syswonder/robonix

What it solves

Robonix provides a system substrate (an "operating system") for embodied AI, addressing the problem of decoupling AI models from specific robot hardware. It prevents the need to re-implement core system services for every new robot, allowing developers to "train once, deploy on any robot."

How it works

Robonix treats AI models and skills as programs and robot hardware as discoverable capabilities. It uses a modular architecture consisting of twelve system components (such as pilot for VLM-based planning and executor for task dispatch) and a package-based system. These packages are categorized into:

• Primitives: Individual devices like cameras, lidar, and arms.
• Services: Runtime functionality like mapping and navigation.
• Skills: Reusable execution flows like grasping or exploring.

Who it’s for

It is designed for developers and researchers building embodied intelligence who want to deploy AI models across heterogeneous robot platforms (wheeled, quadruped, arms, etc.) without rewriting hardware-specific code.

Highlights

• Hardware Agnostic: Supports a wide range of robot bodies, including AgileX, Unitree, and DEEP Robotics, as well as simulated environments like Webots and Minecraft.
• Modular Package System: Decouples hardware drivers from high-level skills through shared capability contracts.
• VLM Integration: Uses OpenAI-compatible VLM endpoints to turn user prompts into executable plans.
• Comprehensive System Services: Includes built-in components for environment tracking (scene), health monitoring (vitals), and inter-component communication (nexus).