Beyond geometry
Describe what a body is, what it can do, what constrains it and where that knowledge came from.
Menu
ROBOT GRAPH DEFINITION
A machine-readablestandard for cognitive embodiment.
OpenRGD gives AI systems a structured way to understand a robot body, its operational limits, capabilities, lifecycle and relationships with other agents.
OPEN STANDARDMACHINE READABLENON-ACTUATING BY DESIGN
01
stableFoundation
Physical body and hardware reality: structure, joints, inertials, actuation evidence and source provenance.
01description
02actuation
03sensors
04hardware
semantic domain01_foundation
Standard bundle0.2.0DRAFT
Python toolchain0.1.1ALPHA
Agent contracts0.1.0CANDIDATE
WHY OPENRGD
The missing semantic layer between AI and robots.
Robot description formats are excellent at describing geometry, joints and simulation state. AI systems need more: explicit capability, operational boundaries, provenance and lifecycle context.
Grounded AI context
Compile robot knowledge into deterministic, machine-readable context without silently authorizing physical execution.
Interoperability boundary
Keep semantic description, middleware export and embodied runtime responsibilities explicit and independently verifiable.
CANONICAL DOMAIN MODEL
One robot. Seven semantic domains.
Each domain carries its own authority and maturity instead of hiding assumptions inside one monolithic robot description.
00
Core
Coordination, manifests and kernel metadata.
01
Foundation
Physical body, hardware reality and structural evidence.
02
Operation
Runtime constraints, safety and physiological operation.
03
Agency
Capabilities, world model and action interfaces.
04
Volition
Values, alignment and decision governance.
05
Evolution
Lifecycle, wear, adaptation and continuity.
06
Ether
Collective, social and inter-agent protocols.
TECHNOLOGY CONTEXT
OpenRGD sits inside a larger embodied systems stack.
These concepts explain why a machine-readable embodiment layer matters and where OpenRGD stops.
PHYSICAL AI
From perception and simulation to real-world action.
Explore →PHYSICAL AI STACKModels, embodiment, safety, runtime and hardware as explicit layers.
Explore →VLAFrom multimodal perception and language to robot actions.
Explore →ROBOT SAFETYWhy safe physical execution is a multilayer systems problem.
Explore →DESCRIPTION FORMATS
One robot can exist across several description layers.
URDF, USD, SDF, MJCF and RGD overlap on physical facts, but each has a different center of responsibility.
SHARED KNOWLEDGE LAYER
Canonical facts. Independent OpenRGD context.
OpenRGD.org reads shared canonical entities from the multi-site knowledge layer, while its summaries, SEO and editorial framing remain specific to OpenRGD.
TECHNOLOGYFEATURED
Embodied AI
AI systems that perceive, reason, learn and act through an embodiment interacting with an environment.
canonical entityembodied-ai
TECHNOLOGYFEATURED
Physical AI
AI systems that perceive, reason and act in the physical world through autonomous machines and robots.
canonical entityphysical-ai
ROBOTFEATURED
RoBee R
A full-size industrial humanoid robot by Oversonic Robotics designed to operate in human-built environments.
canonical entityrobee-r
ROBOTFEATURED
Unitree G1
A compact humanoid robotics platform designed for research, development and embodied intelligence experimentation.
canonical entityunitree-g1
ORGANIZATIONFEATURED
Unitree Robotics
Robotics company developing quadruped and humanoid platforms widely used for research, development and embodied AI experimentation.
canonical entityunitree-robotics
ORGANIZATION
Comau
Italian industrial automation and robotics company with a long history in advanced manufacturing, collaborative robotics and robotic systems.
canonical entitycomau
ROBOT
Comau Racer-5-0.80 COBOT
A 6-axis collaborative robot combining industrial performance with collaborative operating modes.
canonical entitycomau-racer-5-cobot
ORGANIZATION
Oversonic Robotics
Italian robotics company developing cognitive humanoid systems for industrial, healthcare and intralogistics environments.
canonical entityoversonic-robotics
OFFICIAL SURFACES
Only live OpenRGD surfaces are linked from the public hub.
Experimental and future services stay out of the public directory until they are reachable, maintained and ready to represent the project.
openrgd.org/docs→
Developers
Reference material, import/export guides and validation workflows.
openrgd.org/blog→
Engineering Journal
Source-backed architecture notes, experiments and project decisions.
openrgd.org/robots→
Robot Knowledge
Robot profiles and source-backed embodiment records.
openrgd.org/research→
Research
World models, motion, ethics, timing and physical experiments.
openrgd.org/governance→
Governance
Maturity, normative change, RFC and release decision rules.
github.com/OpenRGD/openrgd↗
Canonical Repository
Normative source, contracts, governance, security policy and toolchain.
openrgd / terminal● ● ●
$ rgd init Robot
$ cd Robot
$ rgd import robot.urdf --out partial-robot
✓ foundation evidence extracted
$ rgd hash
OPENRGD_SOURCE_TREE_SHA256_V1
9d0f…a71c
$ rgd check --output json
{ "integrity": "verified" }
$ rgd boot --output json
{ "execution_authorized": false }EVIDENCE → CONTEXT
Import what the source proves. Invent nothing.
OpenRGD can extract source-supported body evidence from URDF and USDA, preserve provenance and compile a deterministic non-actuating grounding context.
01URDF / USD→
02IMPORT→
03RGD GRAPH→
04VERIFY→
05COMPILE
RUNTIME BOUNDARY
Semantic intelligence stops before physical authority begins.
OpenRGD deliberately separates cognitive proposals from actuation. Runtime safety and body control remain explicit downstream responsibilities.
01CognitionProposalCOGNITION
02ActionIntent
03Somatic Translator
04CapabilityPlan
05Operation Safety Gate
06DecisionTrace
07Body Adapter
08HardwarePHYSICAL
OPEN ECOSYSTEM
Built to connect, not to replace everything.
OpenRGD is a semantic standard and toolchain root. It can coexist with robot description formats, middleware and physical runtimes.
Robot descriptions
URDF and supported OpenUSD evidence can enter the graph without losing provenance.
Static exports
Deterministic outputs can target middleware configuration without crossing the actuation boundary.
Contracts
Cross-component interfaces expose maturity, provenance and explicit responsibility.
Shared knowledge
Canonical robot facts can be reused across documentation, research and ecosystem tooling.
OPEN SOURCE · MIT LICENSE