Real-Time Neuroevolution to Imitate a Game Player
realtime neuroevolution for games

Game users move their units back generally when the HP of these units is low; however such control is tedious labor. This paper presents an algorithm to imitate a game user's play patterns using real-time neuroevolution (neural network + genetic algorithm). Learning and imitation processes can be performed during gameplay. To test effectiveness of the algorithm, we made an application similar to the Starcraft, which is a popular real-time strategy game. By using our method, game users can avoid tediously repeating labor to control their units, and we can also make "human-like" agents easily.

Download:
- Paper (revision; not yet)
- Video (.mp4)
Note that in this demo only blue marines (top) were trained. Although the user does not control the blue marines, they battle efficiently as the user do.



Selected Publication:
Real-Time Neuroevolution to Imitate a Game Player Hyunwoo Ki, Jihye Lyu, and Kyoungsu Oh. "Real-Time Neuroevolution to Imitate a Game Player". Lecture Notes in Computer Science (Proc. International Conference on E-learning and Games), Volume 3942, Mar 2006, pp. 658-668.

Related Award (Domestic):
The Bronze Prize Hyunwoo Ki, and Jihye Lyu. "The Bronze Prize". The 2005 Software Contest of the Colleague of the Information & Science, Soongsil University. September 2005.

BibTeX:
@ARTICLE{KiH06NE,
author = {Ki, H. and Lyu, J. and Oh, K.},
title = {Real-Time Neuroevolution to Imitate a Game Player},
journal = {Lecture Notes in Computer Science (Proc. International Conference on E-learning and Games 2006)},
year = {2006},
volume = {3942},
pages = {658-668},
month = {March}, }

Acknowledgement:
- Starcraft
- Korea Research Foundation Grant (KRF-2004-005-D00198)