Application of reinforcement learning for mechatronic system control

Ovaj diplomski rad bavi se primjenom Actor-Critic algoritama, metode podržanog učenja, u kontroli mehatroničkih sustava, sa specifičnim fokusom na problem obrnutog njihala. Mehatronički sustavi, koji kombiniraju mehaničko inženjerstvo, elektroniku, računalne znanosti i informacijske tehnologije, iznimno su složeni za kontrolu i upravljanje. Unatoč izazovima, primjena umjetne inteligencije, posebno podržanog učenja, postaje sve relevantnija u ovom polju. Glavni fokus rada je teorijska analiza i praktična primjena Actor-Critic algoritama u kontroli mehatroničkog sustava, posebno sustava s pendulom. Cilj je razviti kontrolni algoritam koji može balansirati njihalo u uspravnom položaju, što predstavlja klasični problem obrnutog njihala.This thesis explores the application of Actor-Critic algorithms, a method of reinforcement learning, in the control of mechatronic systems, specifically focusing on the inverted pendulum problem. Mechatronic systems, integrating mechanical engineering, electronics, computer science, and information technologies, are extremely complex to control and manage. Despite the challenges, the application of artificial intelligence, particularly reinforcement learning, is becoming increasingly relevant in this field. The main focus of the thesis is the theoretical analysis and practical application of Actor-Critic algorithms in controlling a mechatronic system, specifically a pendulum system. The goal is to develop a control algorithm that can balance the pendulum in an upright position, which represents the classic inverted pendulum problem