Game-theoretic Occlusion-Aware Motion Planning: an Efficient Hybrid-Information Approach

We present a novel algorithm for motion planning in complex, multi-agent scenarios in which occlusions prevent all agents from seeing one another. In this setting, the fundamental information that each agent has, i.e., the information structure of the interaction, is determined by the precise configurations in which agents come into view of one another. Occlusions prevent the use of existing pure feedback solutions, which assume availability of the state information of all agents at every time step. On the other hand, existing open-loop solutions only assume availability of the initial agent states. Thus, they do not fully utilize the information available to agents during periods of unhampered visibility. Here, we first introduce an algorithm for solving an occluded, linear-quadratic (LQ) dynamic game, which computes Nash equilibrium by using hybrid information and switching between feedback and open-loop information structures. We then design an efficient iterative algorithm for decision-making which exploits this hybrid information structure. Our method is demonstrated in overtaking and intersection traffic scenarios. Results confirm that our method outputs trajectories with favorable running times, converging much faster than recent methods employing reachability analysis.