Roomba robotic vacuum cleaners have been great for YouTube videos of dogs and babies going for rides, but they have been, well, kind of random in the way they move around. As Matt noted last year, Dyson had built one that was different, with a vision system that could see where it was going, but it's still not on the market. Meanwhile iRobot, the maker of the Roomba, has introduced a new model that does much of what the Dyson promised: It has "visual simultaneous localization and mapping" (vSLAM) that uses a camera and sensors to map your home to ensure that it gets into every nook and cranny. According to Ars Technica,
Using its updated software, iAdapt 2.0, and two new sensors, the 980 will actually learn the floor plan of your home with all your furniture in it. It uses a vision localization sensor and a low-res camera to search for "landmarks" where it cleans — think a chair or a coffee table — and then recognizes its shape and memorizes it. Combine that knowledge with a tracking sensors, which help the machine know how fast its moving in your home's space, and the Roomba has the power to map every space in your home that it touches.
The video is mesmerizing, watching the Roomba go back and forth, around every chair leg, until you see every inch of the apartment covered. The videos with puppies and babies won't be nearly as interesting as it isn't random at all; it now moves in neat parallel lines, although they don't say what happens if a baby is sitting on the camera.
The new unit also detects the material it's traveling over, cranking up the suction when on carpet. If it runs out of juice, it rolls over to its charging station and sucks up a few watts until it's ready to go at it again.
The new Roomba is also now WiFi-connected and controlled by an app separate from the unit itself, letting you start it remotely and see all the jobs it has completed. Oh, and you can name your robot and celebrate its birthday.
Interestingly, according to Quartz, the Roomba doesn't store the maps, but wipes it clean at the end of the session and starts over from scratch. This makes some sense, given that the furniture and other obstructions get moved and it would have to be pretty sophisticated to know the difference between what's fixed and what's not; it's probably easier just to rebuild the map each time. I suppose like most of us, it knows where it has been, it can see what's right in front of it, but in the longer term really has no idea where it is going.