Multiagent trajectory models via game theory and implicit layer-based learning

For prediction of interacting agents' trajectories, we propose an end-to-end trainable architecture that hybridizes neural nets with game-theoretic principles, has interpretable intermediate representations, and transfers to robust downstream decisions. It combines a differentiable implicit layer, that maps preferences to local Nash equilibria, with a learned equilibrium refinement concept and preference revelation, upon initial trajectories as input. This is accompanied by a new class of continuous potential games, theoretical results for explicit gradients and soundness, and several measures to ensure tractability. In experiments, we evaluate our approach on two real-world data sets, where we predict highway driver merging trajectories, and a simple decision-making transfer task.