Skip to content
GenManip

GenManip Documentation

GenManip Suite is a comprehensive robotics simulation platform built upon NVIDIA Isaac Sim, designed to advance research in general manipulation.
It integrates data generation, benchmarking, and baseline methods, enabling seamless development, extension, and evaluation of robotic manipulation policies.

This manual includes installation guides, tutorials, usage instructions, API references, and detailed descriptions of the provided baselines.

About GenManip

Section titled “About GenManip”

From well-crafted, hand-designed scenes to large-scale procedurally generated datasets, GenManip offers a complete workflow spanning precision design to large-scale generalization.
With its streamlined toolchain, you can easily create, extend, and share your own benchmarks and datasets.

We introduce the concept of a GenManip Package — allowing you to install official benchmarks or community-contributed extensions just like adding expansion packs in a game.
All benchmarks share a unified communication interface, offering a black-box experience so you can focus purely on model evaluation without worrying about internal details.
For data organization, GenManip follows LeRobot GR00t best practices, ensuring interoperability with your preferred training frameworks.

For users eager to explore and create, a few simple edits to the configuration file let you launch a custom data generation pipeline or a new benchmark within minutes.
Isaac Sim’s intuitive GUI makes scene editing effortless — focus on what matters, and let GenManip handle the rest.

The “Tutorials” section provides a complete learning path from basic to advanced usage, helping you get started smoothly.
More content is on the way — stay tuned!

For questions or collaboration, feel free to open an Issue or contact us directly at [email protected].
Our technical team actively monitors and responds as soon as possible.

Key Features

Section titled “Key Features”

  1. GenManip Package
    Use simple scripts to install or publish benchmark assets with a single command — deploy and share benchmarks as easily as game expansions.

  2. Benchmark System
    Supports multiple benchmarks, including GenManip Scaling Pick-and-Place and GenManip IROS Benchmark.
    A unified communication interface allows you to seamlessly switch and evaluate across all benchmarks within the GenManip ecosystem.

  3. User-Friendly Documentation & Config Templates
    Comprehensive documentation and rich configuration templates help you get started quickly.
    After setup, you can truly get started with GenManip in seven minutes.

  4. Domain Randomization
    Supports extensive randomization strategies for Objects / Layouts / Dome Lights / Cameras / Textures / Rooms, enabling robust and diverse data generation.

  5. Cross-Embodiment Support
    Out-of-the-box compatibility with multiple robot embodiments: Franka Emika Panda with Panda Hand, Franka with Robotiq 2F-85, Aloha Split, and Lift2.

  6. Multi-Process & Distributed Execution
    Run thousands of Isaac Sim instances concurrently across multiple servers with synchronized progress via shared filesystems.
    In stress tests, up to 1500 Isaac Sim instances ran simultaneously on 500 RTX 4090 (48GB) GPUs for data generation or evaluation.

  7. Rule Set System
    Define task completion logic through a structured syntax supporting top / left / right / front / back / in relations and complex logical compositions — with more under development.

  8. High-Performance Data Generation Pipeline
    A fully decoupled data generation architecture powered by cuRobo motion planning and generalized oracle rules, enabling large-scale dataset creation across single to hundreds of GPUs.

  9. Meta Object System
    Define Meta Objects for flexible scene composition and object substitution, supporting scalable dataset and benchmark expansion.

Example Use Cases

Section titled “Example Use Cases”

SHAILAB IROS Challenge 2025


A well-designed benchmark with ten dual-arm manipulation tasks, supporting both data generation and evaluation.
Learn more at SHAILAB IROS Challenge 2025.

Large-Scale Data Generation & Benchmarking

Large-scale data generation and benchmark pipeline supporting InternData M1, which powers training and evaluation of InternVLA-M1.
See InternData M1 and InternVLA-M1.