GenesisLab#

GenesisLab is a unified, extensible framework for robot learning built on top of Genesis, the world’s fastest physics simulation platform.

What is GenesisLab?#

GenesisLab is a comprehensive robotics learning framework that provides:

  1. Manager-Based Architecture: A modular design with specialized managers for observations, actions, rewards, commands, and more.

  2. Multi-Robot Support: Pre-configured support for various robot platforms including Unitree (Go2, H1, G1), ANYmal-D, and more.

  3. Gymnasium Integration: Seamless integration with the Gymnasium API for RL training.

  4. Flexible Task System: Easy-to-extend task definitions for various robot learning scenarios.

  5. Advanced Sensor System: Support for both fake (fast) and Genesis native sensors including cameras, LiDAR, and IMU.

  6. Terrain Generation: Procedural terrain generation with curriculum support.

Key Features#

Compared to other robot learning frameworks, GenesisLab offers:

  • 🚀 Ultra-Fast Simulation: Leveraging Genesis’s parallel simulation capabilities for unprecedented training speed

  • 🎯 Manager-Based Design: Clean separation of concerns through specialized manager components

  • 🤖 Multi-Robot Ready: Support for legged robots, humanoids, and wheeled robots out-of-the-box

  • 🔧 Highly Configurable: Configuration-driven design using @configclass decorators

  • 📊 Rich Observation Space: Support for privileged observations, proprioception, exteroception, and more

  • 🎮 Command-Based Control: Flexible command system for directing robot behaviors

  • 🌍 Diverse Terrains: Procedural terrain generation with multiple terrain types and difficulty curricula

Architecture Overview#

GenesisLab follows a manager-based architecture where different aspects of the robot learning problem are handled by specialized managers:

  • Scene Management: LabScene provides the central interface for simulation management

  • Observation Manager: Handles all observation terms and computes observation tensors

  • Action Manager: Processes actions and applies them to the robot

  • Reward Manager: Computes multi-term reward functions

  • Termination Manager: Defines episode termination conditions

  • Command Manager: Generates and manages robot commands

  • Curriculum Manager: Implements training curricula for progressive learning

  • Event Manager: Handles randomization and domain randomization events

  • Object Manager: Manages dynamic objects in the scene

Getting Started#

Quick Installation#

# Clone the repository
git clone https://github.com/yourusername/genesislab.git
cd genesislab

# Install GenesisLab
cd source/genesislab
pip install -e .

# Install Genesis
pip install genesis-world

# Install PyTorch
pip install torch torchvision torchaudio

First Steps#

  1. Run Your First Environment

import gymnasium as gym
import genesislab.envs  # Register environments

# Create environment
env = gym.make("GenesisLab-Go2-Flat-v0", num_envs=1024)

# Reset and run
obs, info = env.reset()
for _ in range(1000):
    action = env.action_space.sample()
    obs, reward, terminated, truncated, info = env.step(action)
    
env.close()

  1. Train with RL Libraries

GenesisLab is compatible with popular RL libraries:

# Example with stable-baselines3
from stable_baselines3 import PPO
import genesislab.envs

env = gym.make("GenesisLab-Go2-Rough-v0", num_envs=4096)
model = PPO("MlpPolicy", env, verbose=1)
model.learn(total_timesteps=10_000_000)

Documentation Structure#

User Guide

API Reference

Community and Support#

  • GitHub Issues: Report bugs and request features

  • Discussions: Ask questions and share ideas

  • Contributing: We welcome contributions! See our contributing guidelines

Citation#

If you use GenesisLab in your research, please cite:

@misc{GenesisLab,
  author = {GenesisLab Contributors},
  title = {GenesisLab: A Unified Framework for Robot Learning with Genesis},
  year = {2024},
  url = {https://github.com/yourusername/genesislab}
}

And don’t forget to cite Genesis:

@misc{Genesis,
  author = {Genesis Authors},
  title = {Genesis: A Generative and Universal Physics Engine for Robotics and Beyond},
  month = {December},
  year = {2024},
  url = {https://github.com/Genesis-Embodied-AI/Genesis}
}

License#

GenesisLab is released under the MIT License. See LICENSE for details.