Robots can now feel surfaces to inspect manufacturing quality — and describe what they find in words

A research team built a system that combines computer vision, touch sensors, and language processing so robots can identify material properties like hardness and roughness that cameras alone can't detect. In lab tests, the robot sorted defective parts with 94% accuracy and could describe surface qualities in natural language — a capability that could let manufacturers automate quality control without human inspectors.

Manufacturing quality inspection has always required human touch — literally, someone running their hand over a surface to feel if it's rough or cracked or soft in ways cameras can't see. This work shows that combining tactile sensors with vision and language models lets a robot do that autonomously and describe what it finds. The practical effect: factories can deploy a single robotic system to catch defects that vision-only systems miss, without needing to program specific rules for each material type. The robot learns to describe properties in natural language, which means factory workers can ask it questions about what it detected instead of reading numbers off a display.