Search papers, labs, and topics across Lattice.
3
0
3
0
A lightweight Q-value model can boost a 9B VLA's performance beyond that of a 27B model while reducing inference latency by 27%.
Organizing visual attention before camera motion can dramatically enhance narrative coherence and viewer engagement in dynamic 3D environments.
LLMs can effectively decompose interactions into phases and roles, but struggle to generate dynamic, realistic motion without a structured approach.
