Discovering Diverse Multi-Agent Strategic Behavior via Reward Randomization

Zhenggang Tang · Chao Yu · Boyuan Chen · Huazhe Xu · Xiaolong Wang · Fei Fang · Simon Du · Yu Wang · Yi Wu

Keywords: [ multi-agent reinforcement learning ] [ strategic behavior ] [ reward randomization ] [ diverse strategies ]


We propose a simple, general and effective technique, Reward Randomization for discovering diverse strategic policies in complex multi-agent games. Combining reward randomization and policy gradient, we derive a new algorithm, Reward-Randomized Policy Gradient (RPG). RPG is able to discover a set of multiple distinctive human-interpretable strategies in challenging temporal trust dilemmas, including grid-world games and a real-world game, where multiple equilibria exist but standard multi-agent policy gradient algorithms always converge to a fixed one with a sub-optimal payoff for every player even using state-of-the-art exploration techniques. Furthermore, with the set of diverse strategies from RPG, we can (1) achieve higher payoffs by fine-tuning the best policy from the set; and (2) obtain an adaptive agent by using this set of strategies as its training opponents.

Chat is not available.