Deep Learning Bot for League of Legends

  • Aishwarya Lohokare University of Southern California
  • Aayush Shah University of Southern California
  • Michael Zyda University of Southern California


In this paper, we take the first step towards building comprehensive bots capable of playing a simplified version of League of Legends, a popular 5v5 online multiplayer game. We implement two different agents, one using Reinforcement Learning and the other via Supervised Long Short Term Memory Network. League of Legends provides a partially observable game environment with an action space much larger than games like Chess or Go. Source code and demonstrations can be found at