Secrets of the Most Popular Robot in America
A significant advancement in artificial intelligence took place this month — involving dog poo.
As icky as that sounds, it’s fur-real: On September 9, 2021, Colin Angle, the founder and CEO of iRobot, the company best known for its line of popular Roomba robotic home vacuum cleaners, Tweeted:
“Yes, it’s true: the new Roomba j7/j7+ detects and avoids pet waste. Powered by the intelligence of iRobot Genius 3.0 [software], this robot vacuum harnesses advanced AI capabilities to work smarter in the home. And importantly, no more poop-tastrohpes! Here’s the scoop on poop,” along with a video illustrating how his company managed the feat.
A few weeks before that, Angle came on the No Parking Podcast hosted by Bryan Salesky, CEO Argo AI, and Ground Truth columnist, Alex Roy, to discuss iRobot’s past, present and future. Angle touched upon the difficulties of developing a robotic vacuum that knows what to suck up and what to avoid within the millions of unique home layouts in which it’s deployed.
Listen to the full episode here:
The Roomba “pet waste” problem, to put it as delicately as Angle does, has been dogging the company for years — perhaps since shortly after the earliest Roomba model was released on the consumer market almost two decades ago, in 2002.
In summary, the issue boils down to the fact that the robotic vacuum is designed to suck up solids like crumbs and dirt and debris, but not so much with semi-solids or mushy substances like, say a dog accident. When such accidents happened previously, a well-intentioned Roomba might try to vacuum up the offending waste product and instead track it and mash it around a home’s floors.
One traumatized puppy owner described the scene in a viral 2016 Facebook post: “Do not, under any circumstances, let your Roomba run over dog poop. If the unthinkable does happen, and your Roomba runs over dog poop, stop it immediately and do not let it continue the cleaning cycle. Because if that happens, it will spread the dog poop over every conceivable surface within its reach, resulting in a home that closely resembles a Jackson Pollock poop painting.”
So how is the new version of the Roomba able to avoid this fate? As Angle explains in iRobot’s new video, “poop comes in many shapes and sizes. We built over a hundred physical models of poop,” and trained the Roomba’s artificial intelligence software algorithms with them.
iRobot’s engineers did this by showing the Roomba software system “over 100,000 images” of different varieties of poop and then tagging them as such, similar to the machine-learning process which trains autonomous vehicle systems using “labeled” objects seen on the road. This way, the Roomba can learn the visual characteristics of various feces and compare them to the obstacles it encounters when vacuuming. “At the end of the day, it worked. And the Roomba j7 comes fully equipped to both identify and avoid number two,” Angle says.
For more on Angle’s work designing and developing cutting-edge robotic vacuums, listen to the No Parking Podcast now and subscribe.