A Deep Reinforcement Learning Approach for Finding Non-Exploitable Strategies in Two-Player Atari Games

TL;DR

This paper introduces deep reinforcement learning algorithms designed to find Nash equilibrium strategies in two-player Atari games that are resistant to exploitation by adversaries, improving robustness over existing methods.

Contribution

It presents novel algorithms, Nash-DQN and Nash-DQN-Exploiter, that effectively learn non-exploitable strategies in two-player zero-sum Markov games, including Atari games.

Findings

Existing methods are vulnerable to exploitation by adversaries.

Proposed algorithms produce more robust, non-exploitable policies.

Empirical results show superior performance over prior approaches.

Abstract

This paper proposes new, end-to-end deep reinforcement learning algorithms for learning two-player zero-sum Markov games. Different from prior efforts on training agents to beat a fixed set of opponents, our objective is to find the Nash equilibrium policies that are free from exploitation by even the adversarial opponents. We propose (a) Nash-DQN algorithm, which integrates the deep learning techniques from single DQN into the classic Nash Q-learning algorithm for solving tabular Markov games; (b) Nash-DQN-Exploiter algorithm, which additionally adopts an exploiter to guide the exploration of the main agent. We conduct experimental evaluation on tabular examples as well as various two-player Atari games. Our empirical results demonstrate that (i) the policies found by many existing methods including Neural Fictitious Self Play and Policy Space Response Oracle can be prone to exploitation by adversarial opponents; (ii) the output policies of our algorithms are robust to exploitation, and thus outperform existing methods.