Page 265 - Introduction to Autonomous Mobile Robots

250

5.8 Autonomous Map Building Chapter 5

All of the localization strategies we have discussed require human effort to install the robot
into a space. Artificial environmental modifications may be necessary. Even if this not be
case, a map of the environment must be created for the robot. But a robot that localizes suc-
cessfully has the right sensors for detecting the environment, and so the robot ought to build
its own map. This ambition goes to the heart of autonomous mobile robotics. In prose, we
can express our eventual goal as follows:
Starting from an arbitrary initial point, a mobile robot should be able to autonomously
explore the environment with its on-board sensors, gain knowledge about it, interpret the
scene, build an appropriate map, and localize itself relative to this map.
Accomplishing this goal robustly is probably years away, but an important subgoal is
the invention of techniques for autonomous creation and modification of an environmental
map. Of course a mobile robot’s sensors have only a limited range, and so it must physically
explore its environment to build such a map. So, the robot must not only create a map but
it must do so while moving and localizing to explore the environment. In the robotics com-
munity, this is often called the simultaneous localization and mapping (SLAM) problem,
arguably the most difficult problem specific to mobile robot systems.
The reason that SLAM is difficult is born precisely from the interaction between the
robot’s position updates as it localizes and its mapping actions. If a mobile robot updates
its position based on an observation of an imprecisely known feature, the resulting position
estimate becomes correlated with the feature location estimate. Similarly, the map becomes
correlated with the position estimate if an observation taken from an imprecisely known
position is used to update or add a feature to the map. The general problem of map-building
is thus an example of the chicken-and-egg problem. For localization the robot needs to
know where the features are, whereas for map-building the robot needs to know where it is
on the map.
The only path to a complete and optimal solution to this joint problem is to consider all
the correlations between position estimation and feature location estimation. Such cross-
correlated maps are called stochastic maps, and we begin with a discussion of the theory
behind this approach in the following section [55].
Unfortunately, implementing such an optimal solution is computationally prohibitive. In
response a number of researchers have offered other solutions that have functioned well in
limited circumstances. Section 5.8.2 characterizes these alternative partial solutions.

5.8.1 The stochastic map technique
Figure 5.38 shows a general schematic incorporating map building and maintenance into
the standard localization loop depicted by figure 5.28 during the discussion of Kalman filter