Hu, Huan (Southeast University), Wang, Qingling (Southeast University)

Implementation on Benchmark of SC2LE Environment with Advantage Actor - Critic Method

Scheduled for presentation during the Invited Session "Artificial Intelligence and its Applications to Unmanned Flight Systems" (WeC1), Wednesday, September 2, 2020, 18:00−18:20, Macedonia Hall

2020 International Conference on Unmanned Aircraft Systems (ICUAS), September 1-4, 2020 (Postponed from June 9-12, 2020), Athens, Greece

This information is tentative and subject to change. Compiled on April 25, 2024

Abstract

Deep reinforcement learning has already surpassed human performance in many video games, which is mainly achieved with reinforcement learning algorithms based on the actor-critic framework. With the release of PySC2 reinforcement learning environment by Google DeepMind and Blizzard Entertainment, deep reinforcement learning algorithms have attracted the attention of many AI developers on StarCraft II games. In this paper, we used the advantage actor - critic algorithm to achieve the training of agents on seven mini-maps released by DeepMind, we obtain the experimental results and compare them with DeepMind's experimental benchmark. The experimental results show that the agent trained based on the advantage actor-critic algorithm has excellent performance on SC2LE environment.