Deep Reinforcement Learning for robot control
Recent trends in robotics aim to enable robots to work in dynamic, open environments co-occupied by humans, which present several new challenges. When working in these complex scenarios, a robot must be equipped with certain sensors that allow it to perceive its surroundings and the objects it has to interact with. Research has shown that applying Deep Reinforcement Learning to solve complex robotics tasks is a promising solution to the shortcomings of traditional methods.
After giving a brief introduction to DRL we will present robo-gym, an open-source framework that we developed to train DRL models for robot control. The framework allows to train the model on multiple robot simulations in parallel and to deploy it directly on the physical robots.We showcase the capabilities and the effectiveness of the framework with two real world applications featuring industrial robots: a mobile robot and a robot arm. Finally, we discuss the challenges of applying DRL to real-world problems.
Matteo Lucchi: Ricercatore in AI e robotica presso l’istituto di ricerca JOANNEUM RESEARCH Robotics a Klagenfurt in Austria.
Dopo essersi laureato in Ingegneria Meccatronica all’Università degli Studi di Modena e Reggio Emilia ha approfondito le sue conoscenze nel campo dell’intelligenza artificiale per applicarle al controllo di robot. I suoi interessi di ricerca sono lo sviluppo di sistemi per il controllo di robot volti ad automatizzare operazioni complesse in campo industriale.