We often take for granted how easy it is for us to complete basic tasks just using our hands.
Typing out an email or setting up a pot of coffee are things we can do bleary eyed in the morning, but the movements our hands have to make to complete such tasks are truly complex, human skills.
That’s why for robots to advance to the point where they can take care of our elderly or help clean our homes, it will be necessary for their hands to demonstrate the same level of finesse ours do.
Researchers at the University of Washington are inching closer to making that a reality. Their software system, dubbed the Adroit manipulation system, is powering a robot hand that is considered one of the most highly capable in the world. The highly dexterous, five-fingered robot hand is not only deft in its execution, but can learn from its own mistakes to perform better the next day.
The system runs on a form of artificial intelligence called machine learning, which is when machines (in this case the robot hand) learn to improve over time.
Lead author Vikash Kumar, a UW doctoral student in computer science and engineering, compared the hand’s learning process to a child going to kindergarten.
“You go to a class, you are told this is something you will be learning today, and try a lot of things,” he told Tech Insider. “The next day you come up with a better understanding. So it’s a lot of trial and error.”
The hand can move all of its joints, one at a time from end to end, in just 2.44 seconds.
It also demonstrates reflexive behaviour — meaning the robot hand will respond if you touch it the same way a human would if you tapped them. It can also perform highly complex skills for robot hands, like spinning a tube of coffee beans without dropping it.
The hand is still in its early days. The Adroit platform costs a whopping $300,000, making it far too expensive for commercialization. And even if price weren’t a concern, the hand isn’t close to production ready.
“We have learned how to make words right now, but the eventual goal is to make real sentences,” Kumar said.
For example, the hand could endow a robot with the ability to make a pot of coffee. But right now the hand can only demonstrate the steps necessary to do that one at a time — it can’t string them all together to actually make the coffee yet.
Kumar said he could see the hand being used to create a robot butler or caregiver for the elderly in the future.
“What we have achieved in this moment is fundamental and basic,” he said. “It’s a small step in the right direction, but doesn’t take us all the way.”