Replicable-MARL/MARLlib

Health v2 · maintenance only

Adoption not part of the score

1345 stars · 198 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

MARLlib is a unified multi-agent reinforcement learning library built on Ray and RLlib with PyTorch, providing implementations of many MARL algorithms and integration with numerous multi-agent environments. It offers a unified API for training, tuning, and benchmarking MARL agents across cooperative, competitive, and mixed settings.

Use cases

• train multi-agent reinforcement learning agents
• benchmark MARL algorithms across environments
• run cooperative and competitive multi-agent RL experiments
• use PettingZoo environments with RLlib
• reproduce multi-agent RL research results
• compare centralized vs decentralized training approaches

When to choose

• you need many MARL algorithms under one unified API
• you want Ray/RLlib scalability for multi-agent training
• you want built-in support for popular multi-agent environments like MPE, SMAC, and PettingZoo

When to avoid

• you need single-agent RL only
• you require the latest RLlib versions or active feature development
• you need lightweight training without Ray dependencies

Facets

library · maturity maintenance

reinforcement-learning machine-learning sdk reinforcement-learning machine-learning artificial-intelligence python multi-agent-reinforcement-learning rllib ray pytorch marl-algorithms pettingzoo research-library research gpu linux macos

Member repositories

For agents

markdown · JSON · MCP: product_card(name="Replicable-MARL/MARLlib")