Search papers, labs, and topics across Lattice.
2
1
5
2
Ditching pixel-space translation unlocks a unified model (LatentUM) that reasons across modalities with SOTA results, opening doors to more efficient and aligned visual AI.
Mantis achieves 96.7% success on the LIBERO benchmark by decoupling visual foresight from the VLA backbone, proving that disentangled prediction boosts performance and reduces the burden on the VLA backbone.
