Search papers, labs, and topics across Lattice.
2
0
5
A quadruped robot can now reliably grasp objects in cluttered scenes from partial observations, thanks to a new pipeline that combines language grounding, point cloud completion, and safety-oriented grasp selection.
Surprisingly, fine-tuning a tiny 0.5B parameter language model achieves robust role classification in human-robot interaction, rivaling much larger models in zero-shot settings.
