syswonder/robonix

Robonix – An Agentic Operating System for Embodied AI

What it is – Robonix is an open‑source, general‑purpose operating system that lets you control many different robots with natural‑language commands. It sits between large language / vision‑language models (LLMs/VLMs) and the robot hardware, turning a user’s request into a Robot Task Description Language (RTDL) program that the robot then executes.

Why it matters – By decoupling hardware from software through layered abstractions (primitive → service → skill → task), the same high‑level code can run on wheeled bases, quadrupeds, arms, hands, or even simulated avatars. This makes it far easier to reuse perception, planning, and manipulation capabilities across platforms and to let LLMs generate robot‑specific plans on the fly.

Key concepts & architecture

Layer	Role
Primitive	Direct hardware I/O (sensors, actuators) exposed via a versioned interface.
Service	Reusable software that implements a Robonix interface – e.g., SLAM, navigation, object detection.
Skill	Learned model or algorithm packaged as an executable unit (e.g., pick‑and‑place, pouring, VLA models).
Task	A user‑level goal with constraints and success criteria.

The system is built around several core components (see the architecture diagram in the README):

Liaison – front‑end that authenticates users via Keystone and receives natural‑language requests.
Pilot – the orchestrator that asks an LLM/VLM to write an RTDL program for the requested task, validates it, and forwards it to the Executor.
Executor – runs the RTDL program on the robot.
Sentinel – monitors execution for safety and health.
Scene – maintains a semantic map of objects and a spatial map of the environment.
Atlas, Nexus, Chronos, Scribe, Vitals – provide service discovery, messaging, timing, logging, and health‑checking infrastructure.
Soma – the body abstraction (URDF + YAML) that describes the robot’s geometry and the primitives it offers.

All components register versioned contracts with Atlas, so callers bind to the contract rather than a concrete implementation. This enables swapping a skill or service without changing the higher‑level code and even moving a task from one robot body to another.

Getting started (from the README)

Install prerequisites – Rust, the Python‑oriented tool uv, Docker (for the Webots simulator), NVIDIA Container Toolkit (if you have an NVIDIA GPU), plus standard build tools (build-essential, pkg-config, OpenSSL headers, ALSA utils).

Clone and build

git clone --recursive https://github.com/syswonder/robonix.git
cd robonix
make install   # builds Rust components and the `rbnx` CLI

Run a simulator (Webots example) in one terminal:

export DISPLAY=:0
bash examples/webots/sim/start.sh

Start Robonix in another terminal, pointing it at an OpenAI‑compatible VLM endpoint:

export RMW_IMPLEMENTATION=rmw_zenoh_cpp
export VLM_BASE_URL=https://api.openai.com/v1
export VLM_API_KEY=sk-…
export VLM_MODEL=your-model-name
cd examples/webots
rbnx build && rbnx boot

Chat with the robot – open a third terminal and run rbnx chat. Try commands like go to room 101, what can you see?, or explore the office.

For developers, the Robonix package template (template‑rbnx) provides a mock primitive, service, and skill that can be built and run without any robot hardware. Use rbnx caps to list live capability providers and rbnx chat to interact with them.

Supported robots & ecosystem

Robonix ships a package catalog (https://packages.robonix.ai) that lists ready‑to‑use robot manifests. The README shows dozens of real and simulated platforms, including:

AgileX Ranger Mini v3, DEEP Robotics Lite3 & Lynx S10, Unitree Go2, Yobotics Y20W, MirrorMe BPX, Wheeltec R550, Yahboom ROSMASTER X3, Hantewin Benben, BeingBeyond D1, AUBO i5H, LinkerHand O6, etc.
Simulated bodies such as Webots TIAGo Lite, Minecraft Bot, MuJoCo‑based Ranger+Piper, and Unitree Go2.

Each robot entry links to a repository containing its hardware drivers and a catalog page that details its primitives, services, and skills. Because all robots expose the same Robonix contracts, the same high‑level RTDL programs can be run on any of them.

License

The project is released under the Mulan PSL‑2.0 license (see the badge in the README).

Bottom line – Robonix provides a modular, contract‑driven stack that lets LLMs generate executable robot programs, abstracts away hardware differences, and offers a growing library of robot‑specific packages. It is positioned as a substrate for research and product teams that want to build “agentic” robots that can be instructed in natural language and verified before execution.