Machine learning applications in robotics

Rad se bavi temom primjene strojnog uˇcenja u robotici, specifiˇcno ojaˇcanim uˇcenjem. Opisuje ideju ojaˇcanog uˇcenja, njegove najbitnije dijelove i glavne podjele algoritama u ojaˇcanom uˇcenju. U nastavku se opisuju testovi s tri razliˇcita algoritma iz skupine glumac-kritiˇcar na dvije razliˇcite okoline i usporeduju se rezultati. Dodatno se joˇs testira utjecaj funkcije nagradivanja na ponaˇsanje agenta. Za testove se koristi simulacijsko okruˇzenje robota iz paketa OpenAI Gym.This paper deals with the subject of machine learning in robotics, specifically reinforcement learning. It describes the idea of reinforcement learning, its fundamental parts and the classification of reinforcement learning algorithms. Further it describes the tests performed with three different algorithms from the actor-critic group, on two different environments, and compares the results. The tests are performed in simulation robot environments from the package OpenAI Gym