Page 213 - Designing Autonomous Mobile Robots : Inside the Mindo f an Intellegent Machine

Chapter 13

4. The drive or steer servo has experienced current limiting or stalling.
5. Excessive numbers of navigation features have been missed.

6. Position uncertainty has grown excessive.
At some level of danger, it is not safe to use a path, even if it is the only route avai-
lable to a destination. At this point, a path is considered to be suspended from service.
Eventually a path must be restored to service or we will gradually find that there are
fewer and fewer routes available to the robot. Paths suspended by problems should
always return to service after some period.
In the CyberGuard system, we provided a method by which the operator could click
on a path and either temporarily suspend it or permanently disable it. This feature
proved to be very useful during extended periods of construction or moving. Remember
also that there may be a mirror image path going the other direction. Generally, if
this is the case then this mirror path should receive the same danger penalty.

Forgetting danger

To assure that the robot will eventually use a path again, it is necessary that the danger
factor decay with time. If we decrease (decay) the penalty in a path too quickly over
time we will repeatedly expose the vehicle to situations that may require operator
intervention or even cause damage. If we wait too long, we may disable a useful path
needlessly for hours due to a very transient problem.
One simple solution to this dilemma is to decrease the decay rate every time a path
is taken out of service, and increase the decay rate a little whenever the path has run
well. Both a minimum and maximum decay rate can be specified to keep the recov-
ery time reasonable. In this way, recurring problems will cause the robot to avoid the
path for longer periods.
The base decay rate for danger factors must be a blackboard variable as well. The reason
for this is that different environments tend to exhibit different danger persistence times.
For example, in an office building a security robot will encounter cleaning personnel and
their equipment fairly frequently, especially early in the evening. On the other hand,
these blockages do not stay in one place longer than a few minutes. In a warehouse,
however, blockages encountered after hours are almost sure to remain in place all
night.
An ideal system would take this concept to the next level by associating dangers with
times of day, days of the week, and other conditions to modify danger values even

196