arXiv Robotics — research abstracts· Skand Peri, Hung Nguyen, Chanho Kim, Li Fuxin, Stefan Lee·· 10 days agoEditorial score66
3D Point World Models: Point Completion Enables More Accurate Dynamics Learning
3D Point World Models: Point Completion Enables More Accurate Dynamics Learning
Summary
This research proposes 3D Point World Models (3DPWM), a task-agnostic world model that operates entirely in 3D space by first completing partial point clouds and then learning action-conditioned dynamics. The method enables reliable long-horizon rollouts and supports sim-to-real transfer, as shown through experiments on robotic embodiments and tabletop manipulation benchmarks.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
Source：arXiv Robotics — research abstracts · arxiv.org