Nvidia announced Cosmos 3 Edge on July 16, and this is the first real world model built specifically for robots to see and navigate physical space.
The model takes what a robot camera sees and builds a mental simulation of what comes next. It understands physics. It understands occlusion. It understands how objects move and interact. That is harder than you think when you are trying to run it on the edge, which means the robot itself, not in a data center.
Nvidia is pushing this in Japan specifically because Japan has the density of industrial robots and the manufacturing infrastructure to actually use it. Fujitsu, Hitachi, and Kawasaki are joining the coalition. This is not vaporware. These are companies with factories and actual work that needs doing.
The physics piece matters. You can train a model on internet video and it learns correlation. Train a model that has to actually work on a robot arm, and it has to learn causation. Cosmos 3 Edge is trained on that harder problem.
The infrastructure bet from Nvidia used to be about gaming and data centers. Now it is about machines that have to touch the world.
——
Follow: @Ali Demi
Book your free AI clarity call, NOW!
https://buff.ly/TpWy277
——
Repost this. Thanks.

