1 paper
Zhenghao Zhang, Yuanxiang Wang, Zhenyu Guan +11
Learning visual world models for planning requires compact latent dynamics that remain sensitive to actions, yet reconstruction-free joint-embedding objectives can collapse to acti…