Abstract
This paper describes an AI agent that plays the popular first-person-shooter(FPS) video game `Counter-Strike; Global Offensive' (CSGO) from pixel input.The agent, a deep neural network, matches the performance of the mediumdifficulty built-in AI on the deathmatch game mode, whilst adopting a humanlikeplay style. Unlike much prior work in games, no API is available for CSGO, soalgorithms must train and run in real-time. This limits the quantity ofon-policy data that can be generated, precluding many reinforcement learningalgorithms. Our solution uses behavioural cloning - training on a large noisydataset scraped from human play on online servers (4 million frames, comparablein size to ImageNet), and a smaller dataset of high-quality expertdemonstrations. This scale is an order of magnitude larger than prior work onimitation learning in FPS games.