Search papers, labs, and topics across Lattice.
5
0
7
7
Gaussian Splatting can now handle complex articulated objects with multiple movable parts, thanks to a new geometry-motion alternating refinement strategy that leverages 2D segmentation priors.
LVLM hallucinations can be slashed without retraining, simply by adaptively integrating contextual information during inference.
By learning visual representations from scene-level semantics down to pixel-level details, C2FMAE overcomes the limitations of both contrastive learning and masked image modeling.
Robots can now learn manipulation affordances from human demos with better contact pose accuracy, thanks to a new pipeline that automatically infers 3D pose-centered labels.
Forget manually annotating robot skills: Uni-Skill automatically evolves a skill library from unstructured video, enabling zero-shot generalization to novel manipulation tasks.
