Page 200 - Introduction to Autonomous Mobile Robots

Mobile Robot Localization

5.2.3 Effector noise 185
The challenges of localization do not lie with sensor technologies alone. Just as robot sen-
sors are noisy, limiting the information content of the signal, so robot effectors are also
noisy. In particular, a single action taken by a mobile robot may have several different pos-
sible results, even though from the robot’s point of view the initial state before the action
was taken is well known.
In short, mobile robot effectors introduce uncertainty about future state. Therefore the
simple act of moving tends to increase the uncertainty of a mobile robot. There are, of
course, exceptions. Using cognition, the motion can be carefully planned so as to minimize
this effect, and indeed sometimes to actually result in more certainty. Furthermore, when
the robot’s actions are taken in concert with careful interpretation of sensory feedback, it
can compensate for the uncertainty introduced by noisy actions using the information pro-
vided by the sensors.
First, however, it is important to understand the precise nature of the effector noise that
impacts mobile robots. It is important to note that, from the robot’s point of view, this error
in motion is viewed as an error in odometry, or the robot’s inability to estimate its own posi-
tion over time using knowledge of its kinematics and dynamics. The true source of error
generally lies in an incomplete model of the environment. For instance, the robot does not
model the fact that the floor may be sloped, the wheels may slip, and a human may push
the robot. All of these unmodeled sources of error result in inaccuracy between the physical
motion of the robot, the intended motion of the robot, and the proprioceptive sensor esti-
mates of motion.
In odometry (wheel sensors only) and dead reckoning (also heading sensors) the posi-
tion update is based on proprioceptive sensors. The movement of the robot, sensed with
wheel encoders or heading sensors or both, is integrated to compute position. Because the
sensor measurement errors are integrated, the position error accumulates over time. Thus
the position has to be updated from time to time by other localization mechanisms. Other-
wise the robot is not able to maintain a meaningful position estimate in the long run.
In the following we concentrate on odometry based on the wheel sensor readings of a
differential-drive robot only (see also [4, 57, 58]). Using additional heading sensors (e.g.,
gyroscope) can help to reduce the cumulative errors, but the main problems remain the
same.
There are many sources of odometric error, from environmental factors to resolution:
• Limited resolution during integration (time increments, measurement resolution, etc.);
• Misalignment of the wheels (deterministic);

• Uncertainty in the wheel diameter and in particular unequal wheel diameter (determin-
istic);

• Variation in the contact point of the wheel;