Attributes of a sensor

Attributes of a sensor

As can be seen by the above example, robots may have dead reckoning capabilities,
but will always have some type of exteroceptive sensor. Otherwise,
the robot cannot be considered reactive: there would be no stimulus from
the world to generate a reaction. The set of sensors for a particular robot is
called a sensor suite. Following SENSOR SUITE Sensors for Mobile Robots,52 in order to construct
a sensor suite, the following attributes should be considered for each
sensor:
1. Field of view and range. Every exteroceptive sensor has a region of
space that it is intended to cover. The width of that region are specified by
the sensor’s field of view, often abbreviated as FOV. The field of view is usually
expressed in degrees; the number of degrees covered vertically may be
different from the number of degrees covered horizontally. Field of view is
frequently used in photography, where different lenses capture different size
and shape areas. A wide angle lens will often cover up to 70, while a “regular”
lens may only have a field of view around 27. The distance that the
field extends is called the range.
FIELD OF VIEW (FOV) The field of view (FOV) can be thought of in terms of egocentric spherical

coordinates, where one angle is HORIZONTAL FOV the horizontal FOV and the other is the vertical
VERTICAL FOV FOV. The other aspect is the range, or how far the sensor can make reliable
RANGE measurements. In spherical coordinates, this would be the values of r that
defined the depth of the operating range.
Field of view and range are obviously critical in matching a sensor to an
application. If the robot needs to be able to detect an obstacle when it’s 8 feet
away in order to safely avoid it, then a sensor with a range of 5 feet will not
be acceptable.
2. Accuracy, repeatability, and resolution. Accuracy refers to how correct
the reading from the sensor is. But if a reading for the same conditions is
accurate only 20% of the time, then the sensor has little repeatability. If the
sensor is consistently inaccurate in the same way (always 2 or 3 cm low),
then the software can apply a bias (add 2 centimeters) to compensate. If the
inaccuracy is random, then it will be difficult to model and the applications
where such a sensor can be used will be limited. If the reading is measured
RESOLUTION in increments of 1meter, that reading has less resolution than a sensor reading
which is measured in increments of 1 cm.
3. Responsiveness in the target domain. Most sensors have particular
environments in which they function poorly. Another way of viewing this is
that the environment must allow the signal of interest to be extracted from
noise and interference (e.g., have a favorable signal-to-noise ratio). As will
be seen below, sonar is often unusable for navigating in an office foyer with
large amounts of glass because the glass reflects the sound energy in ways
almost impossible to predict. It is important to have characterized the ecological
niche of the robot in terms of what will provide, absorb, or deflect
energy.
4. Power consumption. Power consumption is always a concern for robots.
Since most robots operate off of batteries, the less power they consume, the
longer they run. For example, the battery life on a Nomad 200, which carries
five batteries,was improved from four hours to six by shutting off all sensors.
Power is so restricted on most mobile robots that many robot manufacturers
will swap microprocessor chips just to reduce the power drain (which was
part of themotivation for the Transmeta Crusoe chip). Sensors which require
a large amount of power are less desirable than those which do not. In general,
passive sensors have less power demands than active sensors because
they are not emitting energy into the environment.
The amount of power on a mobile robot required to support a sensor package
(and any other electronics such as amicroprocessor and communications
HOTEL LOAD links) is sometimes called the hotel load. The sensor suite is the “guest” of the

platform. The power needed LOCOMOTION LOAD to move the robot is called the locomotion load.
Unfortunately, many robotmanufacturers focus on only the locomotion load,
balancing power needs with the desire to reduce the overall weight and size.
This leads to a very small hotel load, and often prevents many sensors from
being added to platform.
5. Hardware reliability. Sensors often have physical limitations on how
well they work. For example, Polaroid sonars will produce incorrect range
reading when the voltage drops below 12V. Other sensors have temperature
and moisture constraints which must be considered.
6. Size. The size and weight of a sensor does affect the overall design. A
microrover on the order of a shoebox will not have the power to transport
a large camera or camcorder, but it may be able to use a miniature “Quick-
Cam” type of camera.
The above list concentrated on considerations for the physical aspects of
the sensor. However, the sensors only provide observations;without the software
perceptual schemas, the behaviors cannot use the sensors. Therefore,
the software that will process the information from a sensor must be considered
as part of the sensor seletion process. 7. Computational complexity.
Computational complexity is the estimate of how many operations an algorithm
or program performs. It is often written as a function O, called the
“order,” where O(x) means the number of operations is proportional to x. x
is often a function itself. Lower orders are better. An algorithm that executes
with O(n) equally consuming operations is faster than one with O(n2) operations.
(If you doubt this, see if you can find a positive, whole number value
of n such that n > n2.) Computational complexity has become less critical
for larger robots, with the rapid advances in processors and miniaturization
of components. However, it remains a serious problem for smaller vehicles.
8. Interpretation reliability. The designer should consider how reliable
the sensor will be for the ecological conditions and for interpretation. The
robot will often have no way of determining when a sensor is providing
incorrect information. As a result the robot may “hallucinate” (think it is
seeing things that are not there) and do the wrong thing. Many sensors produce
output which are hard for human to interpret without years of training;
medical X-rays are one example, and synthetic aperature radar (SAR) which
produces polar plots is another. If a sensor algorithm was not working properly
in these modalities, the designer might not be skilled enough to notice
it. Therefore, the algorithms themselves must be reliable.