Model-Physical Decoupling · Task-Centered Runtime

Robonix

One OS, Intelligence Across Embodiments.

Robonix is the open, agentic operating system that turns models, skills, services, and hardware into composable capabilities. Define once, deploy anywhere, and without vendor lock-in or repeated integration.

Get Started ⭐ Star on GitHub

10+Robot Vendors6Robot Form Factors3Simulation EnginesFully Open Skill Forge

Scroll

Problem Robonix Solves

Robonix aims to address the tight coupling of robot models and skills with specific embodiments, which results in duplicated development, poor portability and reuse, and difficulties in scaling robotic applications.

Problem / Tightly coupled robot hardware and AI models

System-layer hardware-model-skill decoupling

Robonix abstracts robot bodies and heterogeneous hardware capabilities downward, while connecting upward to LLMs, VLMs, VLAs, and world models through stable system contracts.

01

Problem / Fragmented interfaces across vendors and devices

Unified primitives, services, skills, and tasks

Robot vendors wrap joints, sensors, mobile bases, arms, and other capabilities as hardware primitives. Applications and skills call perception, mapping, navigation, speech, and motion through consistent interfaces.

02

Problem / High cost to reuse and migrate robot software

Install, compose, reuse, and migrate skills

Robonix aims to make acquiring new robot capabilities as easy as installing applications: skills can be developed once, combined with services, and moved across compatible robot bodies.

03

Problem / Robots need a brain system, not isolated scripts

Task planning, validation, execution, management, and safety

Positioned as the robot brain system, Robonix connects user interaction, task planning, skill orchestration, service calls, hardware primitives, task lifecycle management, and safety guards into one runtime.

04

Twelve Modules, One Robot Brain

From user intent to physical action, Robonix separates interaction, planning, execution, scene state, body state, transport, and capability discovery into clear runtime roles.

Task-Centered Runtime

Every user request becomes an explicit RTDL plan that can be observed, streamed, cancelled, and broken into concurrent capability calls.

Chronos syncs system time

User

intentfeedbacktaskplanfeedbackmonitorqueryservice callprimitives

Custom Services

Scene-specific services

External hardware adapters

Hardware primitivesmotion · sensing · voice

Skillsclassic control · VLA

Hardwarechassiscameralidararmspeakermicrophone

Active module

Pilot

VLM-driven planning engine. Queries Atlas for capabilities, builds prompts, emits RTDL plans for Executor.

Understand task, inspect scene/body state, generate and validate task plans.

GitHubsystem/pilot

Application Scenarios

The same task runtime spans simulation, real robots, natural-language interaction, scene understanding, and VLA skill deployment.

SIM

Webots Simulation

Run office, apartment, complete apartment, break-room, and kitchen worlds with RGB-D camera, lidar, IMU, mapping, navigation, and task execution in one reproducible stack.

5 built-in worlds

Multi-container ROS graph

rmw_zenoh by default

CI-friendly bring-up

MAP

Scene-Aware Tasks

Scene maintains live objects, semantic relations, occupancy grids, and approach goals. Pilot can ask what exists now and plan against the current environment.

Object registry

Semantic relations

2D/3D web viewer

Goal-near planning

BOT

Real Robot Deployment

Published deployments cover AgileX Ranger Mini v3 and DEEP Robotics Lite3, with manifests that assemble primitives, services, skills, and body descriptions.

Mobile robot

Quadruped

RGB-D + lidar

Robot catalog

VLA

Skill Installation

Skills wrap reusable task flows or learned VLA policies. The Robonix Skill Toolkit supports data collection, OpenVLA-OFT fine-tuning, and robot-arm deployment.

Task-facing packages

VLA policy workflow

Independent install

Composable services

Live Demo

See Robonix in action — from natural language commands to real robot execution.

Cross-Platform Hardware Decoupling Verification

To verify its hardware decoupling execution capabilities, Robonix successfully conducted cross-platform demonstrations of the same task workflow on both the AgileX Ranger rover and the DeepRobotics Lite3 quadruped robot. The tests confirm that Robonix can seamlessly bridge different hardware morphologies, directly mapping natural language navigation commands into the robot's actual execution behaviors, while maintaining a real-time feedback mechanism based on environmental perception.

6:51

Supported Hardware

Real robot platforms, standalone devices, and simulation targets validated with Robonix.

Manufacturer	Model	Type	Status	Task	Tested
AgileX Robotics	AgileX Ranger Mini v3 ↗	Wheeled	Integrated System	Navigation & Patrol	✓ Tested
DEEP Robotics	DEEP Robotics Lite3 ↗	Quadruped	Integrated System	Navigation & Patrol	✓ Tested
AgileX Robotics	AgileX Piper	Robotic Arm	Standalone Device	Grasping	— Not Tested
AgileX Robotics	AgileX Nero	Robotic Arm	Standalone Device	Grasping	— Not Tested
BeingBeyond	BeingBeyond D1 ↗	Dexterous Hand	Standalone Device	Grasping	✓ Tested
Hantewin	Hantewin Benben	Service Robot	Standalone Device	Service	— Not Tested
Unitree	Unitree Go2 ↗	Quadruped	Standalone Device	Navigation & Patrol	✓ Tested
DEEP Robotics	DEEP Robotics Lynx	Quadruped	Standalone Device	Navigation & Patrol	— Not Tested
Mojang / Forge	Minecraft ↗	Simulation	Simulation	Simulation	✓ Tested

Ecosystem

Robot vendors, model providers, research institutions, and open-source communities in the Robonix ecosystem.

Team & Community

Developed by the Syswonder open-source community, released by the CCF Ubiquitous Operating System Open Community, and built by researchers and engineers from leading institutions.