The paper proposes a solution for two-player general-sum differential games using a Pontryagin-mode neural operator. The authors address the curse of dimensionality and convergence issues through physics-informed neural networks. The key contribution is the introduction of a costate loss, which is computationally cheap and effectively enables the learning of discontinuous values. The close relationship between costates and policies makes the former critical in learning feedback control policies with generalizable safety performance.
Publication date: 5 Jan 2024
Project Page: https://arxiv.org/abs/2401.01502v1
Paper: https://arxiv.org/pdf/2401.01502
