Page 212 - Designing Autonomous Mobile Robots : Inside the Mindo f an Intellegent Machine

Instilling Pain, Fear and Confidence

In the above example, there are two equally fast ways to get from G to A. For sim-
plicity, we will assume that the transit time for each path is the same as the calculated
time. If a danger factor is added to the transit cost to reflect things that have gone
wrong on a path in the past, then this element can be increased with each unpleas-
ant episode, and can be gradually diminished with the passing of time. This way, the
robot will eventually go back to use the path again. Transiting the path without inci-
dent can also be used as an opportunity to decrement its danger factor by an additional
amount.

Different types of painful or delaying experience may occur on the same path. Each
type of unpleasantness will increment the path cost by an amount proportional to its
3
seriousness. These weighting factors should be blackboard variables so that they can
be modified later for differences in environments.
Thus, if path DA is the scene of a bumper strike, then the slightest fear penalty to its
cost will send the robot around the route GE-EB-BA. If the EB segment subsequently
begins collecting danger factor, it would take more of a danger factor to force the robot
around the long route than it took to force it from the first route because the transit
cost of this alternate route is higher. This should be intuitively logical. In fact, the
robot might go back to using the first path until and unless additional problems occurred
on the DA segment.
On the other hand, consider a case where the robot had been penalized on the DA
path and reverted to the GE-EB-BA route, and then experienced pain on the BA
path. In this case, it would not consider the longer route as this would also have the
BA danger factor cost. With nothing more than addition and subtraction, we have
created a very effective protective response for the robot.

Sources of virtual pain

Some of the sources of virtual pain and annoyance that may indicate when a path
should receive a danger penalty include:

1. Bumper strikes have occurred.

2. The robot has been forced to circumnavigate obstacles.
3. The robot has been recalled from the path due to blockages.

3 See Chapter 6 on the importance of blackboards and of avoiding embedded constants.

195