Practical Issues in Evolving Neural Network

Controllers for Video Game Agents

Kenneth O. Stanley

Abstract: This tutorial will draw upon a significant accumulation of practical knowledge that is rarely explained comprehensively in research publications or textbooks to illuminate how evolutionary algorithms can be combined effectively with neural networks in video games.

Kenneth O. Stanley is an assistant professor in the School of Electrical Engineering and Computer Science at the University of Central Florida. He received a B.S.E. from the University of Pennsylvania in 1997 and received a Ph.D. in 2004 from the University of Texas at Austin. He is an inventor of the Neuroevolution of Augmenting Topologies (NEAT) and HyperNEAT algorithms for evolving complex artificial neural networks. His main research contributions are in neuroevolution (i.e. evolving neural networks), generative and developmental systems, coevolution, machine learning for video games, and interactive evolution. He has won best paper awards for his work on NEAT, NERO, NEAT Drummer, and HyperNEAT. He is an associate editor of IEEE Transactions on Computational Intelligence and AI in Games, the chair of the IEEE Task Force on Computational Intelligence and Video Games, and has chaired the Generative and Developmental Systems track at GECCO for the last three years.

Modeling and Simulation Framework of Real

Urban and Board Games to Train Players

James M. Vaccaro

Abstract: This tutorial will include: (1) a modeling approach for generating an urban terrain model from a Compact Terrain DataBase (CTDB) for computer-simulation of an urban search and rescue operation (US&RO), (2) a modeling approach for implementing the game RISK and generating autonomous players (3) a generalized implementation strategy for integrating both models into an autonomous dynamic planning and execution (ADP&E) framework for gaming simulations, and (4) an evolutionary strategy for using autonomous simulation results to improve player’s abilities. The game RISK a multi-player non-cooperative stochastic game problem, and the US&RO simulation is a single-player multi-agent cooperative game problem. Accumulated results from both applications have given insight into a more general framework to ADP&E for Very Large Partially Observable and Uncertain Environments. The generalization of this approach will be described in terms of its hierarchy, modular components, and dynamic processes.

James M. Vaccaro is a Ph.D. Candidate, at the University of California San Diego (UCSD) and Staff Research Scientist at Lockheed Martin. He has a background in reliability science, modeling and simulation, and dynamic planning technology with a focus on their computational intelligence aspects such as game theory, recurrent systems, pulse-coupled neural networks, Hebbian learning, genetic algorithms, Bayesian networks, reinforcement learning, and hybrid systems. He has worked at a variety of research institutes: Air Force Research Laboratory, UC Berkeley, University of Southern California, Institut fur Neuroinformatik, Bochum, Germany, Centre D’Etudes et de Recherches, Toulouse, France and UCSD. His current interests are in self-aware planning algorithms, where risk, probability, expectations and temporal cost of planning are monitored during execution.