Page 259 - Introduction to Autonomous Mobile Robots

244

Figure 5.34 Chapter 5
Kalman filter estimation of the new robot position: By fusing the prediction of robot position (thin)
with the innovation gained by the measurements (thick) we get the updated estimate p ˆ k k( ) of the
robot position (very thick).

to enable finding the best matches while eliminating all other remaining observed and pre-
dicted unmatched features.

5. Estimation. Applying the Kalman filter results in a final pose estimate corresponding
to the weighted sum of (figure 5.34)
• the pose estimates of each matched pairing of observed and predicted features;

• the robot position estimation based on odometry and observation positions.

5.7 Other Examples of Localization Systems

Markov localization and Kalman filter localization have been two extremely popular strat-
egies for research mobile robot systems navigating indoor environments. They have strong
formal bases and therefore well-defined behavior. But there are a large number of other
localization techniques that have been used with varying degrees of success on commercial
and research mobile robot platforms. We will not explore the space of all localization sys-
tems in detail. Refer to surveys such as [5] for such information.
There are, however, several categories of localization techniques that deserve mention.
Not surprisingly, many implementations of these techniques in commercial robotics