Research question 7: How far can CNN be used/involved in an evolutionary computing/control context example? CNN has a great potential for signal processing tasks and it can generate very complex
nonlinear waves and oscillation patterns at the output of CNN cells. Controlling both the
kinematic and the inverse-kinematic of complex robots with high
Degree of Freedom (DOF)
is a very complex scenario whereby classical solutions fail to solve it easily. Complex ADAS
solutions may also be seen as systems with a high degree of freedom, for example CACC
(cooperative adaptive cruise control). Therefore, the evolution of CNN templates to
generate the optimum wave for both the driving motor(s) and the robot actuators is a
particularly interesting idea for research. In nature, organisms’ systems are evolving as
explained by the theory of natural selection. Overall, the art is to define a fitness function
for evaluating the performance of the actual robot locomotion. Hence, by evolving the CNN
template using a genetic algorithm and a fitness function, we can generate very complex
waves for optimally controlling the r
obot hinges without directly involving the robot’s
kinematic equation in controller design. A two dimensional CNN should be sufficient for
evolving a spatial wave over time for controlling the robot’s actuators and hinges.
A fitness function is a particular type of objective function that quantifies the optimality of
a solution [36]. Input data for the fitness function are based on measurements from robot
parts’ orientations, locations and displacements. In the fitness function we do not define
any behavioral locomotion exactly. On the other hand, we define a function that satisfies
the target or destination. With this method involving genetic algorithms an optimum
template ensures that the robot can move or act according to our desires. The most
15
important point in this learning method is that we do not need to (explicitly) predefine any
robot’s kinematics for movement in the fitness function.