Page 228 - Introduction to Autonomous Mobile Robots

213
Mobile Robot Localization
starts to move, say from a precisely known location, it can keep track of its motion using
odometry. Due to odometry uncertainty, after some movement the robot will become very
uncertain about its position (see section 5.2.4). To keep position uncertainty from growing
unbounded, the robot must localize itself in relation to its environment map. To localize,
the robot might use its on-board sensors (ultrasonic, range sensor, vision) to make observa-
tions of its environment. The information provided by the robot’s odometry, plus the infor-
mation provided by such exteroceptive observations, can be combined to enable the robot
to localize as well as possible with respect to its map. The processes of updating based on
proprioceptive sensor values and exteroceptive sensor values are often separated logically,
leading to a general two-step process for robot position update.
Action update represents the application of some action model Act to the mobile robot’s
proprioceptive encoder measurements o and prior belief state s to yield a new belief
t t 1
–
state representing the robot’s belief about its current position. Note that throughout this
chapter we assume that the robot’s proprioceptive encoder measurements are used as the
best possible measure of its actions over time. If, for instance, a differential-drive robot had
motors without encoders connected to its wheels and employed open-loop control, then
instead of encoder measurements the robot’s highly uncertain estimates of wheel spin
would need to be incorporated. We ignore such cases and therefore have a simple formula:

(
,
s' = Act o s t 1 . ) (5.16)
–
t
t
Perception update represents the application of some perception model See to the
mobile robot’s exteroceptive sensor inputs i and updated belief state s' to yield a refined
t t
belief state representing the robot’s current position:
,
(
s = See i s' ) (5.17)
t t t
The perception model See and sometimes the action model Act are abstract functions
of both the map and the robot’s physical configuration (e.g., sensors and their positions,
kinematics, etc.).
In general, the action update process contributes uncertainty to the robot’s belief about
position: encoders have error and therefore motion is somewhat nondeterministic. By con-
trast, perception update generally refines the belief state. Sensor measurements, when com-
pared to the robot’s environmental model, tend to provide clues regarding the robot’s
possible position.
In the case of Markov localization, the robot’s belief state is usually represented as sep-
arate probability assignments for every possible robot pose in its map. The action update
and perception update processes must update the probability of every cell in this case.
Kalman filter localization represents the robot’s belief state using a single, well-defined