Pixel: a full-stack mobile manipulator, fully open source
Arthur Allshire called Pixel probably the most impressive full stack learning project ever, and the whole thing is open source. Elsewhere, a world model that runs at 50+ FPS on one consumer GPU got praised and picked apart in the same breath.
Shown this week
Pixel open-sources a low-cost mobile manipulator and its learning stack
Jason Liu and Tony Tao released Pixel, a low-cost mobile manipulator designed together with its data collection rig and learning stack, plus a second paper, Pixel-UMI, for whole-body manipulation. Tao says they built it because in December 2025 the robots on offer were either too expensive or not good enough.
probably the most impressive full stack learning project ever, and fully open source
A world model runs at 50+ FPS on one GPU, and Paxton finds the seams
Jimmy Wang released 3D World Simulator, an interactive world model (a learned simulator you can poke at) that he says runs at 50+ FPS on an RTX 5090 and predicts novel views, multi-view RGB-D, 3D geometry and spatially aligned tactile signals.
you can see how incredibly *data limited* this sort of approach is
Out in the world
Tutor rents robots by the hour, with humans in the loop on every deployment
More detail landed on Tutor Intelligence's $75M Series B, and the model is the story: customers pay by the hour or by the pick, with no upfront purchase and no long-term contract. Remote operators Tutor calls tutors give the robots input and guidance, and those interactions become training data.