Abstract:

As mobile robots continue to play an increasingly vital role in diverse applications, ensuring their safe and efficient navigation in unstructured environments remains a paramount challenge. The navigation problem can be effectively decomposed into distinct components: perception, localization, planning, and control. However, the multitude of algorithms and approaches available for each of these tasks underscores the complexity of the challenge, with no one-size-fits-all solution. In recent years, the remarkable capabilities of deep neural networks in function modeling have opened up many possibilities for addressing these challenges. In the context of robotics, Deep Reinforcement Learning (DRL) is especially promising, since it enables the training of neural networks through interaction with the environment, without the need for labeled samples. Leveraging model-free DRL, our research has focused on harnessing the power of neural networks for mobile robotics. I will present our work in applying DRL to train neural networks for navigation of mobile robots. I will talk about the practical and theoretical considerations essential for the implementation of DRL in the field of robotics.

Brief bio:

Matej Dobrevski is a control engineer, turned roboticist, turned PhD student at the Visual Cognitive Systems Laboratory at the University of Ljubljana, Faculty of Computer and Information Science under the supervision of Prof. Danijel Skočaj. His interests are deep learning computer vision and robotics. He has worked on various projects in computer vision.