Machine learning algorithm for tracking a swarm of mobile millirobots

Strojno učenje, podveja umetne inteligence, se je v zadnjih letih uveljavilo kot obetavno orodje s sposobnostjo inoviranja industrij, lajšanja vsakdana ter oblikovanja prihodnosti družbe. Zaradi njegovega izjemnega računalniškega potenciala se strojno učenje vedno pogosteje pojavlja tudi za reševanje problemov v večrobotskih sistemih. Detekcijski algoritem YOLOv5 smo s pomočjo strojnega učenja izurili za prepoznavanje milirobotov z računalniškim vidom. Izdelali smo program v programskem jeziku Python, ki s pomočjo izurjenega detekcijskega algoritma in algoritma sledenja DeepSORT omogoča detekcijo in sledenje posameznih milirobotov v roju. Program shranjuje podatke o lokacijah milirobotov in omogoča vizualizacijo njihovega gibanja. Uspešno realizacijo programa smo preverili s pomočjo robotske testne celice s sistemom štirih kamer.Machine learning, a branch of artificial intelligence, has established itself in recent years as a promising tool with the ability of innovating industries, making everyday life easier and shaping the future of our society. Due to its extraordinary computational potential, machine learning is increasingly used to solve problems in multi-robot systems. We trained the YOLOv5 detection algorithm with the help of machine learning to recognize individual millirobots using computer vision. We coded a program using the programming language Python, which, with the help of the trained detection algorithm and the DeepSORT tracking algorithm, detects and tracks millirobots in a swarm. The program stores the location data of individual millirobots and enables the visualization of their movements. The program has been tested in a robot test cell with a four-camera system