Search papers, labs, and topics across Lattice.
7
0
7
0
Achieve state-of-the-art robot manipulation success rates while slashing inference costs by up to 72% with A1, a fully open-source VLA framework that adaptively truncates computation.
Current robot manipulation benchmarks fail to capture the messy reality of real-world deployment, so this work introduces a new benchmark, ManipArena, to close the sim2real gap.
Achieve controllable multi-character animation with arbitrary numbers of characters by preventing identity entanglement and improving identity-pose binding via instance-isolated latent representations and decoupled attention.
Unlock the power of web videos for embodied AI: implicit geometry representations let agents learn to navigate from real-world room tours without relying on fragile 3D reconstruction.
Visuomotor policies can learn to ignore distracting visual variations simply by preprocessing raw RGB images into task-aware, semantic-geometric representations *before* feeding them to the policy.
Forget monolithic action decoders: AtomicVLA's skill-guided mixture-of-experts unlocks significant gains in long-horizon robotic manipulation and continual learning.
Reconstructing realistic 3D hand avatars from messy, real-world video just got a whole lot better thanks to a new method that explicitly models and suppresses visual "noise" like motion blur and object interactions.
