Supervised by Prof Colin Fyfe since 2005
Computer Science, Video Games and Artificial Intelligence.
Video Games, Rigid Body Simulation, Optimisation, 3D engines, Motocrossing, Artificial Neural Networks, Genetic Algorithms.
Motocross The Force is a motocross game featuring terrain rendering and rigid body simulation applied on bikes and characters. The game has been developed and is still being developed in conjunction with Eric Breistroffer (2D and 3D artist, animator) and Thierry Malet (2D and 3D artist).
The two videos below are examples of the game in action; in these videos some motorbikes controlled by ANN's are ridden along a long track. The ANN's are presented with many different obstacles and situations. The ANN can be trained using the back propagation algorithm, then optimised for any particular track using an optimisation technique such as the Genetic Algorithm or another optimisation technique. The training data for the back propagation algorithm is the recording of the main author playing the game on various tracks.
In these two videos, the back propagation algorithm is not used. In the first video, the ANN's are initialised with random weights; hence the bad drunk behaviours. In the second video, the same ANN's have been optimised for a long time (5832 hours) using our New Evolution Technique inspired by the Genetic Algorithm and Reinforcement Learning. These particular ANN's can perform better than a good human player at riding motorbikes on this particular long track.At GDTW 2006, I presented a paper in which we explain changes made to the motocross game to allow other researchers to develop and experiment their own artificial intelligence techniques in the game. The game is split between an executable and AI DLL's.
The latest free version of the game for research (0.964R, 21MB) and the corresponding AI SDK (0.964R, 12/05/2010, 1MB) can be found here and here.
An older version of the game (0.951) and its corresponding AI SDK can be found here and here.
Three good training sets: 1, 2 and 3 (to copy to the trainingset folder).
One good AI here (for track long, to copy to the AI folder).
Part of the source code is inspired by ai-junkie; some of the AI DLL projects make use of the LUA library for reading parameters from files, and the DirectX SDK for matrix and vector operations.
Feel free to contact me if you have any question or comment about the AI SDK.
"Computational Intelligence applied to Computer Games"
B. Chaperot, supervised by C. Fyfe, University of the West of Scotland, June 2010
Thesis
"Motocross And Artificial Neural Networks"
B. Chaperot, C.Fyfe , Game Design And Technology Workshop 2005, Liverpool, UK, November 2005
Paper
Presentation
"Issues in the Physics of a Motocross Simulation"
B. Chaperot, C.Fyfe , Game-On 2005, Leicester, UK, November 2005
Paper
Presentation
"Improving Artificial Intelligence in a Motocross Game"
B. Chaperot, C. Fyfe, IEEE Symposium on Computational Intelligence and Games, CIG06, USA, May 2006
Paper
Presentation
"Creating an AI-Test Platform"
B. Chaperot, C.Fyfe , Game Design And Technology Workshop 2006, Liverpool, UK, November 2006
Paper Presentation
"Topographic Products of Experts applied to a Motocross Simulation and Simulation Stabilisation"
B. Chaperot, C.Fyfe , Game Design And Technology Workshop 2007, Liverpool, UK, November 2007
Paper Presentation
"Advanced artificial intelligence techniques applied to a motocross game"
B. Chaperot, C. Fyfe, Paisley University Computing and Information Systems journal, May 2006
Paper
"Anomaly Detection in Magnetic Motion Capture using a 2-Layer SOM network"
Iain Miller, Stephen McGlinchey, B. Chaperot, IEEE Symposium on Computational Intelligence and Games, CIG06, USA, May 2006
Paper