# marlbenchmark/on-policy

This is the official implementation of Multi-Agent PPO (MAPPO).

Repository: https://github.com/marlbenchmark/on-policy
Canonical: https://ross.abutalabs.com/products/on-policy
Homepage: https://sites.google.com/view/mappo
Language: Python
License: MIT
License Family: permissive
Topics: hanabi, mappo, smac, mpes, starcraftii, ppo, multi-agent, algorithms
Last push: 2024-07-18T10:00:36+00:00

## Health v2 (maintenance only)
Score: 32/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 0, release rhythm 35, longevity 100
- inputs: {"age_days": 2017, "days_push": 776, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2085, forks 385 (observed 2026-08-28T04:06:11.871456+00:00)

## What it is
The official PyTorch implementation of Multi-Agent PPO (MAPPO), an on-policy reinforcement learning algorithm for cooperative multi-agent settings. It includes environment wrappers, training runners, and tuned hyperparameter scripts for benchmarks such as SMAC, SMACv2, Hanabi, MPEs, and Google Research Football.

## Use cases
- train multi-agent reinforcement learning policies with PPO
- reproduce MAPPO benchmark results on SMAC or Hanabi
- compare on-policy vs off-policy algorithms in cooperative multi-agent games
- get a strong baseline for multi-agent RL research
- run RL experiments on StarCraftII SMAC v2 or Google Research Football
- study ablation factors that affect PPO performance in multi-agent settings

## When to choose
- you need a well-tuned, paper-backed MAPPO baseline for cooperative multi-agent benchmarks
- you want reproducible training scripts with published hyperparameters and curves
- your research targets SMAC, Hanabi, MPEs, or Google Research Football environments

## When to avoid
- you need off-policy multi-agent algorithms like QMIX or MADDPG
- you want a general-purpose RL library with many algorithms rather than a focused MAPPO implementation
- you lack a GPU or the specific environment dependencies (StarCraftII, Hanabi, etc.)

## Facets
- artifact type: library
- maturity: active
- function: reinforcement-learning, machine-learning, benchmarking
- domain: reinforcement-learning, machine-learning, artificial-intelligence, gaming-tools
- platform: python
- tags: mappo, ppo, multi-agent, marl, smac, hanabi, starcraft2, google-football, pytorch, research-code, gpu, linux

## Member repositories
- marlbenchmark/on-policy (main) score 32

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:11.871456+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T02:55:50.459923+00:00, confidence not recorded.
  - readme: https://github.com/marlbenchmark/on-policy (fetched 2026-08-28T04:06:11.871456+00:00, sha 031d006c12c2)
  - homepage: https://sites.google.com/view/mappo (fetched 2026-08-29T10:35:47.604358+00:00, sha 39ffe723e5ad)
- Data as of 2026-08-30T08:39:29.467469+00:00.