Page 420 - Sensing, Intelligence, Motion : How Robots and Humans Move in an Unstructured World

SALIENT CHARACTERISTICS OF A SENSITIVE SKIN 395

Section 8.4 has roughly this density of sensors. With this sensitive skin, a round
object 10 cm in diameter will be perceived by the robot as about 13–14 cm in
diameter. This also means that a robot equipped with this skin will be able to
move past the obstacle at about 2 cm distance from it, but not closer—otherwise
it risks a collision.
What this resolution signifies is that two objects located at a distance about
2 cm from each other may be perceived by the robot as one obstacle. Imagine
that the robot arm contemplates moving between two obstacles: The diameter of
the arm links is d; the distance between the obstacles is 2 cm + d. Then, based
on the information from the skin, the robot will not attempt to pass between the
two obstacles, although it actually could. It is insufficient skin resolution that
will make the robot miss the passage.
In real life the 2-cm resolution would be quite good; cases with obstacles requir-
ing motion as tight as in the example above would be rare. The point here is that
the skin resolution affects the dexterity of robot motion in the most direct way.
Note also the quadratic dependence between the skin resolution and the number
of sensors on the skin: For example, decreasing in half the distance between
neighboring sensors will increase fourfold the number of sensors on the skin.
Consider now another example, a skin with capacitance sensors. A capacitance
sensor is a passive sensor: It works by measuring properties of the electric field
that the sensor’s two electrodes create. An object entering the field changes the
field characteristics, and this allows the sensor to detect it. As with any electric
field, the detection effect depends on the object’s material and its distance to
the sensor. The sensor’s sensitivity area can take different shapes depending on
the shape and mutual positions of the sensor’s electrodes. The sensitivity area of
the sensor shown in Figure 8.2b is a hemisphere that extends outwards from the
robot body, with the sensor at its center.
With the sensitivity area sphere of radius, say, 10 cm (about 4 in.), a detection
signal from the sensor will tell the robot that an object has entered a hemisphere
of diameter 20 cm centered at the sensor. In Figure 8.2b, the robot will not
know where within the sensitivity area the object is because an object in the
position Ob1 or Ob2 or anywhere else within the sensitivity area may generate
the same signal. To be on the safe side, the robot will have to conclude that
the object occupies the whole sensitivity area. That is, an obstacle of 2 cm in
diameter will become in the robot’s “mind” an “obstacle” of over 20 cm in
diameter. When planning its motion past the 2-cm obstacle, the robot will have
to leave a large margin between itself and the obstacle, which is equivalent to
suddenly increasing the dimension of every obstacle by 20 cm in every direction.
That is, this capacitance-sensitive skin will effectively make the robot workspace
dramatically more crowded than it actually is. Someone diving into a dirty pool
without a diving mask will likely have a better vision resolution. The robot’s
bad maneuverability here is not the robot’s fault—it is just that its sensors have
unacceptably low resolution.
Compared to the 1200 to 1500 infrared sensors above, achieving a full coverage
for the same large arm manipulator with our capacitance sensors will require