Solution Concepts in Hierarchical Games Under Bounded Rationality With Applications to Autonomous Driving

With autonomous vehicles (AV) set to integrate further into regular human traffic, there is an increasing consensus of treating AV motion planning as a multi-agent problem. However, the traditional game theoretic assumption complete rationality too strong for purpose driving, and need understanding driving bounded rational activity through behavioral lens. To that end, we adapt three metamodels behavior; two based on Quantal level-k one Nash equilibria with quantal errors. We formalize different solution concepts can be applied in context hierarchical games, framework used planning, creating models behavior. Furthermore, contributed dataset at busy urban intersection total ~4k agents ~44k decision points, evaluate behavior basis model fit naturalistic data, well their predictive capacity. Our results suggest among evaluated, modeling pure strategy errors level maneuvers bounds sampling actions trajectories provides best behavior, significant impact situational factors performance models.