On the Design of Interaction-Aware SCMPC for Highway Merging Scenarios

This paper addresses interaction-aware decision making and motion planning for highway merging situations using Scenariobased Model Predictive Control (SCMPC). Given tactical decision options for the autonomous vehicle (AV), a traffic prediction algorithm intends to identify the most likely evolutions from the current traffic scene, which are then evaluated by an ensemble of SCMPCs to determine the most efficient decision regarding velocity tracking cost and safety margin satisfaction. This way, we aim to leverage interaction-aware predictions to gain insights about possible target vehicle reactions to the decisions of the AV with the incentive to solve merging situations more efficiently and enhance safety by considering target vehicle intentions. We demonstrate the approach in comparison to a non-interaction-aware baseline method in a multi-vehicle simulation study.