Roomba cleaned homes by bouncing and sensing dirt instead of mapping the room

Domestic robots had been a vision for decades, and mobile robot vacuums existed for industrial and commercial settings for more than fifteen years before Roomba. The home versions had to be smaller, lighter and far cheaper, which meant the intelligence had to be affordable.

iRobot's answer was reactive, not cognitive. The Roomba uses infrared sensors to avoid falling down stairs, sensors that detect dirty spots needing more attention, and simple random-motion heuristics instead of planning or learning. It is, by the researchers' own account, not capable of planning or learning, yet it cleans three average rooms per charge and returns to its base.

The patent filed in December 2002 covered the cleaning mechanism that made this work: a self-adjusting head with counter-rotating asymmetric brushes plus an independent vacuum, and a side brush that sweeps dirt from outside the robot's envelope into the cleaning path. The design protected the product while the low-cost approach opened a consumer market.