Generate large-scale synthetic training data for robot manipulation without real demonstrations.
Teleoperate a simulated dual-arm robot manipulating cloth in real time.
Export training trajectories in LeRobot-compatible formats.
Reproduce or extend the SIM1 paper's experiments on deformable-object simulation.
| internrobotics/sim1 | nvlabs/isaaclabeureka | murphylmf/unish | |
|---|---|---|---|
| Stars | 141 | 138 | 145 |
| Language | Python | Python | Python |
| Last pushed | — | 2025-10-28 | — |
| Maintenance | — | Quiet | — |
| Setup difficulty | hard | moderate | hard |
| Complexity | 5/5 | 4/5 | 5/5 |
| Audience | researcher | researcher | researcher |
Figures from each repo's GitHub metadata at analysis time.
Requires a CUDA GPU, Conda, cloning with submodules, and downloading large asset files before use.
SIM1 is a research codebase from InternRobotics that accompanies an academic paper about a physics based simulator for robots handling cloth and other deformable objects, such as folding or manipulating fabric with two robotic arms. The core idea, as described in the README, is that this simulator can generate large amounts of realistic training data for robot learning without needing real world demonstrations for every scenario, because its physics closely match reality. It is built on top of two existing physics and simulation projects, called Newton and Warp, and the repository covers the entire pipeline from interactive teleoperation and synthetic data generation through to rendering and exporting data in a format compatible with LeRobot, a popular robot learning framework. This is unambiguously a tool for robotics and machine learning researchers rather than casual users. Installation requires Python 3.11 managed through Conda, and CUDA 12.4 or newer if GPU acceleration is desired. The setup process involves cloning the repository with its submodules included, running a provided setup script that installs all necessary dependencies, and then running a separate script to download required assets such as robot models and cloth simulation files, which the README notes must happen before any data generation can occur. Once installed, the project offers a keyboard controlled interactive teleoperation mode, where a person can manually operate a simulated dual arm robot to manipulate cloth, with an option to stream this session over a websocket for remote viewing and control. There is also a full data generation pipeline, run through a single script, that produces synthetic training trajectories by generating data, smoothing it, replaying it in two different export formats, and filtering it before it is ready for use. The project is released under the Apache 2.0 license, a permissive open source license allowing reuse and modification with proper attribution, including for commercial purposes. Given the heavy dependence on GPU hardware and physics simulation libraries, getting this running is a substantial undertaking rather than a quick install.
A research simulator for generating synthetic robot training data on deformable objects like cloth, from an academic paper on physics-aligned simulation.
Mainly Python. The stack also includes Python, Newton, Warp.
Use freely for any purpose, including commercial use, as long as you keep the copyright and license notice.
Setup difficulty is rated hard, with roughly 1day+ to a first successful run.
Mainly researcher.
This repo across BitVibe Labs
Verify against the repo before relying on details.