High Performance on Atari Games Using Perceptual Control Architecture Without Training

Abstract Deep reinforcement learning (DRL) requires large samples and a long training time to operate optimally. Yet humans rarely require periods of perform well on novel tasks, such as computer games, once they are provided with an accurate program instructions. We used perceptual control theory (PCT) construct simple closed-loop model which no within video game study using the Arcade Learning Environment (ALE). The was programmed parse inputs from environment into hierarchically organised signals, it computed dynamic error signal by subtracting incoming for each variable reference drive output signals reduce this error. tested same across three different Atari games Breakout, Pong Video Pinball achieve performance at least high DRL paradigms, close good human performance. Our shows that models, based assumptions, can without learning. conclude specifying parsimonious role may be more similar psychological functioning.