NePPO: Near-Potential Policy Optimization for General-Sum Multi-Agent Reinforcement Learning
NePPO is a new Multi-Agent Reinforcement Learning method designed to find approximate Nash equilibria in complex, general-sum games where agents have different goals. It works by learning a player-independent 'potential function' that captures how much an agent's utility changes when they unilaterally change their strategy. By minimizing an objective based on this potential function, the algorithm finds a policy that approximates the desired equilibrium.