ABSTRACT
The capacity of genetic programming (GP) to evolve a 'hero' character in the Dota 2 video game is investigated. A reinforcement learning context is assumed in which the only input is a 320-dimensional state vector and performance is expressed in terms of kills and net worth. Minimal assumptions are made to initialize the GP game playing agents - evolution from a tabula rasa starting point - implying that: 1) the instruction set is not task specific; 2) end of game performance feedback reflects quantitive properties a player experiences; 3) no attempt is made to impart game specific knowledge into GP, such as heuristics for improving navigation, minimizing partial observability, improving team work or prioritizing the protection of specific strategically important structures. In short, GP has to actively develop its own strategies for all aspects of the game. We are able to demonstrate competitive play with the built in game opponents assuming 1-on-1 competitions using the 'Shadow Fiend' hero. The single most important contributing factor to this result is the provision of external memory to GP. Without this, the resulting Dota 2 bots are not able to identify strategies that match those of the built-in game bot.
- B. Andersson, P. Nordin, and M. Nordahl. 1999. Reactive and memory-based genetic programming for robot control. In European Conference on Genetic Programming (LNCS), Vol. 1598. 161--172. Google ScholarDigital Library
- M. Brameier and W. Banzhaf. 2007. Linear Genetic Programming. Springer. Google ScholarDigital Library
- S. Brave. 1996. The evolution of memory and mental models using genetic programming. In Proceedings of the Annual Conference on Genetic Programming. Google ScholarDigital Library
- J. Cartlidge and S. Bullock. 2004. Combating coevolutionary disengagement by reducing parasite virulence. Evolutionary Computation 12, 2 (2004), 193--222. Google ScholarDigital Library
- J. M. Font and T. Mahlmann. 2019. The Dota 2 Bot Competiton. IEEE Transactions on Games (2019). to appear.Google Scholar
- P. García-Sánchez, A. Tonda, A. M. Mora, G. Squillero, and J. J. Merelo. 2015. Towards automatic StarCraft strategy generation using genetic programming. In IEEE Conference on Computational Intelligence and Games. 284--231.Google Scholar
- F. Haddadi, H. Günes Kayacik, A. N. Zincir-Heywood, and M. I. Heywood. 2013. Malicious Automatically Generated Domain Name Detection Using Stateful-SBB. In EvoApplications (LNCS), Vol. 7835. 529--539. Google ScholarDigital Library
- Max Jaderberg, W. M. Czarnecki, I. Dunning, L. Marris, G. Lever, A. García Castañeda, C. Beattie, N. C. Rabinowitz, A. S. Morcos, A. Ruderman, N. Sonnerat, T. Green, L. Deason, J. Z. Leibo, D. Silver, D. Hassabis, K. Kavukcuoglu, and T. Graepel. 2018. Human-level performance in first-person multiplayer games with population-based deep reinforcement learning. CoRR abs/1807.01281 (2018).Google Scholar
- S. Kelly and M. I. Heywood. 2017. Emergent Tangled Graph Representations for Atari Game Playing Agents. In European Conference on Genetic Programming (LNCS), Vol. 10196. 64--79.Google Scholar
- S. Kelly and M. I. Heywood. 2018. Emergent Solutions to High-Dimensional Multitask Reinforcement Learning. Evolutionary Computation 26, 3 (2018), 347--380. Google ScholarDigital Library
- S. Kelly, Robert J. Smith, and M. I. Heywood. 2019. Emergent policy discovery for visual reinforcement learning through tangled program graphs: A tutorial. In Genetic Programming Theory and Practice, Wolfgang Banzhaf, Lee Spector, and Leigh Sheneman (Eds.). Vol. XVI. Chapter 3, 37--57.Google Scholar
- W. B. Langdon. 1998. Genetic Programming and Data Structures. Kluwer Academic. Google ScholarDigital Library
- P. Lichodzijewski and M. I. Heywood. 2010. Symbiosis, complexification and simplicity under GP. In Proceedings of the ACM Genetic and Evolutionary Computation Conference. 853--860. Google ScholarDigital Library
- S. Ontanón, G. Synnaeve, A. Uriarte, F. Richoux, D. Churchill, and M. Preuss. 2013. A survey of real-time strategy game AI research and competition in StarCraft. IEEE Transactions on Computational Intelligence and AI in Games 5, 4 (2013), 293--311.Google ScholarCross Ref
- A. Sapienza, H. Peng, and E. Ferrara. 2017. Performance dynamics and success in online games. In IEEE International Conference on Data Mining Workshops. 902--909.Google Scholar
- R. J. Smith and M. I. Heywood. 2018. Scaling Tangled Program Graphs to Visual Reinforcement Learning in ViZDoom. In European Conference on Genetic Programming (LNCS), Vol. 10781. 135--150.Google Scholar
- R. J. Smith and M. I. Heywood. 2019. A Model of External Memory for Navigation in Partially Observable Visual Reinforcement Learning Tasks. In European Conference on Genetic Programming (LNCS), Vol. 11451.Google Scholar
- L. Spector and S. Luke. 1996. Cultural Transmission of Information in Genetic Programming. In Annual Conference on Genetic Programming. 209--214. Google ScholarDigital Library
- G. Synnaeve and P. Bessière. 2016. Multiscale Bayesian modeling for RTS games: An application to StarCraft AI. IEEE Transactions on Computational Intelligence and AI in Games 8, 4 (2016), 338--350.Google ScholarCross Ref
- A. Teller. 1994. The evolution of mental models. In Advances in Genetic Programming, K. E. Kinnear (Ed.). MIT Press, 199--220. Google ScholarDigital Library
- M. Čertický and D. Churchill. 2017. The current state of StarCraft AI competitions and bots. In AIIDE Workshop on Artificial Intelligence for Strategy Games. 1--7.Google Scholar
- S. Whiteson, N. Kohl, R. Miikkulainen, and P. Stone. 2005. Evolving soccer keepaway players through task decomposition. Machine Learning 59, 1 (2005), 5--30. Google ScholarDigital Library
Index Terms
- Evolving dota 2 shadow fiend bots using genetic programming with external memory
Recommendations
Why Players use Pings and Annotations in Dota 2
CHI '17: Proceedings of the 2017 CHI Conference on Human Factors in Computing SystemsGroupware research has long focused on representing gestures as a means to facilitate collaboration. However, this work has not led to wide support of gesturing in commercial groupware systems. In contrast, Dota 2, a popular MOBA game, provides two ...
DotA 2 bots win prediction using naive bayes based on adaboost algorithm
ICCIP '17: Proceedings of the 3rd International Conference on Communication and Information ProcessingDotA 2 is a multiplayer game that is widely played today. In DotA 2, the players are divided into two teams, i.e., radiant and dire to against each other. Each team consists of five heroes. One hero is played by human and four heroes are controlled by ...
Game Play Differences by Expertise Level in Dota 2, A Complex Multiplayer Video Game
Dota 2, a complex team based video game, was used to study expertise and attentional allocation in a multiplayer online battle arena MOBA setting. Pre-and post-play survey questions and eye-tracker data were collected from 67 video game players during a ...
Comments