Motocross The Force, An AITest Platform

1
Motocross The Force,An AI-Test Platform

• Benoit Chaperot, Colin Fyfe,
• School of Computing,
• University of Paisley, SCOTLAND

2
The game

• The game offers a very good environment to
  experiment with advanced Artificial Intelligence
  and data mining techniques
• Most of our research in AI is done in the context
  of this motocross game
• A demo is available for download at
• http//cis.paisley.ac.uk/chap-ci1
• http//www.jstarlab.com

3
The Original Game Architecture

• The game is composed of one executable and some
  data files.

4
Problems with the Original Game Architecture

• The AI source code is in the middle of the game
  engine source code this makes it difficult to
  read and maintain.
• Intellectual property issues only the authors
  can implement AI for the game.
• Only one kind of AI can be used at a time in the
  game. It is not possible to directly compare AI
  techniques.

5
The New Architecture

• The game is made of one executable, some DLLs
  and some data files. DLL stands for Dynamic Link
  Library.

6
Problems solved with the New Game Architecture

• The AI source code is separated from the game
  engine code, with one small DLL project per AI
  easy to implement and maintain.
• The AI part can now be open source. Anyone can
  now implement AI for the game.
• It is possible and easy to directly compare AI
  techniques.

7
Summary
8
What can the AI see ?
9
Illustrative Results 1/2

• Two different kinds of AI have been implemented
  using the new architecture, feed forward multi
  layered perceptrons (MLP), and self organising
  map (SOM).
• The two AIs were trained using the same training
  set and can be compared in real time at racing
  bikes along one of the tracks.

10
Illustrative Results 2/2

• The MLP performs slightly better than the SOM.
• The two AIs behave differently and make
  different decisions and mistakes.

11
Video and Questions ?