Magazine article AI Magazine

The Home-Vacuum Event

Magazine article AI Magazine

The Home-Vacuum Event

Article excerpt

When the original vacuuming contest committee--Pete Bonasso, Erann Gat, and Sebastian Thrun--began devising a contest, the big question in their minds was, "What does AI buy you in this task?" As early as 1993, it was apparent that industrial vacuuming robots were emerging as autonomous and well behaved at least in large industrial areas (Bonasso, Miller, and Kuipers 1993). Devising a sweep pattern on a bounded uncluttered surface to ensure complete coverage is a well-formed and solved problem. AI had nothing to contribute with regard to the basic cleaning task.

A domestic or small office venue offered more complexity. The areas were smaller and contained more furniture. However, all but the first AAAI Mobile Robot Competition had office-navigation tasks, and the top finishers showed that many efficient algorithms existed (for example, Nourbakhsh, Powers, and Birchfield [1995] and Yang et al. [1996]).

After some discussion, organizers agreed that the power of the intelligence in a home-cleanup task lay in (1) knowing how to clean efficiently, (2) knowing when there might be a need to touch up an area because of human intervention, and (3) being as unobtrusive as possible in the presence of humans. Efficient cleaning concerns deal with limited resources, for example, energy, time, and bag capacity, whereas touching up requires checking for messes in areas of recent human activity. Being unobtrusive involves adapting behavior to avoid interfering with both expected and unexpected activities of any humans in the environment.

The Rules

We combined our venue with that of the Find-the-Remote contest, which resulted in a five-room house (figure 1). The Find-the-Remote participants would use the kitchen and the living room, and the Home-Vacuum participants would use the hallway, the bedroom, the den, and the family room.


We designed three phases, the first of which required a one-time cleaning of all the rooms. We hoped phase one would be an easy navigation round where the entrants' robots would show competent navigation capabilities well established in past contests. The only wrinkle concerned bag capacity: if the robot encountered messes in the rooms (one-foot-diameter piles of white confetti), it had to return to a deposit area after it met two such messes. Points were awarded for cleaning the messes (or just moving over them) and making deposits. As in the past, points were deducted for colliding and for engineering the environment to accommodate the robot.

We planned for the defining part of the contest to be phases two and three. These phases were designed to highlight the intelligent aspects of a home-vacuuming task. Phase two, called Tidy Up, was intended to tease out efficiency issues. The robot was to station itself at the disposal area, face down the hall, and watch for humans entering and leaving the rooms. Whenever a human (one of the judges) entered a room, there was the possibility that a mess had been made; the robot had to check the room and, if necessary, clean it up. Again, after every two messes, the robot had to return to the disposal area before continuing. The objective was to obtain the highest score in a 15-minute run. We pushed for intelligence by making the family room twice as profitable as the other rooms and giving the teams the expected frequency of human visits for each room. Thus, the robot had to concern itself with missing a family room opportunity while cleaning a lesser room or emptying the bag. In this phase and the next, discretionary points were awarded for innovative vacuuming mechanisms.

Phase three, Clean My Room, was designed to emphasize unobtrusive activity in the presence of humans. The robot was to start at the disposal station. A human (one of the judges) would come into the hallway and indicate to the robot that a room needed to be cleaned. The robot was to then move to the room, clean any messes there, and return to base to deposit the trash. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed


An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.