This paper investigates the suitability of iRobot's Roomba as a low-cost robotic platform for use in AI research and education. Examining the sensing and actuation capabilities of the vacuum base led us to develop sensor and actuation models more accurate than those provided by the raw API. We validate these models with implementations of Monte Carlo Localization and FastSLAM, algorithms that suggest the Roomba's viability for AI research. Classroom trials incorporated the Roomba into CS 1 and CS 2 courses in the spring of 2006, and student feedback has been similarly promising for educational uses. While the platform has both benefits and drawbacks relative to similarly-priced alternatives, we conclude that the Roomba will interest many educators, especially those focusing on the computational facets of robotics or applications involving large, homogeneous groups of physical agents.