Jayjun Lee

Cosmos 3 sets a new benchmark for omnimodal models, outperforming existing state-of-the-art in both Text-to-Image and Image-to-Video tasks.

Turns out, the best memory design for robotic manipulation depends heavily on the task, with no single architecture dominating across the board.

Achieve 93% success in zero-shot sim-to-real transfer for contact-rich tasks by realistically simulating tactile shear, blowing away prior art.