Modelling a learning algorithm for a mobile robot with FPGA

A neural network algorithm has been developed for a mobile robot that learns to avoid obstacles in a room and learns the environment. The algorithm has been also simulated for implementing into a FPGA. The learning algorithm has been modelled with using PNN (Probabilistic Neural Network) architecture and some qualities have implemented to this architecture to avoid the disadvantages of the algorithm. The hardware architecture has been constructed with VHDL (Very High Speed Integrated Circuits Hardware Description Language) language and simulated with Modelsim. The performance of the algorithm has been also tested with Matlab simulations. And finally, we have developed a simulation interface in Delphi environment which shows a demo room and a robot that learns the room. The interface program enables the user to put the mobil robot, obstacles and the target point to the desired places in an empty room. When the start button is pressed mobil robot begins to move through the target place with calculating the shortest distance and also avoiding from the obstacles in the room. The avoidance from the obstacles has been realised with using the learning algorithm which is briefly introduced in this study. Avoidance from the obstacles algorithm was firstly constructed with using classical PNN architecture. But classical PNN architecture has brung some disadvantages such as long computition time, complicated hardware architecture and choosing wrong navigation direction in some cases. To overcome these disadvantages, some new qualifications have been implemented into neural network architecture and the raise of the performance has been shown with computed results. The success of the learning algorithm of the mobil robot has been calculated as 88.88% in Matlab environment. The performance results and graphics are also presented. Key words: Artificial Intelligence, Neural Network, FPGA, robot navigation