Applications of cellular neural networks in robotics

Smart systems are one of the most interesting improvements in technology that plot an evaluation trajectory from the science-fiction to the real life. Robotics is one of the few technologies that people easily accept the ?intelligence? in it. In fact robotics, besides of being the science and technology of robots, is the mimicking process of the nature. Considering many billion years of ongoing evolutions that naturally select only the survivors, it is not surprising that mimicking is one of the most common design techniques in robotics. The spectrum of the mimicking process is very wide. For example both the control problem of an insects leg and analising the same insects retina structure could be individual mimicking processes. This thesis consists of two parts that have both robotics and Cellular Nonlinear/Neural Networks (CNN) in common. The beginning of the first part introduces some techniques that are already introduced in the literature to control the legs of multi-pod robots synchronously. Then it is shown that it is possible to convert analog representation of the CNN to digital in order to emulate Reaction Diffusion CNN (RD-CNN) structure on a digital environment. Finally simulation results are given to show that it is possible to implement RD-CNN on a digital system. Second part of this thesis starts with a brief introduction of the Gauss-like and the Gabor-like linear CNN filters to mimic the preprocessing procedures of the retina. Secondly some linear algebra techniques are introduced in order to solve state equations of the discrete linear CNN structure. Thirdly some optimisations have been made on these techniques that increase the calculation speed and decrease the memory requirements. Finally some simulation results are given with comparisons. Keywords: Cellular neural networks, robotics, reaction diffusion, sparse matrices.