| Abstract: |  The thesis deals with the evolution of a discrete-time reccurent aritificial neural network to control an autonomous mobile robot. Standard genetic algorithm is employed to solve the obstacle avoidance problem. Evolution takes place in several specific simulated environments in Player/Stage. The obtained solutions are analyzed in all environments and compared with each other. The ability of adapting to particular robot configuration is explored by evolving a new control system with higher frequency of sensors readings. Robustness is tested by placing the best control system into the real mobile robot Morbot in the robotic arena SyRoTek. Reliability of its navigation is at least 60 %. Achieved results show that evolved control system is well adapted to specific possibilities of the robot and the environment, so that it maximizes its fitness function. It is also able to control the real robot properly. From that a conclusion might be drawn, that it is possible to evolve an aritificial neural network in a simulated environment and to use the found solution for controlling of the real robot. |
|---|