Reproduce “MOMAland: A Set of Benchmarks for Multi-Objective Multi-Agent Reinforcement Learning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Florian Felten, Umut Ucak, Hicham Azmani, Gao Peng, Willem Röpke, Hendrik Baier, Patrick Mannion, Diederik M. Roijers and 5 more
MOMALAND introduces the first publicly available, standardized benchmark suite for Multi-Objective Multi-Agent Reinforcement Learning (MOMARL). This framework provides over 10 diverse environments and standardized APIs, enabling researchers to systematically evaluate and compare MOMARL algorithms, with baselines demonstrating its utility in cooperative and competitive settings.
Key insights
- Standardized benchmarks are crucial for unifying evaluation practices and accelerating algorithmic development in complex, emerging fields like MOMARL.
- Adapting existing MARL algorithms with multi-objective decomposition (e.g., scalarization via weight vectors) can effectively discover diverse Pareto-optimal policies in cooperative MOMARL tasks.
- Applying single-agent MORL algorithms to centralized MOMARL environments provides a viable pathway for initial exploration and development in this domain.
- In strategic multi-agent settings, even a partial shift in agent preferences from pure self-interest to other objectives can guide the system from sub-optimal Nash equilibria towards socially optimal outcomes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.