Paper: From OpenUSD to DSX: NVIDIA's Modular Physical-Simulation Infrastructure for Digital Twins, Robotics, and AI Factories
Status: Working draft — actively seeking contributors
This paper argues that NVIDIA's physical-simulation offering is best understood as a layered infrastructure — not a monolithic Omniverse application. The analysis combines a documentation-grounded architecture review, a development timeline, and three comparative case studies (DSX for AI factories, robotics via Isaac Sim and Newton, and CAE/CFD via Kit-CAE).
The central result is that NVIDIA's industrial significance lies in connecting interoperable scene data (OpenUSD), GPU-native simulation, and blueprint-packaged deployment rather than in any single simulator considered in isolation.
The current draft is written primarily from a systems architecture perspective. To make the paper genuinely useful to the community, we need domain depth that one author can't provide alone.
| Area | What's Needed | Ideal Contributor |
|---|---|---|
| OpenUSD standards & AOUSD | Governance process details, spec evolution timeline, cross-vendor adoption data, interoperability testing results | USD/AOUSD working group members, DCC tool developers |
| DSX production deployment | Real-world deployment experiences, scaling observations, integration patterns, operational challenges | Data center engineers, NVIDIA partners, infrastructure teams |
| Robotics workflows | Sim-to-real transfer benchmarks, Isaac Lab training results, OSMO orchestration at scale | Robotics researchers, Isaac Sim users, Physical AI teams |
| CAE/CFD digital twins | PhysicsNeMo surrogate model accuracy data, Kit-CAE production readiness assessment, solver comparison | CAE engineers, CFD researchers, ISVs |
| Area | What's Needed | Ideal Contributor |
|---|---|---|
| Newton & Warp benchmarking | Performance comparisons, differentiable simulation use cases, integration examples | Physics simulation researchers |
| Competitor analysis | Structured comparison with Unity, MuJoCo ecosystem, Gazebo, cloud digital twin platforms | Anyone with cross-platform experience |
| Industry adoption patterns | How organizations are actually adopting (or struggling to adopt) this stack | Enterprise users, system integrators, consultants |
| App Streaming & browser delivery | Latency measurements, scalability data, user experience studies | Infrastructure engineers, UX researchers |
You contribute a section or case study and are listed as a co-author.
- Open an issue in this repo titled
[Paper] Contribution: <your topic> - Describe what you can contribute and your background
- We'll coordinate on scope, timeline, and the section outline
- You write the section; we integrate and align with the overall argument
- All co-authors review the final draft before submission
You review the draft and provide substantive feedback.
- Read the current draft
- Open issues for factual errors, missing nuances, or structural suggestions
- Reviewers are acknowledged in the paper
You contribute data, benchmarks, or case evidence that strengthens the analysis.
- Open a PR adding your data to the
paper/data/directory - Include a brief description of methodology and context
- Contributors are acknowledged; substantial contributions may warrant co-authorship
- Documentation-grounded: Claims must be traceable to official docs, published benchmarks, or described methodology. No unsourced marketing claims.
- Balanced tone: The paper is analytical, not promotional. Honest assessment of maturity and limitations is as important as describing capabilities.
- Open process: All contributions happen through GitHub issues and PRs. No private side channels for paper content.
- Publication target: We're aiming for a venue that reaches both the simulation/digital-twin community and the broader AI infrastructure audience. Specific venue TBD based on final scope.
The working draft is available at draft.md. It represents the paper's current state and is subject to significant revision as co-authors join.
| Phase | Target |
|---|---|
| Call for co-authors opens | Now |
| Co-author onboarding & scope alignment | 4 weeks |
| Section drafts due | 8 weeks |
| Internal review & integration | 10 weeks |
| Final draft for submission | 12 weeks |
- Open an issue in this repo (preferred)
- Reach out to @tsubasakong on GitHub
- Twitter/X: @gpt3_eth