Fail-Safe Behavior and Motion Planning Incorporating Shared Control for Potential Driver Intervention
Wei Xue, Zheng Wang, Rencheng Zheng, Xutao Mei, Bo Yang, Kimihiko Nakano
Abstract
Automated vehicles require an efficient fail-safe system for unexpected failures in normal operation. Although existing methods aiming at L4 driving automation leverage conservative motion planning to independently guide fail-safe behaviors when a normal planning function fails, they usually ignore human intervention demands. In fact, manual takeover is still a reliable choice of fail-safe maneuver today. Moreover, human drivers tend to handle critical situations, rather than trust the automated system with the vehicle control. Here, we propose a fail-safe architecture for vehicle behavior and motion planning module incorporating shared control in response to the driver intervention. The proposed method not only estimates driver intention to cooperate with the driver on maneuver selection, but also distributes the driving authority in accordance with the types of achievable minimal risk condition. By using the receding horizon planner with composition planning horizons, the driver is able to obtain an appropriate takeover condition on the basis of safety guarantee. Test in the library of critical scenarios shows that the proposed method is effective in terms of safety and assistance to the driver.