Robotic learning has quickly become one of the most vibrant categories of automation – and understandably so. Programming a robot has traditionally required a lot of technical knowledge, but what if there was an easier way for non-programmers/robotics to teach these systems to do what we want?
Imitation and reinforcement learning are currently two of the most popular methods. The first involves taking over the robot to teach it to perform a task, while the second involves training a system on millions of images.
A number of researchers are investigating an even more intuitive method that effectively trains a system by watching a person complete a task. A team from Carnegie Mellon University demonstrates in-the-Wild Human Imitating Robot Learning, or WHIRL, an algorithm that can train a system by watching a video.
In their demos, a ready-to-use mobile robotic arm teaches 20+ household chores, including opening and closing drawers and appliances, and taking out the trash.
“Imitation is a great way to learn,” said PhD student Shikhar Bahl of the Robotics Institute in a press release. “It remains an unresolved problem in the field to actually get robots to learn from looking directly at people, but this work takes an important step towards enabling that ability.”
It’s easy to see how a feature like this can be especially useful in a home environment, where roboticists expect these systems to one day be deployed to help elderly homeowners and other people with limited movement.
In the case of WHIRL, no special add-ons are required. The robot just tries to perform a certain task until it is successful, even if it takes several times to fully master it. As CMU points out, its own approach may not be identical to humans – instead, the system looks for the best method to complete the task based on its own hardware limitations.
Right now the system is being trained by watching videos and the team is looking to expand things with clips from services like YouTube.