# NVIDIA just put 2,575 simulation-ready 3D assets on Hugging Face ![[images/NVIDIA-SimReady-Assets-thumbnail.jpg]] - **Post:** [x.com/oliviscusAI/status/2114556298167505225](https://x.com/oliviscusAI/status/2114556298167505225) - **Author:** [@oliviscusAI](https://x.com/oliviscusAI) ("Oliver Prompts") - **Posted:** 2026-10-25, 10:14 PM CT (2026-10-26 03:14 UTC), one video, **33 seconds, silent**. - **Numbers read from the X API on the evening of 2026-10-29 (CT):** 687,849 impressions, 217,031 video views, 6,160 likes, 697 reposts, 7,524 bookmarks, 74 replies, 78 quotes. > **Transcript source:** none needed. The video has **no speech**; faster-whisper `small.en` on the audio returned nothing usable. The outline is built from on-screen text read from frames I extracted with ffmpeg; the screenshots are from the video at the times in their file names. ## Post text (verbatim) > NVIDIA just put 2,575 simulation-ready 3D assets on Hugging Face 🤯 > > A 472GB collection covering everyday objects, tools and industrial equipment. > > It's called NVIDIA SimReady Assets. The models include physics settings and semantic labels for use in simulated environments. > > - Train robots to pick up and move objects > - Build virtual factories and warehouses > - Generate labelled training images > - Populate 3D scenes with ready-made objects > > Useful for robotics research, digital twins and anyone building detailed virtual environments. > > Download the collection below 👇 ## One-paragraph gist NVIDIA has published a free library of about 2,500+ OpenUSD 3D assets, tagged with physics and meaning, on Hugging Face as `nvidia/simready-assets`. The post pitches it for robot training, virtual factories, synthetic labelled images and scene building. The video is a polished silent promo over the dataset page that shows the asset count rising, the categories, one asset's physics runtimes and a plain-English search tool. ## Timestamped outline (video, 0:00 to 0:33, silent) The whole video is laid over a cropped Hugging Face dataset header (`Datasets: nvidia/simready-assets`, 209 likes, "Follow NVIDIA 72.6k", Modalities: Image, Size: <1K) with animated 3D asset cards on top. ### 0:00 A single asset appears - [0:00] A light grid with one isometric orange pallet-stacker/forklift model, file label `vehicle_forklift_orange_a01`. ![[images/NVIDIA-SimReady-Assets-t0000.jpg]] ### 0:06 The headline count - [0:06] Big counter settles on **"2,549 assets."** with the tag line **"OpenUSD · PhysX · MuJoCo · Newton"** (the file format and the three physics engines). The counter is still climbing at 0:05 and the number fades after 0:07; the settled value is 2,549, not the 2,575 in the post text. ![[images/NVIDIA-SimReady-Assets-t0006.jpg]] ### 0:11 and 0:13 Categories - [0:11] Category tile **Automotive, 373 assets**. ![[images/NVIDIA-SimReady-Assets-t0011.jpg]] - [0:13] **Medical & Healthcare, 118 assets**. (Other tiles in the sequence include Tools, 234 assets, and Warehouse & Robotics, 269 assets, seen in neighbouring frames.) ![[images/NVIDIA-SimReady-Assets-t0013.jpg]] ### 0:17 One asset in detail - [0:17] Card for a **Cordless Drill**: file `powerTool_drill_a01.usd`, category powerTool, size 0.23 x 0.06 x 0.23 m, 231K triangles, 12 PBR materials, physics "PhysX · MuJoCo · Newton". Behind it faded tiles read "Tools, 234 assets". ![[images/NVIDIA-SimReady-Assets-t0017.jpg]] ### 0:22 Tag line - [0:22] Text: **"Simulation-ready, from racks to robots."** ![[images/NVIDIA-SimReady-Assets-t0022.jpg]] ### 0:26 Plain-English search - [0:26] Caption "Search in plain English." A screenshot of the Space `hf.co/spaces/nvidia/simready-central` with a "Search using natural language..." box in "Hybrid" mode and "50 of 3,599 results" (a different number again; the Space's index is not the same count as the dataset's asset folders). Thumbnails: robot arm (`standardbots_thor.usd`, 45.19 MB), kitchen stove (46.32 MB), surgical robot (595.12 MB). ![[images/NVIDIA-SimReady-Assets-t0026.jpg]] ### 0:31 Example search - [0:31] The search box reads "something to carry heavy boxes around a warehouse" and returns palletized loads, loaded pallets, a storage crate and palletized boxes, each with a match percentage (for example 87%). ![[images/NVIDIA-SimReady-Assets-t0031.jpg]] ## Claims check | Claim | Verdict | Source or note | | --- | --- | --- | | 2,575 simulation-ready assets | **Supported** | A third-party audit (2026-10-26) listed the repository root and got 2,575 asset directories plus one `.gitattributes` file; NewsNix independently counted 2,575. HF's own SimReady Central blog says "more than 2,000" | | The video's "2,549 assets." | **Does not match the post** | Differs by 26 from 2,575. I did not find why (older snapshot, a different counting rule or a promo-only figure are all possible; I did not verify any of them) | | A 472GB collection | **Supported at the time, now stale** | The audit says 472 GB was the Hugging Face page's displayed total on 2026-10-26 (it did not re-sum the bytes). The page text I fetched on 2026-10-29 says "Total file size 490 GB", so the figure has since changed | | "Physics settings and semantic labels" included | **Supported, with a limit** | The audit found mass, centre of mass, inertia, friction and restitution fields and semantic labels in the one package it opened (`box_corrugated_brown_b08`: 43 files, about 58.3 MB). It says field presence is not proof the values are real-world measured, and it ran no physics engine | | Covers everyday objects, tools, industrial equipment | **Consistent** | The video shows Automotive, Tools, Medical and Warehouse and Robotics categories; HF's blog lists robot arms, grippers, racks and cooling equipment | | Use cases: train robots, virtual factories, labelled images, 3D scenes | **Marketing-level, plausible** | NVIDIA's blog describes "physical AI training environments and AI factory digital twins". No audit ran a robot-training or synthetic-image test | | "Download the collection" | **Link missing from the post text** | The dataset is at huggingface.co/datasets/nvidia/simready-assets; the audit says you do not need all 472 GB, you can pin a revision and download single asset packages | | NVIDIA published it free | **Supported** | The HF blog calls SimReady Central "free, ungated". NewsNix reports a CC BY 4.0 licence file in each asset folder while the HF page's license field is blank; I did not read a licence file myself | | Metrics: 687,849 impressions, 217,031 video views | **Supported (as read)** | X API public metrics, 2026-10-29 evening CT | Sources: [Hugging Face dataset page](https://huggingface.co/datasets/nvidia/simready-assets); [SimReady Central announcement](https://huggingface.co/blog/nvidia/simready-central-assets-robotics-ai-factories); [mushroom.cv audit, 2026-10-26](https://blog.mushroom.cv/blog/nvidia-simready-assets-audit/); [NewsNix licence and download report](https://newsnix.com/articles/nvidia-simready-2575-assets-cc-by-licence-invisible/); [SimReady Central docs](https://docs.omniverse.nvidia.com/simready/latest/simready-central.html). ## Things mentioned - NVIDIA SimReady Assets (`nvidia/simready-assets`) and SimReady Central, on Hugging Face - OpenUSD (the 3D scene format), PhysX, MuJoCo and Newton (physics engines) - Robotics research, digital twins, virtual factories and warehouses, synthetic labelled training images - The Space `nvidia/simready-central` (natural-language search, "Hybrid" mode) - Related repo: [NVIDIA/simready-foundation](https://github.com/NVIDIA/simready-foundation) (the SimReady specification) ## Tutorial section: using the assets (stated vs. inferred) | Step | Stated in the post or video | Added context (labelled) | | --- | --- | --- | | 1 | "Download the collection below" (link not in the text) | Go to the dataset page on Hugging Face and pick assets rather than pulling everything; the audit says you can pin a revision and download one package | | 2 | The video shows searching in plain English in the SimReady Central Space | Shown in the video: type a description such as "something to carry heavy boxes around a warehouse" | | 3 | The video shows asset cards with USD file names and physics runtimes | Inferred: open the `.usd` in a USD-capable tool (for example NVIDIA Omniverse or Isaac Sim) and choose a PhysX, MuJoCo or Newton variant (inferred from the audit's description) | | 4 | Not stated | Before relying on the physics values, test a small set; the audit says it did not run any simulator, and per-asset capability varies | | 5 | Not stated | Check the licence yourself: HF shows the licence field blank, while NewsNix reports CC BY 4.0 text inside each asset folder | ## Biographies - **Oliver Prompts (@oliviscusAI):** X account created 2022-05-18; bio "Helping you understand AI. Software Engineer. Opinions are personal. DM for collabs." About 22,989 followers, 1,635 posts, 786 media posts at the time of reading. - **NVIDIA:** the chip and AI company that published the dataset (general knowledge). On Hugging Face the video shows "Follow NVIDIA 72.6k". - **Hugging Face:** the AI model and dataset hosting platform that hosts it (general knowledge). ## Tag note - Used: video-summary, ai. - Proposed (not in the approved list): **robotics**, **simulation**.