Page 264 - Introduction to AI Robotics

247
6.8 Case Study: Hors d’Oeuvres, Anyone?
The initial behaviors for Borg Shark are given in the behavior table below:
Releaser Behavior Motor Schema Percept Perceptual Schema
always on avoid() vfh() most-open-direction polar-plot(sonar)
FOOD-REMOVED= track-face center-face(face-centroid) face-centroid find-face(vision)
treat-removal(laser) track-face()
check-VIP() ribbon-color look-for-ribbon(VIP-color)
SERVING-TIME-OUT, move-to-goal pfields.attraction(waypoint) waypoint list of waypoints
TRAY-FULL=bumper()
FOOD-DEPLETED= track-face center-face(face-centroid) face-centroid find-face(vision)
treat-removal(laser)
The initial behavior table for Puffer Fish was:

Releaser Behavior Motor Schema Percept Perceptual Schema
always on avoid() vfh() most-open-direction polar-plot(sonar)
AT-HOME= sleep() turn-camera-head() obstacle polar-plot(sonar)
dead-reckoning(encoders) cycle-skirt()
AWAKE=radio-signal() move-to-goal() pfields.attraction(location) relative-location read-encoders()
AWAKE=radio-signal() move-to-goal() pfields.attraction(shark) shark find-shark-blue(camera)
TRAY-FULL=bumper() move-to-goal() pfields.attraction(home) relative-location read-encoders()

The vfh behavior is an obstacle avoidance behavior using polar plots de-
rived from models described in Ch. 11. As the team tested the behaviors
individually, the find-face and treat-removal behaviors proved to be
unreliable. While color was a reasonable affordance for a face, the algorithm
often returned false negatives, missing faces unless in bright light. Mean-
while the laser appeared to occasionally get a reflection from the teeth, also
generating false positives, and more than 75% of the time it would miss a
person’s hand if the motion was quick. The rates were:

logical sensor False Positives False Negatives
Face-Find 1.7% 27.5%
Food-Count 6.7% 76.7%

The solution to the find-face performance was to exploit another affor-
dance of a human, one used by mosquitoes: heat. The problem was partial
segmentation; candidate regions were getting rejected on being too small.
Heat would make a good the decision criteria. If a candidate region was co-
located with a hot region, then it was declared a face. Fortunately, the team
2
was able to transfer an E T digital thermometer used on another robot to
Borg Shark. The thermal sensor shown in Fig. 6.30 was intended for deter-