Reinforcement Learning (RL) has seen vast developments in the past decade. Video games are particularly well suited to generate environments where RL can be used and tested. In this paper, we compare several well documented RL methods, both off and on policy, on various Doom-based environments with different learning objectives. This paper shows that the current state-of-the-art methods, namely DDDQN and A3C, perform best on the different doom environments and surpass human performance.
Pierre Le Jeune
PhD Student in Deep Learning
My research interests include computer vision, deep learning and applications in low-data regime.