Page 158 - Introduction to AI Robotics

141
4.4 Potential Fields Methodologies
member, in this example, the robot is not projecting past or future boundaries
of the corridor; the visualization of the field makes it appear that way.
The follow-corridor behavior is using the same sonar data as avoid;
therefore, walls will produce a repulsive field, which would generally push
the robot onto the middle of the corridor. Why not just use a single uniform
parallel field for follow-corridor? First, behaviors are independent. If
there is a corridor following behavior, it must be able to follow halls without
depending on side effects from other behaviors. Second, the polar symmetry
of the repulsive fields may cause see-sawing, so there is a practical advantage
to having a separate behavior.
The use of behavior-specific domain knowledge (supplied at instantia-
tion time as an optional initialization parameter) can further improve the
robot’s overall behavior. If the robot knows the width of the hall a priori,
follow-corridor can suppress instances of avoid which are using obsta-
cles it decides form the boundary of the wall. Then it will only avoid ob-
stacles that are in the hall. If there are no obstacles, follow-corridor
will produce a smooth trajectory. If the obstacles are next to a wall, the
follow-corridor will treat the profile of the obstacle as a wall and move
closer to the center.
SEQUENCING AND The motor schemas for a behavior may be sequenced. One example of
PARAMETERIZING this is the docking behavior. 12 Docking is when a robot moves to a specific
POTENTIAL FIELDS
location and orientation relative to a docking station. This is useful for robots
performing materials handling in industry. In order to accept a piece of ma-
terial to carry, the robot has to be close enough to the correct side of the end
of a conveyor and facing the right way. Because docking requires a specific
position and orientation, it can’t be done with an attraction motor schema.
That field would have the robot make a bee-line for the dock, even if it was
coming from behind; the robot would stop at the back in the wrong position
SELECTIVE and orientation. Instead, a selective attraction field is appropriate. Here the ro-
ATTRACTION bot only “feels” the attractive force when it is within a certain angular range
of the docking station, as shown in Fig. 4.24.
Unfortunately selective attraction does not cover the case of when the ro-
bot approaches from behind or to the side. How does the robot move to
an area where the selective attraction field can take effect? One way to do
this is to have tangential field, which makes the robot orbit the dock until
it gets into the selective attraction area. The combination of the two motor
schema produces a very smooth field which funnels the robot into the correct
position and orientation, as shown in Fig. 4.25.