Robbyant Open-Sources LingBot-Vision Robot Vision Model With 161M Images
A widely shared X post says Ant Group's robotics arm has released open weights, code, and a paper for a robot vision model trained on web-scale image data.
“I'll always root for a team that open-sources its best work, and Robbyant just did it properly.”
KimmonismusTECH#1364In a post on X, @kimmonismus highlighted Robbyant's release of LingBot-Vision, a robot vision foundation model the post says was trained on 161 million images filtered from 2 billion raw web images, without human labels, edge detectors or depth sensors. A follow-up post linked the project's Hugging Face collection and GitHub repository, where weights, code and a paper are listed as publicly available.
“161M images is a big pile, but static images don't teach force, contact, or failure recovery. The scarce asset in embodied AI is teleoperated action data.”
0xascaleRelated Stories
2 linkedCommentary on X
Highest ranked
@kimmonismus Got impressed with the 21m version here. That size runs on the robot itself with no need for API call out for every frame and no cloud in the loop for perception. A robot that needs the cloud to see after every step goes blind the second the connection drops.
