Formats
URDF/Xacro, MJCF, SDF and USD serve different composition, physics, rendering and tooling needs.
Open engineering references · 04
Simulation models, robot descriptions & digital assets
Find model collections and simulation ecosystems, then judge geometry, dynamics, licence, provenance and validation separately. A model that loads is not necessarily a model that predicts.
Read the asset type first
Different files answer different questions; one cannot stand in for another.
Parameters
Mass, inertia, friction, damping, actuator response, contact and sensor noise require provenance and calibration.
Validation
Unit tests, trajectory comparisons, timing, contacts and task-level transfer reveal where the model is trustworthy.
Rights
Code, XML, meshes, textures, trademarks and converted vendor assets may each carry different terms.
Official and first-party project gateways
Every record states what is available—and what it still cannot prove.
Sources checked 4 August 2026. External projects change; confirm version, licence and maintenance status again at the original source before use.
01
Cyberbotics and Webots community
Cyberbotics · Webots assets
- Robot, actuator and sensor PROTO assets
- Worlds, controllers and reusable nodes
- webots.cloud catalogue and external repositories
The primary Webots repository is Apache-2.0; externally hosted PROTOs and vendor-derived assets retain their own terms.
Mature open simulator with bundled and community-hosted assets
Discover diverse robots, devices and environments for controller and simulation workflows.
Catalogue inclusion does not establish provenance, physical calibration or redistribution rights for every asset.
Freeze every PROTO dependency, hosting commit and licence before treating a Webots world as repeatable evidence.
02
Drake development team
Drake · model directives and package maps
- YAML model directives
- URDF and SDFormat assembly
- PackageMap local and remote resource resolution
Drake uses BSD-3-Clause with noted permissive exceptions; referenced model packages retain independent licences.
Established model-based design and verification toolbox
Assemble traceable multibody scenarios for dynamics, planning, optimisation and control.
A resolved model graph is not a physical validation, manufacturing package or rights clearance for its dependencies.
Publish flattened directives, resolved package checksums and the exact proposition supported by each verification result.
03
Google DeepMind and model contributors
Google DeepMind · MuJoCo Menagerie
- MJCF robot models
- Meshes and scenes
- Per-model README, licence and changelog
Every model directory can have different terms; the top-level Apache-2.0 licence does not override per-model licences.
Curated simulation collection with model-specific quality and maintenance
Discover arms, quadrupeds, humanoids, mobile manipulators and end effectors for MuJoCo research.
The repository explicitly distinguishes model XML, assets and licences. Models are not manufacturing drawings and physical accuracy varies.
Start from a model’s own README and licence, then create a validation card for the behaviours your experiment actually relies on.
- USD robot assets across multiple form factors
- Physics and articulation metadata
- Validated and in-validation asset indicators
Isaac Sim and asset rights follow NVIDIA and source-asset terms; availability in the browser does not imply an open-hardware or redistribution licence.
Vendor-maintained simulation ecosystem with asset-specific validation status
Locate wheeled, manipulator, quadruped, aerial and humanoid assets for Isaac workflows.
A USD asset supports simulation workflows, not physical reproduction. Check product version, asset root, source rights and the page’s validation marker.
Use the catalogue to compare representation coverage, then retain exact asset paths and versions so experiments remain reproducible.
05
Open Robotics and Gazebo community
Gazebo · SDF worlds and models
- SDF world and model structure
- Model inclusion and resource paths
- Physics, sensors and environment composition
Gazebo code and each downloaded model or Fuel asset can have different licences; inspect asset metadata before use or redistribution.
Mature open simulation framework; model quality is contributor- and configuration-dependent
Learn how robot models, sensors and environments are composed into versioned simulation worlds.
A tutorial model demonstrates format and workflow. It does not establish dynamic fidelity, safety or licence for unrelated assets.
Preserve the SDF version, model URI, resource snapshot, physics settings and test outputs; otherwise a world may change beneath a published result.
- Loader access to 185+ descriptions
- URDF, Xacro and MJCF paths
- Integration with multiple robotics frameworks
Loader code is Apache-2.0; every downloaded description retains its own repository and licence.
Convenient discovery and loading layer over heterogeneous upstream models
Quickly locate robot descriptions and inspect them across supported frameworks.
Automatic download is not automatic rights clearance or validation. Record the upstream commit, licence and generated Xacro arguments.
Use it as a catalogue and reproducibility aid, while keeping provenance and model validation attached to each individual robot description.
Before adoption
Move from ‘I found the files’ to ‘I know the next validation step’.
- 01
Record source repository, commit or release, model variant, format version and every conversion step.
- 02
Inspect licences separately for code, description files, meshes, textures, trademarks and vendor-derived assets.
- 03
Validate only the behaviours required by the task: kinematics, dynamics, contacts, actuators, sensors, timing or rendering.
- 04
Never treat a successful simulation or sim-to-real transfer as evidence of complete physical safety.