2018/08/15 by Per‐Arne Andersen, Andersen, Per-Arne, Morten Goodwin +3 · 1 citation
Computer Science · Social Sciences · #Artificial Intelligence in Games #Digital Games and Media #Reinforcement Learning in Robotics
paper · pdf · doi:10.48550/arxiv.1808.05032
Reinforcement learning (RL) is an area of research that has blossomed\ntremendously in recent years and has shown remarkable potential for artificial\nintelligence based opponents in computer games. This success is primarily due\nto the vast capabilities of convolutional neural networks, that can extract\nuseful features from noisy and complex data. Games are excellent tools to test\nand push the boundaries of novel RL algorithms because they give valuable\ninsight into how well an algorithm can perform in isolated environments without\nthe real-life consequences. Real-time strategy games (RTS) is a genre that has\ntremendous complexity and challenges the player in short and long-term\nplanning. There is much research that focuses on applied RL in RTS games, and\nnovel advances are therefore anticipated in the not too distant future.\nHowever, there are to date few environments for testing RTS AIs. Environments\nin the literature are often either overly simplistic, such as microRTS, or\ncomplex and without the possibility for accelerated learning on consumer\nhardware like StarCraft II. This paper introduces the Deep RTS game environment\nfor testing cutting-edge artificial intelligence algorithms for RTS games. Deep\nRTS is a high-performance RTS game made specifically for artificial\nintelligence research. It supports accelerated learning, meaning that it can\nlearn at a magnitude of 50 000 times faster compared to existing RTS games.\nDeep RTS has a flexible configuration, enabling research in several different\nRTS scenarios, including partially observable state-spaces and map complexity.\nWe show that Deep RTS lives up to our promises by comparing its performance\nwith microRTS, ELF, and StarCraft II on high-end consumer hardware. Using Deep\nRTS, we show that a Deep Q-Network agent beats random-play agents over 70% of\nthe time. Deep RTS is publicly available at https://github.com/cair/DeepRTS.\n