Search papers, labs, and topics across Lattice.
2
0
3
0
Geometry-Change VLA (GC-VLA), which learns to predict multiview future-current geometry-change tokens from current observations and applies Geometry-Conditioned Residual Flow (GCRF), using a binary intervention router and a single bounded residual velocity policy learned from closed-loop feedback.
Fine-tuning VLMs for compositionality frequently degrades their grasp of basic attributes and relations below vanilla CLIP levels, a failure mode corrected here by mining maximally contradictory negatives directly from scene graphs.
