In 2001, after many years of development, the Electrolux Trilobite, a vacuum cleaning robot, became the first commercial home robot. It had a simple control system to do obstacle avoidance, and some navigation. A year later, iRobot introduced Roomba, which was a tenth the price of the Trilobite and, with only 512 bytes of RAM, ran a behavior based controller. The most intelligent thing it did was to avoid falling down stairs. Since then, sixteen million Roombas have been deployed all over the world and several other competing brands now exist.
As the processing power and RAM capacity of low cost embedded processors improved from its dismal state in the year 2000, the AI capabilities of these robots also improved dramatically. Simple navigation, self-charging, and actions for dealing with full dust bins were added, followed by ability to deal with electrical cords and rug tassels, enabled by a combination of mechanical improvements and sensor based perception. More recently, the addition of full VSLAM (Visual Simultaneous Location and Mapping)— an AI technology that had been around for twenty years—has enabled the robots to build a complete 3D world model of a house as they clean, and become more efficient in their cleaning coverage.
Early expectations that many new applications would be found for home robots have not materialized. Robot vacuum cleaners are restricted to localized flat areas, while real homes have lots of single steps, and often staircases; there has been very little research on robot mobility inside real homes. Hardware platforms remain challenging to build, and there are few applications that people want enough to buy. Perceptual algorithms for functions such as image labeling, and 3D object recognition, while common at AI conferences, are still only a few years into development as products.