Science

Robot Learns to Sort and Organize After Watching a Human Do It Only Once

by Danny Paez

Having a robotic butler hand you a steaming cup of coffee and the newspaper in the morning is something science fiction has made us yearn for and modern robotics has brought into the realm of possibility. Yet roboticists are still having trouble teaching machines how to complete tasks that even children are capable of. That’s why two researchers at the University of California, Berkeley have begun teaching a robot as if it were a five-year-old in the hopes of turning them into the taskmaster robots of the silver screen.

“We’re teaching robots how to pull off sorting and organizational tasks by simply watching a human do them once,” Tianhe Yu co-author of the study tells Inverse. “Today’s robots are able to perform a few specific tasks well, but they still don’t come close to what a human is capable of. We hope that by teaching robots through demonstration we can enable them to carry out more general tasks.”

PR2 learns to place the peach into the red bowl after watching Yu do so.

Berkeley Artificial Intelligence Research / Tianhe Yu / Chelsea Finn

In a paper published online, Yu and his colleague Ph.D. student Chelsea Finn trained a commercially-available robot, named PR2, by showing, not “telling” or giving it explicit instruction. They fed a neural network a single video clip of Yu doing simple tasks, like putting a peach into a red bowl, and PR2 was able to imitate those actions after just one demonstration.

Previous research was able to teach robots through demonstration, but they all required another robot to demonstrate the task instead of a human. That’s because human limbs don’t move like robotic arms making it difficult for A.I. to track and imitate the motions of an arm.

Learning to pick up a novel object and place it into a previously unseen bowl.

Berkeley Artificial Intelligence Research / Tianhe Yu / Chelsea Finn

That’s why Finn and Yu made their system focus on the object being moved, instead of the arm doing the moving. This way PR2 could just focus on getting the peach where it needed to be, without worrying about exactly how Yu did it himself.

“Companies like FedEx or UPS could use this technology to teach their robots how to pick up and place boxes in specific places,” he explains. “We managed to get it to learn what to do with just one demonstration, being able to show a robot where a package is supposed to go instead of having to hardcode or control it would be extremely helpful.”

Berkeley Artificial Intelligence Research / Tianhe Yu / Chelsea Finn

Moving forward, Finn and Yu want to look past sorting fruit and get their robot playing sports and using tools, they stated in a blog post. Building off of the success they had with simpler tasks, this could allow PR2 and its successors to become expert builders, cleaners, and maybe even baseball players.

If the next phase of this research works out, a team of contractors in the future could be one human ordering around a bunch of bots. Or better yet, there could be an entirely robotic baseball league for our entrainment.

Related Tags