Adaptive Approachs for Optimizing Player Satisfaction

Recently returned from Rome, where I attended Workshop 6 of the SAB'06 conference. This international workshop on optimizing the experiences of computer game players through adaptivity was one of the first of its kind. The emphasis was on technical solutions and methods, which was encouraging because the field is so new, and the subject such a difficult one to define, that a lot of the work so far has been propositional in nature. Indeed it is almost worth thinking of this as the intersection of two fields, that of adaptive games (a subset of adaptive software) and player modelling (a subset of user-centred design). Then the intersection takes on an identity of its own, because all the definitions change when you talk about games as opposed to software, or players as opposed to users.

Thus the submissions can be loosely categorised. Firstly, those that mark ground on that long road toward understanding players, secondly those that provide some in-game adaptivity. Neither area is less important - and even though I haven't time to describe the papers here in detail, I'd like to commend all the authors for their work. I'd also like to thank all the delegates for their insights, and especially thank the organisers Georgios Y. and John H.
Below I've categorised the papers, and even previewed a couple of them!

1. Understanding players
   
   Utilisation of Evolutionary Algorithms to Increase Excitement of the COMMONS Game
   Norio Baba, Hisahi Handa
   
   Using Hierarchical Machine Learning to Improve Player Satisfaction in a Soccer Videogame
   Brian Collins, Michael Rovatsos
   
   Using Decision Theory for Player Analysis in Pacman
   Ben Cowley, Darryl Charles, Michaela Black and Ray Hickey
   
   Making Racing Fun Through Player Modelling and Track Evolution
   Julian Togelius, Renzo De Nardi and Simon Lucas
   This paper showed how to design racing game experiences around the player, so that the tracks themselves would adapt based on player modelling.
   Their idea was clever - train car driving agents on models of the humans performance, so that tracks could be evolved (using the agent) that would match how the player naturally drives. Execution proved a little trickier, as modelling human driving behaviour suffers from the death barrier - if the player crashes, how can the agent learn what the player would rather have done (assuming they didn't want to crash)?
   The solution was to model player performance, which is an elegant workaround although it leaves out some detail. Tracks could then be evolved against agents that wouldn't necessarily drive like a human, but would perform like a human. They were evaluated on fitness functions based on heuristics of what the authors thought made driving games fun, which can be found in their paper.
   
   
2. Adaptivity in games
   
   Smarter Teammates - Applying Hidden Markov Models in Sports Games
   Christian Thurau and Christian Bauckhage
   This paper interested me very much as I felt it could be technically relevant, and I know nothing about the methodology :) Simply put, they used statistical ML to recognise basic player behaviour in a real-time multiplayer football game translated into a canonical reference frame. Defining a set of five behaviour classifiers, they managed to get statistically reliable results in three out of five behaviours. This seemed a promising start, and we hope to see more along this line.
   
   Adaptive Generation of Dilemma-based Interactive Narratives
   Heather Barber and Daniel Kudenko
   
   Where Am I? - On Providing Gamebots with a Sense of Location Using Spectral Clustering of Waypoints
   Christian Bauckhage, Martin Roth and Verena V Hafner
   

Take a look at the proceedings here, and keep an eye out for the workshop next year.
Arrivederci!