Page 217 - Designing Autonomous Mobile Robots : Inside the Mindo f an Intellegent Machine

Chapter 13

Calculating virtual confidence

If we normalize the value of virtual confidence to a number from zero to one, then we
can simply multiply the current programmed speed by the confidence and immedi-
ately develop a slowing response to lost confidence. The same can be done for motor
torque limits, acceleration, and deceleration. If we later find this to be a suboptimal
reaction, we can easily add a simple trapezoidal function to the process as we have
discussed for other processes.

Confidence and pain

Any given dangerous experience should erode confidence in proportion to its seri-
ousness. For example, if the robot is forced to circumnavigate an unexpected obstacle, it
might experience a 50% decrease in confidence. A bumper strike on the other hand
might temporarily reduce the confidence to zero, and require an operator “assist” to
bring it back to, say, 33%. An operator assist could be a local or remote restart
command that indicates that the conditions are now safe. The details of this mecha-
nism will necessarily depend on the application and the environment.
Once pain has caused the robot to become less confident, it should stay less confident
until it has successfully driven a sufficient distance without further incident. This
implies that if the robot became less confident due to navigation difficulties (high
uncertainty) then it will need to acquire enough navigational references to reduce
the uncertainty, and then drive a sufficient distance to assure it is out of the area of
danger.

Anyone who has driven country roads late at night knows that if one deer crosses the
road in front of you, there is a very good chance others are following and that it is
wise to slow down. The same is true with most of the environmental problems a mobile
robot will face. For example, if there is one trash can in the aisle, there will probably
be more.

The amount of confidence loss induced by a pain or near-pain experience should there-
fore begin to erode with successful travel at a rate specified in a blackboard variable.

Confidence and danger

The danger of navigating a path segment is the sum of the known inherent danger and
the learned danger we discussed earlier. All areas or paths have an inherent danger as-
sociated with them; the narrower the clearance between obstacles becomes, the
higher the risk.

200