Interaction-Aware Game-Theoretic Motion Planning for Automated Vehicles using Bi-level Optimization.

Planning an interactive and cooperative behavior in the vicinity of multiple decision-makers is a challenging task. The game-theoretic perspective provides a suitable framework to describe such interactive scenarios. In this paper, we introduce a motion planning algorithm to generate interactionaware behavior for highly interactive scenarios. Our algorithm is based upon a reformulation of a bi-level optimization problem which frames interactions among two decision makers as a Stackelberg game. In contrast to existing works in this field, we solve the prediction and planning problem simultaneously, which enables the generation of efficient behaviors even in highly interactive situations. The main novelty of our algorithm evolves around its ability to consider general nonlinear constraints. Further, we present mechanisms to introduce courtesy and cooperation into behavior planning which prevents overly aggressive driving, as issue that has been identified in existing interaction-aware planning approaches. Finally, we evaluate our approach in the context of automated driving. Our evaluation first investigates the algorithm's ability to purposefully influence and exploit the response of surrounding vehicles. We then illustrate how the approach can be used for cooperative and courteous planning.