1 paper · 1 filter
Pengxuan Yang, Ben Lu, Zhongpu Xia +7
Latent World Models enhance scene representation through temporal self-supervised learning, presenting a perception annotation-free paradigm for end-to-end autonomous driving. Howe…