Back to Ecosystem Pulse
EvaluationProduction Ready
on-policy
by marlbenchmark
Official MAPPO implementation for benchmarking cooperative multi-agent policies
Python
Updated Jul 18, 2024
Share:
Summary
Implements Multi-Agent PPO (MAPPO) for training and benchmarking cooperative multi-agent policies. Provides the official algorithm implementation with training loops, environment wrappers, and evaluation scripts for common MARL testbeds like SMAC and Hanabi. Includes reproducible configs and checkpoints to compare MAPPO performance across environments and research baselines. Open Agent Specification (Agent Spec)
The Value Proposition
As multi-agent systems proliferate, consistent evaluation is essential to judge coordination, robustness, and failure modes. MAPPO offers a standardized policy-gradient baseline for comparing cooperative behaviors and emergent failures across environments. For agent-to-agent evaluation and agent track record building, reliable MARL benchmarks like this let teams quantify how policy changes affect interaction quality and reliability. Model Context Protocol (MCP) and Agent Protocol provide shared frameworks to reason about interactions and coordination.
Ideal For
Researchers and engineers benchmarking cooperative multi-agent algorithms or validating agent policies on SMAC, Hanabi, and StarCraft II scenarios. Agent-to-Agent Protocol (A2A)
Use Cases
- Benchmark MAPPO on SMAC or Hanabi to compare cooperative policies
- Validate multi-agent coordination and failure modes before deployment
- Generate reproducible training runs and checkpoints for agent-to-agent evaluation pipelines
Works With
smachanabistarcraftiipytorchpython
Topics
algorithmshanabimappompesmulti-agentpposmacstarcraftii
Similar Tools
pymarlsmac
Keywords
multi-agentmappomulti-agent evaluationmarL