Nvidia Puts a 4B World Model on the Robot [Weekly Physical AI Roundup]
![Nvidia Puts a 4B World Model on the Robot [Weekly Physical AI Roundup]](https://wesearch.press/img-proxy?url=https%3A%2F%2Fsubstackcdn.com%2Fimage%2Ffetch%2F%24s_!OWwu!%2Cw_1200%2Ch_675%2Cc_fill%2Cf_jpg%2Cq_auto%3Agood%2Cfl_progressive%3Asteep%2Cg_auto%2Fhttps%253A%252F%252Fsubstack-post-media.s3.amazonaws.com%252Fpublic%252Fimages%252Fd1713932-079b-4406-a185-4a4bbcef54e2_2048x1152.png&w=800)
NVIDIA Puts a 4B World Model on the Robot World models are usually too heavy for the control loop and run on a server. Cosmos 3 Edge is small enough to run on-device.Jay ChiaJul 23, 2026ShareCosmos 3 Edge, running on a JetsonCosmos 3 Edge is a 4B world foundation model built as a Mixture-of-Transformers, pairing an autoregressive reasoner with a diffusion generator that produces the action chunk. NVIDIA reports it running at roughly 15 Hz on a Jetson Thor at 640×360 and 32 actions per inference, generalizing across humanoids, grippers, and driving.
- ▪NVIDIA Puts a 4B World Model on the Robot World models are usually too heavy for the control loop and run on a server.
- ▪Cosmos 3 Edge is small enough to run on-device.Jay ChiaJul 23, 2026ShareCosmos 3 Edge, running on a JetsonCosmos 3 Edge is a 4B world foundation model built as a Mixture-of-Transformers, pairing an autoregressive reasoner with a diffusion g
- ▪NVIDIA reports it running at roughly 15 Hz on a Jetson Thor at 640×360 and 32 actions per inference, generalizing across humanoids, grippers, and driving.
Hacker News (AI / LLM) files mainly under ai. We currently carry 2,091 of its stories.
Opening excerpt (first ~120 words) tap to expand
NVIDIA Puts a 4B World Model on the Robot World models are usually too heavy for the control loop and run on a server. Cosmos 3 Edge is small enough to run on-device.Jay ChiaJul 23, 2026ShareCosmos 3 Edge, running on a JetsonCosmos 3 Edge is a 4B world foundation model built as a Mixture-of-Transformers, pairing an autoregressive reasoner with a diffusion generator that produces the action chunk. NVIDIA reports it running at roughly 15 Hz on a Jetson Thor at 640×360 and 32 actions per inference, generalizing across humanoids, grippers, and driving. The release includes open weights, a Cosmos-3-Edge-Policy post-trained on DROID for pick-and-place, and the post-training scripts to adapt it to your own setup.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at Hacker News - Newest: ""AI" "LLM"".
Discussion
0 commentsMore from Hacker News - Newest: ""AI" "LLM""
