Machine learning is increasingly emerging as a solution to problems that are difficult to solve with classical approaches. In machine learning, the algorithm is trained on data, rather than written explicitly. In this thesis, we developed a mobile robot controller using methods of machine learning. We developed a line detector, that works well under different conditions. We created a simulation environment, designed for testing robot control algorithms, and developed a line following controller using reinforcement learning methods. The controller was transferred from simulation to the real world. In the simulation, controller performance is comparable to the classic PID controller, whereas in the real world, it is considerably worse. That could be improved in future work. The detector and controller are capable of real-time operation on limited platforms such as Raspberry Pi.
|
