At the DRC Finals (also known as the DARPA Robotics Challenge), Team IHMC unveiled their impressive ATLAS robot and took home second place. And if you’re wondering what has been keeping the ATLAS robot busy these days, check out this video below. As it turns out, this big guy has been learning how to do some chores.
That’s right, ATLAS—a multi-million dollar robot—has been learning how to clean your house.
Unfortunately, it’s not because the world will soon be introduced to a robotic butler. Rather, it’s because ATLAS needs to be run often to ensure that code updates don’t break anything and to ensure that everything continues to run smoothly.
In this case, running the same DRC related tasks seem to be far too tedious for the robot, which prompted IHMC to come up with other activities for ATLAS to do.
To be able to complete the chores, ATLAS barely had to change control systems or the user interface (UI). Apart from minor bug fixes, the only major addition that would allow the robot to operate effectively while doing chores was to add an option to have its hands in front of it as it walks, making it more efficient for sweeping.
The entire process is still not fully autonomous—but it can be controlled with a little human supervision.
“Most of the stuff in this video is controlled by me, but in a co-active way. I’m not simply sitting there with a joystick teleoperating the robot: I tell the robot through the UI that I want to grab a bottle off the table by clicking the bottle and making sure that the resulting hand is in the correct place. Then, the robot tells me how it’s going to move its entire body to reach that location, through a preview in the UI. If I’m okay with the plan the robot has come up with, I tell it to execute that motion. In the future, I can see a lot of what was done in this video moving more to the autonomous side, but I always see there being a human in the loop,” John Carff, ATLAS’ robot operator at IHMC explains.
So far, ATLAS is able to turn on the vacuum cleaner, pick up things off the floor, lift, stack, tidy up and sweep.
While impressive, this still leaves a lot of room for improvement, particularly for its control algorithms. To help with this, ATLAS will need to perform various tasks continuously.