Summary

Summary

The success of a reactive system depends on the suitability of the robot’s sensor
suite. It is often more useful to think of sensing in terms of perceptual
schemas or logical sensors needed to accomplish a task, rather than focus on
the characteristics of a particular transducer or modality. Reactive perception
may appear limited since it uses only behavior-specific representations
and does not involve recognition. However, it supports a diversity of forms
of perception, including behavioral sensor fusion. Advances in electronics
have led to a plethora of range sensors and algorithms. Many of these are
logically equivalent and can be used interchangeably; for example, with obstacle
avoidance behaviors.
The design of a perceptual schema or a sensor suite requires a careful analysis.
Each individual sensor should fit the task, power, and processing constraints.
Likewise, the entire sensor suite should provide complete coverage
of all perceptual processing required by the robot.
Almost all mobile robots have some form of proprioception, most likely
shaft or wheel encoders used to estimate the distance traveled based on the
number of times the motor has turned. Outdoor robots may carry GPS, and
this trend is expected to increase as the cost of receivers goes down and inexpensive
DGPS systems emerge.
Reactive navigation requires exteroception, whereby the robot observes
the environment. Proprioception can guide a robot on a path, but exteroception
can prevent it from hitting an unmodeled obstacle or falling off a cliff.
The most common exteroceptive sensor on reactive robots is an ultrasonic
transducer or sonar. An ultrasonic transducer is an active sensor which returns
a single range reading based on the time-of-flight of an acoustic wave.
Some of the difficulties associated with ultrasonics include erroneous readings
due to specular reflection, crosstalk, and foreshortening. Other popular
proximity sensors are IR and laser rangers.
Due to the low price and availability of consumer electronics, computer
vision is becoming more common in robotic systems. Computer vision processing
operates on images, regardless of the modality which generated it.
Color coordinate systems tend to divide an image into 3 planes. The two

most common color coordinate systems are RGB and HSV. HSV treats color
in absolute terms, but RGB is favored by equipment manufacturers. A color
space used in biomedical imaging, SCT, appears to be less sensitive to lighting
conditions than RGB and RGB-derivedHSV.Many reactive robots exploit
color as an affordance. This can be done by thresholding an image and identifying
regions of the appropriate color. A color affordance method which
works well for objects with multiple colors is color histogramming. Stereo
range finding is an important class of algorithms for navigation, though the
computational complexity has prevented it being ported to many mobile robot
applications. Laser range finders, particularly the inexpensive planar
rangers, have grown in popularity over the past few years.
Despite the diversity of sensors and affordances inherent in the environment,
reactive robotics is remarkable for its lack of sophistication in sensing.
This may stem from the split between computer vision and robotics in the
formative years of the field. Many roboticists still assume algorithms developed
by computer vision specialists are too computationally expensive
to work on commercially available on-board processors. This is no longer
true, in part because of the increased computational power of general purpose
chips. Readers are encouraged to explore the large body of literature on
computer vision and free tools on the web.