What's this about?
Jim Fan argued that the sense of touch is criminally under-explored in robotics, with an analogy that landed: imagine doing sleight of hand wearing thick oven mitts — that is how a robot experiences the world. A magnetic piece snapping into place, a paper cup peeling off a stack, a USB connector finding its way into a port are all invisible to a camera.
What exactly is a "modality"?
A modality is a channel of sensory input — vision, sound, touch. Robotics has poured effort into vision because cameras are cheap and image models are excellent. Touch has lagged because the sensors are harder and there is no internet-scale dataset of tactile experience. Fan's point is that the tasks people actually want robots to do are gated on precisely the signal that is missing.
What's the community saying?
861 likes
- Agreement on the diagnosis, with the examples doing the work — each is a task humans perform without looking.
- Fan's own prescription is that learning to feel must be a full-stack co-designed effort spanning sensor, hardware and model, not a module added later.
- hardmaru's adjacent comment this week — that new real-world locomotion behaviour resembles the novel policies once discovered only in simulation, except that these work outside it.