111
According to Equation 9-7, the fitness value has a direct
relation with the distance
between the initial position point (
Init_Center_Pos
) and the robot position (
Robot_Pos
)
divided by (R). Further, this function has an inverse relation
with the robot movement
(Robot_ Movement).
(9-7)
Fitness = (1 - Dist (Init_Center_Pos, Robot_Pos) / R) *(1/Robot_Movement)
By optimizing the robot movement/locomotion, the fitness value will converge to zero. For
the case of an unstructured robot, we have designed a broken-leg spider. In this test, the
aim is learning the robot for the turning left and right skill
as a complete and perfect
spider. In Figure 9-15 is shown the representation of this type of robot.
Figure
9-14: Wave generated for circular locomotion
R
Figure
9-13: Moving 6-Leg Robot, around the Circle
Path
Path
112
The definition of the fitness function is a little bit sensitive in this case. According to
Equation 9-8, the fitness value has a direct relation with the distance
between the initial
position point (
Init_Center_Pos
) and the robot position (
Robot_Pos
). Further, this function
has an inverse relation with the robot angle (
Robot_Angle
). During the optimization phase,
the aim/goal is converging the fitness function to zero.
(9-8)
Fitness = Dist (Init_Center_Pos, Robot_Pos)*(1/Robot_Angle)
(9-9)
After nearly 3300 chromosome generations and evolution, the robot is able to turn over
its Yaw axis. Figure 9-16 shows the result of the wave pattern for unstructured spider
turning. The usability of the templates in this paper can be summarized as follows.
Templates are stored in a list/memory. By a high level task
management templates are
selected. This selection depends on the high level task
management decision and the
environment situation as well. Further, a factor of high importance is the behavioral
architecture and behavioral programming. In fact, low level skills (e.g.
Moving forward,
