1 paper
Junchao Huang, Ziyang Ye, Xinting Hu +5
Autoregressive video world models predict future visual observations conditioned on actions. While effective over short horizons, these models often struggle with long-horizon gene…