Search papers, labs, and topics across Lattice.
Beijing University of Posts and Telecommunications
2
0
4
Uniformly quantizing the entire diffusion action head of VLAs to W4A4 is not only possible, but can match or exceed FP16 performance, defying conventional wisdom and slashing memory footprint by 71%.
Stop guessing about action spaces for robot manipulation: a massive empirical study reveals that predicting delta actions boosts performance, while joint vs. task space offers a stability vs. generalization tradeoff.
