Page 222 - Designing Autonomous Mobile Robots : Inside the Mindo f an Intellegent Machine

Instilling Pain, Fear and Confidence

levels around the robot. Thus, the flow of information is the mirror of that for the
virtual path method.

Confidence and virtual fitness

Here we add another anthropomorphic concept that is a metaphor for health and
acuteness of the senses. For example, let’s assume that a robot has very low batteries.
If it attempts to accelerate at its full rate to reach its top speed, then the robot will
probably experience a servo stall. Worse, servo performance may become sluggish
during emergency maneuvers, and the robot will be wasting battery power it may
need to conserve for the return to its charger.
In Cybermotion robots, the platform automatically quits accelerating as the servo
begins to approach its current limit or full power command. Even so, we found it bene-
ficial to reduce the velocity target as the batteries became depleted.
Another example of low fitness would be one or more sensor systems being partially
or completely disabled. There are many reasons why this might happen. For example, in
the presence of white noise sound sources like a leaking air line, sonar will be less
effective than in a quiet environment.
In other cases, sensors must intentionally be desensitized to accomplish a maneuver.
As mentioned earlier, it may be necessary for the robot to completely ignore a col-
lision avoidance sensor system in order to complete an operation such as mating with its
dock. If this is done, however, the robot’s speed and power must be severely limited.

Flashback…

An interesting example of the effectiveness of such precautions occurred shortly after the
story of the robot that tried to run me down because of a typo in one of its programs (see
Chapter 2). Because of that experience, we had modified our vehicle code to slow the
robot down as a result of several of these fitness factors. It was in fact an early form of
fitness confidence.

I was visiting a Navy site that was using our robots. For some months, I had been moni-
toring reports on the effectiveness of our collision avoidance, and there had been several
cases in which the robot had failed to stop for people in its way. However, in other tests
it had successfully detected and stopped for very small objects on the floor. The path
programs where these incidents had occurred had been checked, and did not contain
errors such as the one that had resulted in the pushy robot I had experienced.

205