12.2 Localization

Localization is a fundamental problem in robotics. Using its sensors, a mobile robot must determine its location within some map of the environment. There are both passive and active versions of the localization problem:

Passive localization: The robot applies actions, and its position is inferred by computing the nondeterministic or probabilistic I-state. For example, if the Kalman filter is used, then probabilistic I-states are captured by mean and covariance. The mean serves as an estimate of the robot position, and the covariance indicates the amount of uncertainty.
Active localization: A plan must be designed that attempts to reduce the localization uncertainty as much as possible. How should the robot move so that it can figure out its location?
Both versions of localization will be considered in this section.

In many applications, localization is an incremental problem. The initial configuration may be known, and the task is to maintain good estimates as motions occur. A more extreme version is the kidnapped-robot problem, in which a robot initially has no knowledge of its initial configuration. Either case can be modeled by the appropriate initial conditions. The kidnapped-robot problem is more difficult and is assumed by default in this section.

Steven M LaValle 2012-04-20