Nowadays, nearly all mobile robotic tasks require some knowledge of the robot location in the environment. For example, those tasks involving the robot to reach a specific target require knowledge about the current robot pose in order to plan a path to the goal. Also, exploration tasks require some estimate of the robot pose in order to decide whether a specific region has been already visited by the robot or not. The problem of computing the robot pose is known as the mobile robot localization problem. The mobile robot localization problem appears in many flavors. In some cases, only a qualitative pose estimate is needed. For example, for high level spatial reasoning, the robot may only need to know if a certain area, such as a room, has been previously visited or not.
This kind of localization is commonly named weak localization. In some other cases, quantitative pose estimates with respect to a fixed reference frame are required. For example, to build metric maps, such as occupancy grids, the robot needs accurate numerical estimates of its pose in the space. This approach to localization is usually referred to as strong localization. Both weak and strong localization problems can be defined in a global or in a local context,
constituting the so called global localization and local localization problems respectively. The former refers to the obtention of the robot pose without an a priori estimate of its location. It is called global localization by analogy with global function minimization, whereby an optimum must be found without a reliable initial guess. On the contrary, local localization, sometimes named pose maintenance, refers to a continuous refinement of the robot pose, starting with an initial guess. This chapter focuses on the strong localization problem in the local context. From now on, in the context of this document, the terms localization and mobile robot localization will refer to the strong localization problem in the local context. A common approach to confront this localization problem is the use of exteroceptive sensors, such as range finders or cameras, measuring the external environment. Exteroceptive sensor data is correlated at subsequent robot poses to compute displacement estimates, usually based on initial guesses provided by proprioceptive sensors, such as odometers or inertial units. As a consequence of this, the quality of the pose estimates is strongly related to the quality of the measurements provided by the exteroceptive sensors.