Repolode
Submit
mani-skill

mani-skill/ManiSkill

An open-source, GPU-parallelized robot simulation and training framework built on SAPIEN, focused on manipulation skills.

View on GitHubWebsite
Stars
3,495
Forks
561
Open issues
167
Last commit
3 months ago

What is it?

ManiSkill is a robot simulation framework built on SAPIEN, aimed at training and benchmarking robot manipulation skills. Its core idea is to run many simulated environments in parallel on the GPU, which makes it practical to collect large volumes of visual and state-based training data quickly.

The framework supports GPU-parallelized simulation where each parallel environment can hold a different scene or set of objects. Its example tasks span humanoids, mobile manipulators and single-arm robots, covering table-top, drawing, cleaning and dexterous manipulation. A task-building API hides much of the GPU memory management behind an object-oriented design.

ManiSkill also includes tuned baselines for reinforcement learning (PPO, SAC, TD-MPC2), imitation learning (behavior cloning, Diffusion Policy) and vision-language-action models, along with examples for real-to-sim evaluation and sim-to-real deployment. It is installed via pip and works best on Linux with an NVIDIA GPU.

Key features

  • 01GPU-parallelized RGBD and segmentation data collection at high frame rates
  • 02GPU-parallel simulation with heterogeneous scenes across environments
  • 03Object-oriented task building API that abstracts GPU memory management
  • 04Example tasks across humanoids, mobile manipulators and single-arm robots
  • 05Baselines for RL, imitation learning and vision-language-action models

Use cases

  • Training manipulation policies with reinforcement learning in parallel simulation
  • Generating large synthetic datasets for imitation learning
  • Benchmarking robot learning algorithms on standardized manipulation tasks
  • Evaluating real-world policies in simulation before sim-to-real deployment

Quick start

  1. 1

    Install the package with pip

    pip install --upgrade mani_skill
  2. 2

    Install a torch version compatible with your system

    pip install torch
  3. 3

    Set up Vulkan for rendering by following the documentation instructions

Things to know

  • Best supported on Linux; Windows and macOS have limited support.
  • GPU-accelerated simulation is not available on Windows, macOS or WSL.
  • Requires Vulkan setup for rendering.
  • Asset data is licensed CC BY-NC 4.0, which is non-commercial.

In collections

Reinforcement learning frameworks & training environmentsRobotics & embodied AI projects

Topics

#embodied-ai#robotics#robot-learning#reinforcement-learning#simulation#robot-manipulation

FAQ

What is ManiSkill?
An open-source, GPU-parallelized robot simulation and training framework built on SAPIEN, focused on manipulation skills. ManiSkill is a robot simulation framework built on SAPIEN, aimed at training and benchmarking robot manipulation skills. Its core idea is to run many simulated environments in parallel on the GPU, which makes it practical to collect large volumes of visual and state-based training data quickly.
Is ManiSkill free to use? What license does it have?
ManiSkill is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is ManiSkill still maintained?
Updated within 6 months. The last commit to ManiSkill was on Aug 4, 2026.
What can I use ManiSkill for?
Training manipulation policies with reinforcement learning in parallel simulation Generating large synthetic datasets for imitation learning Benchmarking robot learning algorithms on standardized manipulation tasks Evaluating real-world policies in simulation before sim-to-real deployment
How do I install ManiSkill?
Install the package with pip: pip install --upgrade mani_skill Install a torch version compatible with your system: pip install torch Set up Vulkan for rendering by following the documentation instructions
Where can I get ManiSkill?
The source code and installation instructions are on GitHub: https://github.com/mani-skill/ManiSkill

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

Related projects