Repolode
Submit
DLR-RM

DLR-RM/rl-baselines3-zoo

A training framework for Stable Baselines3 reinforcement learning agents, with tuned hyperparameters, pre-trained agents and hyperparameter optimization scripts.

View on GitHubWebsite
Stars
2,900
Forks
612
Open issues
62
Last commit
5 days ago

What is it?

RL Baselines3 Zoo is a training framework for reinforcement learning built on Stable Baselines3. It provides command-line scripts to train and evaluate agents, tune hyperparameters, plot results and record videos of agents.

The project also ships a collection of tuned hyperparameters for common environments and algorithms, along with agents trained using those settings. Hyperparameters are defined per environment in YAML files, so a training run can be started with a single command such as specifying an algorithm and environment ID. Trained agents can be replayed with an enjoy script, and the repository covers environments from Atari, classic control, Box2D, PyBullet, MuJoCo, robotics and MiniGrid.

Integrations include Weights & Biases for experiment tracking and Hugging Face for storing and sharing trained models. The project is intended for users who want a simple way to train RL agents, benchmark algorithms and reuse pretrained settings.

Key features

  • 01Command-line scripts for training, evaluating and replaying agents
  • 02Tuned hyperparameters per environment and algorithm, defined in YAML files
  • 03Hyperparameter optimization tooling
  • 04Pre-trained agents available on Hugging Face
  • 05Integrations with Weights & Biases and Hugging Face
  • 06Plotting scripts for training results

Use cases

  • Training RL agents on Gym or PyBullet environments without writing boilerplate
  • Benchmarking different RL algorithms on the same tasks
  • Searching for good hyperparameters for a new environment
  • Reusing pretrained agents to watch or evaluate policies

Quick start

  1. 1

    Install the rl_zoo3 package from PyPI

    pip install rl_zoo3
  2. 2

    Train a SAC agent on HalfCheetahBulletEnv-v0 with periodic evaluation (run from the cloned repo)

    python train.py --algo sac --env HalfCheetahBulletEnv-v0 --eval-freq 10000 --eval-episodes 10 --n-eval-envs 1

Things to know

  • Full installation needs system packages such as swig, cmake and ffmpeg.
  • The README warns that pybullet may fail to build on macOS without a special CFLAGS workaround.
  • The benchmark results come from single runs, so they are not a quantitative benchmark.

Topics

#deep-reinforcement-learning#stable-baselines3#hyperparameter-optimization#pytorch#gym#robotics#benchmark#hyperparameter-search#hyperparameter-tuning#lab

FAQ

What is rl-baselines3-zoo?
A training framework for Stable Baselines3 reinforcement learning agents, with tuned hyperparameters, pre-trained agents and hyperparameter optimization scripts. RL Baselines3 Zoo is a training framework for reinforcement learning built on Stable Baselines3. It provides command-line scripts to train and evaluate agents, tune hyperparameters, plot results and record videos of agents.
Is rl-baselines3-zoo free to use? What license does it have?
rl-baselines3-zoo is open source under the MIT license. Check the license terms before using it in commercial products.
Is rl-baselines3-zoo still maintained?
Actively maintained. The last commit to rl-baselines3-zoo was on Oct 16, 2026.
What can I use rl-baselines3-zoo for?
Training RL agents on Gym or PyBullet environments without writing boilerplate Benchmarking different RL algorithms on the same tasks Searching for good hyperparameters for a new environment Reusing pretrained agents to watch or evaluate policies
How do I install rl-baselines3-zoo?
Install the rl_zoo3 package from PyPI: pip install rl_zoo3 Train a SAC agent on HalfCheetahBulletEnv-v0 with periodic evaluation (run from the cloned repo): python train.py --algo sac --env HalfCheetahBulletEnv-v0 --eval-freq 10000 --eval-episodes 10 --n-eval-envs 1
Where can I get rl-baselines3-zoo?
The source code and installation instructions are on GitHub: https://github.com/DLR-RM/rl-baselines3-zoo

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects