Visual world models predict the consequences of actions from images, but planning with them is typically done by optimizers such as CEM and MPPI. According to the abstract, these optimizers stall on long-horizon tasks where progress requires motion.
The new arXiv preprint, 'Sampling-based Kinodynamic Planning with World Models,' addresses this limitation. As the title indicates, the authors propose a sampling-based kinodynamic planning method rather than an optimizer-based one.
The abstract provided is truncated, so details of the method and any experimental results are not available from the source. The paper is listed as a new announcement on arXiv.