SimReady assets for
robotics research labs

Build the asset catalog your research actually needs without writing physics by hand. Rigyd converts 3D, images, or text into physics-enabled OpenUSD and MJCF that drop into Isaac Lab, MuJoCo, Gazebo, and Genesis from a single source.

The problem

Why existing workflows fall short.

Asset prep eats research time

Most robotics research papers cite manual asset preparation as days-to-weeks of grad-student time per project. The asset diversity the benchmark numbers depend on rarely matches what the lab can author by hand in the available time.

Reproducibility breaks at the asset layer

Hand-tuned physics values vary across labs. A friction coefficient one team chose by intuition does not transfer when another team rebuilds the benchmark. Reproducibility fails at the asset layer before the policy gets a chance.

Cross-simulator portability is daily friction

Today's robotics research uses multiple simulators in the same paper (Isaac Lab for GPU-parallel training, MuJoCo for contact-rich tasks, Gazebo for ROS validation, Genesis for fast exploration). Re-authoring assets per simulator wastes weeks every project.

How Rigyd helps

AI-native infrastructure that automates the hard parts.

One source, every research simulator

OpenUSD output drops into Isaac Lab, Omniverse, Unreal, Unity, and Gazebo Sim via USD imports. MJCF output drops into MuJoCo, MJX, and Genesis. The same source asset feeds every common research simulator without re-authoring physics multiple times.

Reproducible physics calibration

Every value Rigyd emits is keyed to a documented calibration: mass from volume × material density, friction from surface-material classification, inertia from the inferred mass distribution. The methodology is published, the assumptions are spelled out, and the numbers are reproducible across labs.

Catalog scale for ablation and DR studies

Bulk processing makes domain-randomization sweeps and asset-diversity ablations practical. A 10,000-asset DR study that would take a grad student a month of authoring is a one-day batch job, freeing engineering time for the actual research question.

~5 min

per asset for AI-automated SimReady preparation

5+

simulators consumed from a single Rigyd source

reproducible

methodology published at /research

Spend research time on the policy, not the assets

Generate SimReady catalogs for benchmarks, ablations, and sim-to-real studies in hours rather than weeks.

Starts at $29/month. 30 credits included.

Frequently asked questions

How do I cite Rigyd-generated assets in a paper?

Rigyd Research, June 2026. https://rigyd.com/research. For a specific quantitative claim, cite the section number, e.g. "Rigyd Research §3, June 2026" for the cost-reduction methodology. The /research page documents the calibration model, the engineer-rate assumption, and the sensitivity bands so reviewers can re-derive every number against their own assumptions.

Are Rigyd asset outputs deterministic for reproducibility?

For a given source input and a given Rigyd version, the output is deterministic: same mass, same friction, same collision decomposition. Across Rigyd versions the calibration model can change; for reproducible benchmarks, pin the Rigyd output to a versioned URL (Enterprise API exposes asset versioning so a paper can cite the exact asset hash used at training time, and a follow-up team can pull the same hash).

Does the output work in Isaac Lab and Genesis at the same time?

Yes. Isaac Lab consumes OpenUSD natively (Rigyd's primary output) and MJCF via the MJX backend. Genesis consumes MJCF natively and OpenUSD via its USD loader. The same source asset emitted from Rigyd feeds both research stacks without re-authoring. Many labs run Isaac Lab for GPU-parallel rollout and Genesis for fast contact-rich exploration in the same paper; this works out of the box.

Can a single lab use this for both training and evaluation?

Yes. Most published robotics research separates a training catalog (large, diverse, augmented) from an evaluation catalog (held-out, often hand-curated). Rigyd handles both: bulk generation for training diversity, plus per-asset override for the held-out evaluation catalog where measured ground truth is available. The two flows share a single methodology, which keeps the train-eval comparison honest.

Is academic access available for non-commercial research?

Yes. Academic and non-commercial-research access is available through the Request API Access form. Tell us the lab, the research question, and the rough scale (assets per benchmark, frequency of regeneration), and we will respond with an academic plan. Outputs generated under the academic plan are usable in papers, open-source releases, and reproducibility artefacts without additional licensing.