New Math Makes Robot Teams More Reliable When They Disagree

Game-theoretic models can describe multi-agent interactions in control and robotics, but theoretical and computational complexity makes them intractable in many dynamic, incomplete-information scenarios. Recent work assumes agents share sensor measurements and maintain a common estimate of state uncertainty, but this enforces cooperation in generically noncooperative interactions.

The paper takes an ego agent’s perspective, seeking computational tractability and robustness to unanticipated behavior of other agents due to mismatched state estimates from private rather than shared sensor measurements. It formulates game-theoretic planning in an augmented belief space with perturbed Bayesian dynamics that allows the ego to predict all agents’ potentially divergent state estimates.

A hypothetical adversary, “nature,” chooses perturbations reflecting how non-ego agents’ state estimates—based on measurements the ego agent will not see at runtime—may deviate from the ego’s own belief. The approach is in the spirit of H-infinity robust control extended to noncooperative, stochastic dynamic games. A solver with polynomial-time iterations and two experiments show robust belief-space reasoning leads ego agents to proactively limit future vulnerability to model misspecification of other agents’ beliefs and actions, reducing performance degradation by over 48%.