100
pixels and is represented in a table of numbers called matrix. The size of this matrix
depends upon the number of joints in the walker robot. In Equation9-1, the stars stand for
convolution operations.
The genetic algorithm is a heuristic search technique used in computing to find either
exact or approximate solutions for optimizing a given problem. The GA is an evolutionary
algorithm that uses techniques inspired from biology such as inheritance, mutation,
selection, and crossover. In this paper, this algorithm is used for finding the best templates
for optimum robots locomotion. The complete structure of the system used for the
training process is shown in Figure 9-2. This structure consists of six main parts: (1) Initial
Population; (2) Crossover; (3) Mutation; (4) Fitness Function; (5) Decoding; (6) Cellular
Neural Network Simulator. In Figure 9-3 the connections between the robot
actuators/hinges and the CNN outputs are shown. These connections are exploited in the
control of both robot hinges and actuators. Wave rhythms are generated from the CNN
processor outputs which can drive the walker robot on a specific path and/or direction
depending on the high level task each of which consists of many low level tasks. After the
learning phase, the output waves can drive the robot with a minimum energy and a good
efficiency. This driving depends upon specific choices of templates values. Each template
set is a solution for driving the robot by means of (or by performing) some specific low
level tasks.
