Page 219 - Designing Autonomous Mobile Robots : Inside the Mindo f an Intellegent Machine

Chapter 13

separately in the force field method, we can lump the calculation with that of posi-
tion uncertainty. We will discuss this shortly.

Position confidence

Here we find yet another use of the robot’s uncertainty estimate. If a robot is moving
without good navigational feedback, then its position may be well out of the planned
path or area. This can lead to serious consequences, particularly when there are mapped
dangers that the robot cannot detect with its sensors. Such dangers might include drop-
offs, overhead obstacles, and other moving machinery. The chances of avoiding these
types of obstacles, and the potential damage that could be caused by an encounter with
one of them are increased at higher speeds.
Conversely, slowing down gives the robot’s sensors a better chance of imaging a nav-
igational feature. If traffic is blocking the robot’s view of navigation features, then
not only will slowing down increase the chances of glimpsing some features through
the traffic, but fundamentally it is a safe thing to do in a crowded situation.
We cannot simply slow the robot down in all areas because this would diminish the
work it could accomplish in a given time, and thus the payback period. It is crucial
that the robot run as fast as is safe whenever possible.
For example, assume our robot is operating near the entrance of a theater when the
show ends. The crowd of people may well block the robot’s view of its navigation fea-
tures and disrupt navigation. If the robot continues to move rapidly, it will experience
greater and greater navigational uncertainty and may become unreferenced. Worse
yet, it might experience an exchange of kinetic energy with an unwary patron!
Uncertainty has a lateral, a longitudinal, an azimuth, and possibly a height compo-
nent. A conservative method of evaluating the danger from uncertainty is to use the
vector sum of the lateral and longitudinal components. A less conservative method
is to simply use the lateral uncertainty since it is most likely to represent the threat
of contact with a mapped or unmapped danger. But what should this value be com-
pared against? If the robot is driving through a narrow aisle, the tolerable uncertainty is
obviously less than if it is traveling across an open lobby.
In the virtual path approach of navigation, the allowable deviation from the path is
usually set by an instruction. This can be explicit, or implied from a navigational
reference to walls or aisles. During circumnavigation, the robot will not be permitted
to move away from the path by more than these left- and right-hand limits minus its

202