Self Play Reinforcement Learning

Self-play reinforcement learning (RL) trains agents by having them repeatedly play against themselves, leading to improved performance without needing human-labeled data. Current research focuses on enhancing these methods by incorporating techniques like human-regularized policies to improve real-world applicability (e.g., autonomous driving), employing graph neural networks to better capture complex game structures, and using population-based training to create more robust agents less susceptible to adversarial strategies. These advancements are pushing the boundaries of AI capabilities in complex domains, from strategic board games to autonomous systems, and are contributing to a deeper understanding of RL algorithms and their limitations.

Papers

March 28, 2024

Human-compatible driving partners through data-regularized self-play reinforcement learning
Daphne Cornelisse, Eugene Vinitsky
Autonomous Driving Autonomous Vehicle Multi Agent Channel Partner Level Self Play Reinforcement Learning

November 22, 2023

From Images to Connections: Can DQN with GNNs learn the Strategic Game of Hex?
Yannik Keller, Jannis Blüml, Gopika Sudhakaran, Kristian Kersting
Gene Level GNN New Connection Strategy Game Relational Structure Self Play Reinforcement Learning Game Structure Game Diplomacy

March 6, 2023

Reinforcement Learning Based Self-play and State Stacking Techniques for Noisy Air Combat Environment
Ahmet Semih Tasbas, Safa Onur Sahin, Nazim Kemal Ure
Reinforcement Learning Robust Reinforcement Learning State Abstraction Air Combat Unmanned Combat Self Play Reinforcement Learning

February 23, 2023

Targeted Search Control in AlphaZero for Effective Policy Improvement
Alexandre Trudeau, Michael Bowling
Policy Iteration Self Play Policy Improvement AlphaZero Based Artificial Intelligence Self Play Reinforcement Learning

October 11, 2022

Mastering the Game of No-Press Diplomacy via Human-Regularized Reinforcement Learning and Planning
Anton Bakhtin, David J Wu, Adam Lerer, Jonathan Gray, Athul Paul Jacob, Gabriele Farina, Alexander H Miller, Noam Brown
Multi Agent Task Planning Game Content Regularized Reinforcement Learning Imitation Learned Policy Self Play Reinforcement Learning Reward Maximizing Policy

August 10, 2022

Reducing Exploitability with Population Based Training
Pavel Czempin, Adam Gleave
Adversarial Training Zero Sum Game Adversarial Policy Monotone Adversary Population Based Training Self Play Reinforcement Learning

June 12, 2022

A Unified Approach to Reinforcement Learning, Quantal Response Equilibria, and Two-Player Zero-Sum Games
Samuel Sokota, Ryan D'Orazio, J. Zico Kolter, Nicolas Loizou, Marc Lanctot, Ioannis Mitliagkas, Noam Brown, Christian Kroer
Reinforcement Learning Unified Framework Video Game Mirror Descent Two Player Self Play Reinforcement Learning

May 25, 2022

Impartial Games: A Challenge for Reinforcement Learning
Bei Zhou, Søren Riis
Reinforcement Learning Adversarial Attack Challenge Task Self Play Reinforcement Learning AlphaZero Algorithm