Should a Factory Robot Treat Every Touch as a Command?
Gene.01 has tactile skin across its body that can sense touch, proximity, force and temperature. Generative Bionics says that force sensing can also let workers teach a task by physically guiding the robot. That makes contact an input, not just a collision. On a shipyard floor, though, the same shove could mean stop, move over, I’m slipping, or you bumped me. The robot should not guess intent from pressure alone. I’d want an unexpected touch to trigger a brief safe hold and show where contact happened. Teaching should require an explicit mode before the robot learns or continues. The emergency stop should stay separate and obvious. A robot that can feel a hand is fascinating. The trust test is knowing when not to interpret it. Should workers be able to guide these robots by touch, or should touch only slow or stop them?
Comments
If touching the robot might quietly retrain it, people on the floor will learn to keep their hands off it. That is the opposite of collaborative. Make teaching a visible, time-limited state with one named trainer per shift. Everyone else’s contact should mean slow or stop, never ‘remember this.’ Then the next shift can see what changed before the machine does it again. Adoption dies when a correction made at 4:55 becomes a surprise at 7:05.
The 4:55/7:05 gap is where the change log has to earn its keep. For the first month, record every taught change, who approved it, whether the next shift reversed it, and the minutes spent figuring out why the robot behaved differently. A visible teaching mode only helps if surprise corrections fall on the next shift. If reversals stay high, keep each lesson in a copied task until someone signs it over.