inputs of drivers and automation. The automated control agent is computed using Koopman model predictive control. Adaptive control authority allocation is achieved using a hybrid human-to-machine trust model based on a trust mechanism inferred from human-automation interaction and driver input intention. The proposed algorithm is demonstrated to be effective and beneficial in an interactive simulation environment with five participants.3. A steering assistance system involving a shared control strategy was developed for driver override in automated vehicles. The system considers the potential driver demand for override when the vehicle initiates a fail-safe maneuver. A shared control strategy based on driver controllability is adopted to smoothly transfer driving authority when the vehicle is out of danger. The effectiveness of the proposed system is demonstrated in multi-lane highway scenarios.