1 paper · 1 filter
Liheng Ma, Rui Heng Yang, Amin Abyaneh +8
World Action Models (WAMs) build on pretrained video models, whose representations are grounded in physical dynamics and provide a natural basis for action prediction. Despite this…