Language-Guided Representation Learning for Cross-Sensor Material Recognition

A new arXiv paper proposes using language guidance to learn material representations that transfer across different vision-based tactile sensors. Because each sensor produces distinct readings of the same material, the approach aims to close that gap by grounding learned features in shared language descriptions. The work targets robotic touch perception, where properties such as softness and texture are difficult to recover from vision alone.