Page 418 - Sensing, Intelligence, Motion : How Robots and Humans Move in an Unstructured World

SALIENT CHARACTERISTICS OF A SENSITIVE SKIN 393

The system will be more reliable if it is capable of self-diagnostics, self-
healing,and graceful degradation. Self-diagnostics is a well-developed discipline
in system design. Our system would continuously poll all its sensors and inform
the control unit of detected deviations from the normal. Self-healing implies
an automatic repair of the failed hardware. On the level of sensors today, this
feature is available in live nature but not in technology. The purpose of graceful
degradation property is to avoid the need to shut off the whole system in case of
losing a few elements. In our case, graceful degradation refers to the system’s
capacity to direct sensors in the vicinity of a broken sensor to take over its job
(see, e.g., Ref. 140). Doing this will effectively increase the distance between
well-functioning sensors; while slightly degraded, the system will still function.
As more sensors die, the system will degrade “gracefully.”

Accuracy. Sensing information has to be accurate. Sensor accuracy is defined
by the difference between the sensor measurement and the actual value that
the sensor attempts to measure. How accurate is accurate enough? The answer
depends on many details, including the sensor types and density on the skin, and
on the robot’s mass, kinematics, and maximum speed. Imagine that our robot arm
manipulator is equipped with short-range proximity sensors capable of detecting
an object when the distance to it from the robot falls below 20 cm. The arm is a
heavy body made of steel; its inertia is high. Imagine that at its maximum speed
the robot control system can guarantee a successful collision-avoiding maneuver
only if the distance to an object on the robot’s way at the time of detection is no
less than 15 cm. This means that we would not be happy with a sensor whose
accuracy is ±5 cm, because with it we would run a risk of the arm colliding
with surrounding objects.
Resolution. This characteristic refers to precision with which the robot can pin-
point the location of an object that appears in the vicinity of the robot body.
The better the robot knows the location of an object that it tries to avoid, the
better the dexterity of its motion and, hence, its chance to accomplish its task in a
workspace filled with obstacles. Sensor resolution is tied to the sensor accuracy,
but it characterizes the whole sensor system rather than a single sensor.
As an example, consider a sensitive skin that is based on proximity infrared
sensing. Unlike passive sensors like a vision camera or a temperature sensor, an
infrared sensor is an active sensor: It contains (a) a light-emitting diode (LED)
that sends a ray of infrared light in space and (b) a detector that sits right next to
the LED and detects the light’s reflection from an object in front of the sensor.
For better resolution we are interested in a sensor that would send light in a
narrow cone. Such sensors, each with a tiny lens in front of the LED in order
to produce a cone-like light ray, are common (see Figure 8.2a). They may have
a limited sensitivity distance, say 20 or 30 cm. For full coverage, sensors are
spaced on the skin so that their sensitivity cones overlap, forming a continuous
sensitivity cushion around the robot. The fact of detecting a reflected signal thus
means that the object in question is wholly or partially within the sensitivity cone
and sensitivity distance of a given sensor.