# agi-brain/xuance

XuanCe: A Comprehensive and Unified Deep Reinforcement Learning Library

Repository: https://github.com/agi-brain/xuance
Canonical: https://ross.abutalabs.com/products/xuance
Homepage: https://xuance.org
Language: Python
License: MIT
License Family: permissive
Topics: multi-agent-reinforcement-learning, reinforcement-learning, reinforcement-learning-library, mindspore, pytorch, tensorflow2, atari, mujoco, mpe, a2c, ddpg, dqn, maddpg, starcraft2, magent, google-research-football, mappo, qmix, ppo
Last push: 2026-09-02T14:49:54+00:00

## Health v2 (maintenance only)
Score: 95/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 100, release rhythm 93, longevity 85
- inputs: {"age_days": 1200, "days_push": 0, "days_rel": 46, "gap_med": 25, "n_releases_24m": 16}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1082, forks 160 (observed 2026-09-03T02:15:08.862104+00:00)

## What it is
XuanCe is an open-source Python library of deep reinforcement learning (DRL) and multi-agent reinforcement learning (MARL) algorithm implementations, supporting PyTorch, TensorFlow, and MindSpore backends. It includes a wide range of algorithms (DQN variants, PPO, SAC, QMIX, MAPPO, model-based and offline RL) with benchmark tooling for environments like Atari, MuJoCo, SMAC, and Google Research Football.

## Use cases
- train deep reinforcement learning agents in Python
- run multi-agent RL experiments like QMIX or MAPPO on StarCraft II
- benchmark DRL algorithms on Atari and MuJoCo environments
- implement custom RL algorithms with a unified API
- switch between PyTorch, TensorFlow, and MindSpore backends for RL research
- experiment with offline and model-based reinforcement learning
- train agents in PettingZoo multi-agent environments

## When to choose
- you need a broad zoo of single-agent and multi-agent RL algorithms in one library
- you want backend flexibility across PyTorch, TensorFlow, and MindSpore
- you need reproducible benchmarking against standard RL environments
- you are doing RL research and want readable, extensible implementations

## When to avoid
- you need a production RL serving or deployment system rather than a research library
- you only need a single specific algorithm with minimal dependencies
- you require non-Python or real-time embedded RL support

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, reinforcement-learning, deep-learning, benchmarking, sdk
- domain: reinforcement-learning, machine-learning, deep-learning, artificial-intelligence
- platform: python, cross-platform
- tags: reinforcement-learning-library, multi-agent-reinforcement-learning, marl, pytorch, tensorflow, mindspore, gymnasium, pettingzoo, atari, mujoco, starcraft2, ppo, qmix, maddpg, offline-rl, model-based-rl, algorithms, research, gpu

## Member repositories
- agi-brain/xuance (main) score 95

## Provenance
- Observed fields: from GitHub, fetched 2026-09-03T02:15:08.862104+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 2026-08-30T06:53:02.969890+00:00, confidence not recorded.
  - readme: https://github.com/agi-brain/xuance (fetched 2026-09-03T02:15:08.862104+00:00, sha 4a18154220e8)
  - homepage: https://xuance.org (fetched 2026-08-29T12:55:17.705290+00:00, sha dbc728d8b406)
  - registry_pypi: https://pypi.org/pypi/xuance/json (fetched 2026-08-29T12:55:17.712041+00:00, sha 22d049824e22)
- Data as of 2026-08-30T08:39:29.467469+00:00.