End to End Large Model Predictive Control of Automated Driving Vehicles

In the drive burden alleviation of automated turning, false look ahead of speed and curvature cooperation will cause issues of multiple functional boundary operations. In the author’s previous research from SAEWCX2020, the risk feeling model on multiple lanes was proposed. However, the issues of corner operations were not considered. In real world situations of corner cases with multiple functional boundary operations, it is new and important to utilize more accurate dynamic vehicle model based on end to end large model to deal with issues of wheelbase differential dynamics, slip angle dynamics, and yaw rate dynamics. Human drivers sequentially update their predictive view point along the target lane and predict the intent of other traffic participants, and interact with them to decide the target speed. The above challenges on turning scenarios is tackled with the advance methods of predictive lattice trajectory generation using functional safety modes. In this study, the proposed lattice planner uses trackable path planning methods to reduce the feedback hunting of the downstream motion controller. The hard constraints are tackled with predictive control methods using the look ahead models of wheelbase differential dynamics, slip angle dynamics, and yaw rate dynamics. The behavioral planner is designed to avoid multiple parallel moving intentions. To realize the automated turning features, three sub-features are proposed. The first sub-feature is the lattice planner with short term predictions to realize comfort performance with jerk level. The second sub-feature is the predictive behavioral planner with long term predictions to realize smooth performance with snap level. The third sub-feature is mission planner with global predictions to realize reliable performance with acceleration level. To solve the issues, the predictive lattice control of corner cases is proposed and validated considering multiple boundary corner behavior of soft constraints with ISO26262/ISO21448 functional safety.