This is a great paper from the Rossum Project. I could not find it referenced here on LMR, so I'm submitting it.
An important challenge in small-scale robotics is finding a robot's position when only limited sensor information is available. There are many technologies available for robot localization,
including GPS, active/passive beacons, odometry (dead reckoning), sonar, etc. In each approach, however, improvements in accuracy come at the cost of expensive hardware and additional processing power. For the robotics enthusiast, the key to successful localization is getting the best results out of cheap and widely available sensors.
This paper presents a method for localization and map construction of a mobile robot using data from a sonar-based range sensor. No prior knowledge of the environment is assumed. The map is constructed autonomously by the robot.
Supporting Docs: http://sbp.ri.cmu.edu/papers/sbp_papers/integrated3/brown_loc_wo_land.pdf