The Interplay of Differential Games, Optimal Control, and Energy-based Models in Multi-Agent Interactions

[Submitted on 31 Aug 2023]

Click here to download a PDF of the paper titled “On a Connection between Differential Games, Optimal Control, and Energy-based Models for Multi-Agent Interactions” by Christopher Diehl, Tobias Klosek, Martin Krüger, Nils Murzyn, and Torsten Bertram.

Download PDF

Abstract: Game theory provides a mathematical framework for modeling multi-agent interactions. However, its application in real-world robotics is hindered by challenges such as unknown agent preferences and goals. In this paper, we address these challenges by establishing a connection between differential games, optimal control, and energy-based models. We propose an Energy-based Potential Game formulation that unifies existing approaches. Moreover, we introduce a new end-to-end learning application that combines neural networks for game-parameter inference with a differentiable game-theoretic optimization layer. Our experiments, involving simulated mobile robot pedestrian interactions and real-world automated driving data, demonstrate that this game-theoretic layer improves the predictive performance of various neural network backbones.