Search papers, labs, and topics across Lattice.
2
0
1
A massive 2.6M-scene benchmark and a novel generative model finally unlock robust dexterous grasping in cluttered environments, outperforming prior methods without post-optimization.
Ditching brittle hand-estimator-plus-SLAM cascades for an end-to-end streaming memory network slashes world-space hand tracking error by 21.4% while doubling inference throughput.
