An ICP inspired inverse sensor model with unknown data association

This paper introduces an Iterative Closest Point (ICP) inspired inverse sensor model for robot localisation given multiple simultaneous observations of aliased landmarks. Combined with a Kalman filter, the sensor model offers a robust alternative to maximum likelihood data association, or a computationally inexpensive alternative to a particle filter. The technique can also be used as a means for re-localising a kidnapped robot, or a sensor resetting method for a particle filter. In the RoboCup Standard Platform League, this sensor model is able to localise the robot from a single observation in 42% of field positions where multiple landmarks are visible.