Fictitious Cross-Play: Learning Global Nash Equilibrium in Mixed Cooperative-Competitive Games

TL;DR

This paper introduces Fictitious Cross-Play (FXP), a novel algorithm that effectively learns global Nash equilibria in mixed cooperative-competitive games by combining self-play and best response strategies, outperforming existing methods.

Contribution

FXP combines self-play and best response training to efficiently converge to global Nash equilibria in complex mixed games, overcoming scalability issues of prior approaches.

Findings

FXP converges to global Nash equilibria in matrix games.

FXP achieves higher Elo ratings and lower exploitability in gridworld domain.

FXP defeats state-of-the-art models in a challenging football game with over 94% win rate.

Abstract

Self-play (SP) is a popular multi-agent reinforcement learning (MARL) framework for solving competitive games, where each agent optimizes policy by treating others as part of the environment. Despite the empirical successes, the theoretical properties of SP-based methods are limited to two-player zero-sum games. However, for mixed cooperative-competitive games where agents on the same team need to cooperate with each other, we can show a simple counter-example where SP-based methods cannot converge to a global Nash equilibrium (NE) with high probability. Alternatively, Policy-Space Response Oracles (PSRO) is an iterative framework for learning NE, where the best responses w.r.t. previous policies are learned in each iteration. PSRO can be directly extended to mixed cooperative-competitive settings by jointly learning team best responses with all convergence properties unchanged. However, PSRO requires repeatedly training joint policies from scratch till convergence, which makes it hard to scale to complex games. In this work, we develop a novel algorithm, Fictitious Cross-Play (FXP), which inherits the benefits from both frameworks. FXP simultaneously trains an SP-based main policy and a counter population of best response policies. The main policy is trained by fictitious self-play and cross-play against the counter population, while the counter policies are trained as the best responses to the main policy's past versions. We validate our method in matrix games and show that FXP converges to global NEs while SP methods fail. We also conduct experiments in a gridworld domain, where FXP achieves higher Elo ratings and lower exploitabilities than baselines, and a more challenging football game, where FXP defeats SOTA models with over 94% win rate.