We show how a neural network can be used to allow a mobile robot to de
rive an accurate estimate of its location from noisy sonar sensors and
noisy motion information. The robot's model of its location is in the
form of a probability distribution across a grid of possible location
s. This distribution is updated using both the motion information and
the predictions of a neural network that maps locations into likelihoo
d distributions across possible sonar readings. By predicting sonar re
adings from locations, rather than vice versa, the robot can handle th
e very nongaussian noise in the sonar sensors. By using the constraint
provided by the noisy motion information, the robot can use previous
readings to improve its estimate of its current location. By treating
the resulting estimates as if they were correct, the robot can learn t
he relationship between location and sonar readings without requiring
an external supervision signal that specifies the actual location of t
he robot. It can learn to locate itself in a new environment with almo
st no supervision, and it can maintain its location ability even when
the environment is nonstationary.