Page 216 - Introduction to AI Robotics

6.2 Behavioral Sensor Fusion

Figure 6.2 Example of redundant top and bottom sonar rings. 199

will often produce different false positive and false negative rates. Whether
a robot can tolerate a higher false positive or false negative rate depends on
the task.
When the sensors are both returning the same percept, the sensors are con-
REDUNDANT sidered redundant.An example of physical redundancy is shown in Fig. 6.2,
PHYSICAL where a Nomad 200 has two sonar rings. The sonar software returns the
REDUNDANCY
minimum reading (shortest range) from the two, providing a more reliable
reading for low objects which would ordinarily specularly reflect the beam
LOGICALLY from the upper sonar. Sensors can also be logically redundant, where they re-
REDUNDANT turn identical percepts but use different modalities or processing algorithms.
An example is extracting a range image from stereo cameras and from a laser
COMPETING SENSORS range finder. Sometimes redundant sensors are called competing sensors,be-
cause the sensors can be viewed as competing to post the “winning” percept.
Complementary sensors provide disjoint types of information about a per-
cept. In behavioral sensor fusion for urban search and rescue, a robot may
search for survivors by fusing observations from a thermal sensor for body
heat with a camera detecting motion. Both logical sensors return some aspect
of a “survivor,” but neither provides a complete view. Coordinated sensors
use a sequence of sensors, often for cue-ing or providing focus-of-attention.
A predator might see motion, causing it to stop and examine the scene more
closely for signs of prey.
Most of the work on sensor fusion treats it as if it were a deliberative pro-