Multi-agent reinforcement learning robust to model uncertainty scales better
Interactive Distributionally Robust Multi-Agent Learning with General Function Approximation
Machine LearningArtificial Intelligence
Summary
Multi-agent reinforcement learning often struggles when the rules of the environment are uncertain or changing. The authors study a way to make learning more reliable, even when agents face such uncertainties and interact strategically. They propose a new framework called RoMEX-φ that estimates worst-case outcomes efficiently using data, without needing exact models. This method works well even in large or complex settings and is more resilient to unexpected changes than some existing approaches.
What this means in practice
- •For multi-agent system developers: Build multi-agent systems that remain reliable under uncertain environment changes by using robust exploration and learning methods.
- •For game ai programmers: Create AI that adapts to shifting game dynamics and opponent behaviors with improved robustness to model errors.
Authors
Debamita Ghosh, George K. Atia, Yue Wang
Abstract
Model misspecification poses a fundamental challenge in multi-agent reinforcement learning, where transition uncertainty can be amplified by strategic interactions among agents. Distributionally robust Markov games (DRMGs) provide a principled framework for addressing such uncertainty, yet existing methods often rely on restrictive assumptions or scale poorly to large state and joint action spaces. We study online learning in general-sum DRMGs with general function approximation and $φ$-divergence uncertainty sets. We propose RoMEX-$φ$, a model-free framework that integrates equilibrium-based exploration with dual fitted learning. Through a functional dual representation of the robust multi-agent Bellman operator, RoMEX-$φ$ enables tractable worst-case value estimation from nominal interaction data using a centered empirical robust discrepancy. We introduce the robust Multi-Agent Decoupling Coefficient (robust MADC) to characterize the intrinsic exploration complexity arising from strategic interactions and adversarial transition uncertainty. We establish sublinear robust regret guarantees governed by the robust MADC rather than explicitly by the state and joint action space sizes, replacing tabular dependence with intrinsic function-class complexity. Numerical experiments on a scalable general-sum DRMG under total variation uncertainty show that RoMEX-$φ$ is substantially more resilient to transition shifts than its non-robust counterpart while remaining competitive with an exact tabular robust baseline. Our results provide a scalable framework for distributionally robust multi-agent reinforcement learning with general function approximation.