Connect to the hackathon WiFi
All robots have been pre-configured to connect to the Robot Wifi network at YC. Your laptop and phone must be on the same network to connect to your MARS.The WiFi password is pinned in Discord. Make sure you are connected before attempting to SSH or use the Innate Controller App.
Update your robot first
Hackathon resources repository
The robohacks-utils repo is the central hub for hackathon-specific tooling and assets. Clone it to get started:- HDF5 → LeRobot converter (
hdf5_lerobot.py) — convert recorded HDF5 datasets into the LeRobot format for policy training. Install dependencies withpip install -r requirements.txt. - STEP files (
STEP/) — CAD models of the MARS robot for hardware modifications or custom attachments. - Latest Android APK — attached to the repo releases. The iOS app is available on TestFlight. Leader Arm operation only works on Android. We have some Android phones to share if no one on your team has one.
Innate OS authentication
To run Innate OS or the simulator on your own machine with access to the voice, agent, and training servers, you need an Innate service key.| Track | How to get your key |
|---|---|
| Robot track | Look at ~/innate-os/.env on your robot and copy INNATE_SERVICE_KEY. One key per team. |
| Sim track | Ask in the #innate-cloud-infra channel on Discord. One key per team — you can share it. |
Sim track — MARS URDF
If you are on the simulation track, you must integrate the MARS URDF into your submissions. You can find it here:MARS URDF (maurice_sim)
The simulation description package for the MARS robot, including URDF, meshes,
and launch files.
Training runs
Track your cloud training runs on Weights & Biases:W&B Dashboard
View training metrics, loss curves, and run comparisons for the ACT policy.
Open-source repos & example projects
Innate OS is fully open source. Use these repos and example projects as a starting point for your hack.Innate OS
The full operating system that runs on MARS. Explore it to understand how the robot works under the hood — ROS2 nodes, launch files, drivers, and more.innate-os
The open-source ROS2-based operating system for MARS.
GraspGen — arbitrary object grasping
GraspGen is a project built on top of Innate OS that uses the depth camera and NVIDIA’s GraspGen model to grasp arbitrary objects. It is a great reference for how to:- Access core robot interfaces through ROS Bridge (port 9090)
- Control the robot over WebSockets
GraspGen
Depth-camera-based arbitrary object grasper built on Innate OS and NVIDIA
GraspGen.
Vision-Language Navigation (UniNaVid)
A remote inference server using the UniNaVid model for vision-language navigation. The server processes navigation commands with a large VLM, and the client inside Innate OS executes the resulting movements.- Server — remote inference endpoint for navigation commands.
- Client — the ROS2 package inside Innate OS that calls the server and executes movement primitives. You can find it at
ros2_ws/src/cloud/innate_uninavid/, along with a corresponding skill atskills/navigate_with_vision.py.
Useful docs pages
Quick Start
Power up, connect, and control your MARS in minutes.
Development Setup
SSH in, edit code, restart — the full dev loop.
Agent SDK
Build agents, skills, and inputs on top of Innate OS.
Manipulation & Training
Collect data, train policies, and deploy arm skills.
Need help?
Join the Innate Discord for real-time support from the team and other hackers:Discord
Ask questions, share progress, and get debugging help.