Search papers, labs, and topics across Lattice.
Affiliation:
3
0
3
Frozen multimodal foundation models can achieve near-perfect fine robotic manipulation by simply augmenting their spatial perception with dynamically rendered virtual views.
Even top-performing VLA models degrade sharply the moment instructions rely on implicit linguistic or spatial context rather than explicit object naming鈥攁nd upgrading to stronger external VLMs fails to close the gap.
