Clamp

"Everyone thinks the hard part of a robot is the brain," she said, clicking a pincer. "Then they watch it try to pick something up. This is manipulation, and it's brutal." She lowered the claw onto the egg with a fixed grip — splat. "Too firm." She tried the rock with the same gentle setting — it slipped and thudded. "Too soft. And here's the trap: a claw that squeezes with a fixed force is doomed, because the right force is different for every object, and it changes moment to moment."

She held up a plain claw. "A naive gripper is open-loop — you command a position, it closes to there, done. It has no idea what it's holding or whether it's working. To grip well you need a closed loop: the claw senses and adjusts continuously. So I add force feedback — a sensor that reads how hard the fingers are actually pressing. Now the claw closes not to a fixed position but to a target force: a whisper for the egg, a real squeeze for the rock. Same command, right pressure for each."

"How does it know when it's about to drop something?" a builder asked.

"Slip detection," Clamp said. "A tiny sensor feels the object beginning to slide — micro-vibrations, a shift — before it's gone. The moment it feels a slip, the loop firms up just enough to catch it, then eases back. And when it feels the object straining — starting to deform — it eases off. That's the whole art: firm on a slip, gentle on a strain, all in a fast feedback loop, dozens of times a second."

She picked up a soft foam ball without denting it. "There's one more trick: compliance. A rigid claw fights every bump; a compliant gripper — one with a little built-in give, like a spring or a soft pad — automatically conforms to odd shapes and forgives small errors. Softness, engineered on purpose, grips better than raw stiffness."

"So the right grip depends on the thing," the builder said slowly, "not on how hard I can squeeze."