Every few months a new robot hand goes viral for how human it looks. Then the demo ends, and almost every robot doing real work goes back to using a claw.
A gripper is two metal fingers that close, and that is what you will find on most factory floors today. It is a long way from a real hand.
Here is why the hand is the hard part.
Why do more joints make a hand less reliable?
A gripper has one moving part. Your hand has about twenty seven places that bend. Every joint you add to a robot is another motor, another wire, and another thing that fails on the morning of a demo.
A warehouse does not need fingers, because what it really wants is a clamp that closes ten million times without breaking.
Why is grip force so hard to get right?
When you pick up an egg, you squeeze just hard enough to hold it without cracking it, and you never have to think about it.
A robot has to work out the right force for every new object, and it usually gets one try.
Why can't a robot feel an object slipping?
This is the biggest problem of the five. A robot can see a cup perfectly and still have no idea that the cup is sliding out of its grip.
Your fingertips notice a slip in a fraction of a second. Most robot hands have almost no sense of touch, and a camera usually cannot see a slip until it is too late.
Why is there not enough hand data?
Robots learn by copying people, so teaching hands requires a huge amount of recorded hand work, with the force as well as the video. That kind of recording barely exists today.
This part is my day job, and it is why I find the hand interesting rather than annoying.
Why doesn't simulation solve it?
Most robot training happens in simulation because it is cheap and fast. Touching, rubbing and slipping are the things simulators get most wrong, and almost everything a hand does involves touching something.
So the hand is the place where practice and reality disagree most.
What is the industry doing about it?
There are two ways out, and the industry is trying both.
The first is to rebuild the world for simple grippers, which is what a warehouse already does. The floors are flat, the boxes are the same, and everything arrives in the same place every time. It works, and it is why robots already earn their keep there.
The second is to build a hand good enough for the world we already have. Think about door handles, zips, jar lids and light switches. All of them were designed for human fingers rather than for a claw.
A company can rebuild a warehouse, but nobody can rebuild every kitchen, and that difference is the reason people keep working on robot hands.
Robot models have become smart very quickly. The harder problem now is handling real objects, like picking up an egg without dropping it or crushing it.