A recent IEEE Spectrum feature, "Robots Are Learning to Feel," looks at the growing role of tactile data in robotics. Vision and language have driven much of the recent progress in robot learning. Touch is the sense that lets a machine tell how firmly it is gripping an object, whether the object is slipping, and what the surface is made of.
Touch is difficult for robots. A camera gives a rich view of a scene, but it cannot show the forces acting at the point of contact. Tactile sensing supplies that information, which matters for delicate, contact-heavy work such as handling fragile items, inserting parts, or manipulating objects that are hidden or deformable.
The article's title points to a broader push to treat touch as a data source for training robots, much as images and text are used today. Collecting and using that data at scale is an open challenge, and the topic ties into the wider push toward physical AI, where models learn from real-world interaction rather than from screens alone.
This summary is based only on the article's headline and link. See the original IEEE Spectrum piece for the specific technologies, research groups, and results it covers.