Safe Occlusion-Aware Autonomous Driving via Game-Theoretic Active Perception

• DOI: 10.15607/rss.2021.xvii.066
• Publication Date: 2021-06-27T13:30:56Z
• Authors: Zixu Zhang; Jaime Fisac
• Keywords: Pursuer; Leverage (statistics)

Autonomous vehicles interacting with other traffic participants heavily rely on the perception and prediction of agents' behaviors to plan safe trajectories. However, as occlusions limit vehicle's ability, reasoning about potential hazards beyond field view is one most challenging issues in developing autonomous driving systems. This paper introduces a novel analytical approach that poses trajectory planning under hybrid zero-sum dynamic game between vehicle (evader) an initially hidden participant (pursuer). Due occlusions, pursuer's state unknown evader may later be discovered by sensors. The analysis yields optimal strategies for both players well set initial conditions from which guaranteed avoid collisions. We leverage this theoretical result develop framework provides worst-case safety guarantees while minimizing conservativeness accounting ability actively road users soon they are detected future observations. Our agnostic environment suitable various motion planners. demonstrate our algorithm urban highway scenarios using open-source CARLA simulator.