Walking robot simulation

Cilj je ovog diplomskog rada ostvarenje mogućnosti stabilnog hodanja i savladavanja prepreka robotskog agenta te dolazak do cilja koristeći metode podržanog učenja. Kroz rad je dan kratak pregled područja te teorijska podloga podržanog učenja te implementacija algoritama za rješenje zadanoga problema. Koristeći već postojeću implementaciju algoritma Proximal Policy Optimization (PPO) i simulator Isaac Gym robotski je agent uspješno naučen da prati zadane naredbe i dolazi do cilja zadanog u obliku koordinata na karti.The goal of this thesis is to realize the ability of a robot agent to walk and cross obstacles and reach a given goal using reinforcement learning methods. This paper provides a brief overview of the scientific field and the theoretical background of reinforcement learning, as well as the implementation of algorithms to solve the given problem. Using the existing Proximal Policy Optimization algorithm implementation and the Isaac Gym simulator, the robotic agent was successfully trained to follow commands and arrive at a target position provided to it in the form of coordinates in an environment ma