Jul 2026
KAM-WM: Kinematic Affordance Maps from Latent World Models for Robot Manipulation
KAM-WM is presented, a framework that extracts a coarse directional interaction cue from a frozen latent video world model without rollout or world-model fine-tuning and indicates that a frozen video model can provide a useful first-order visual prior for control without the test-time cost of future rollout.
Xinyu Shao, Keru Zhou, Guowei Huang et al.
· arXiv.org · 1 citation