Search papers, labs, and topics across Lattice.
1
0
3
Egocentric human video can outperform teleoperated real-robot data for pretraining embodied models, achieving 24% lower validation loss and dramatically higher task success rates.
