Search papers, labs, and topics across Lattice.
1
0
3
VLMs get a 24% performance boost and run 56% faster on robot manipulation tasks by explicitly modeling action advantages and exploring multiple future paths, instead of relying on noisy foresight predictions.
