FloAff-Kitchen: Bridging Navigation and Manipulation via Canonical and Progressive Floor Affordance Learning

Researchers introduced FloAff-Kitchen, a new framework for mobile manipulation robots.

It combines canonical representation learning and progressive affordance prior learning to predict floor affordances from egocentric multimodal inputs.

They also released a cross-scene, multi-view benchmark for evaluating floor-affordance prediction across different environments and viewpoints.

Experiments show the method outperforms strong baselines, helping robots choose floor placements that better support downstream manipulation tasks.