AI and Non-Player Characters

This event was held on the 14th of November in Essex. (with optional dinner on evening of November 13th)

The purpose of this event was to provide a lively forum for discussion on the state of the art for Non-Player Character (aka Bot) Artificial Intelligence, and to gain a clearer picture of the main challenges and opportunities facing NPC AI in various game genres. The aim was to foster the development of research grant proposals and greater collaboration between industry and academia.

The morning session revolved around a set of short talks, while the afternoon allowed time for group discussion.

Discussion topics

Discussion topics included:

* Emotional Modelling for Synthetic Characters

* Flexible, dynamic planning

* AI for MMORPGs

* Fast learning algorithms NPCs

* NPC AI for Player Satisfaction

* Cooperation in NPC Teams

* Plug-n-play NPCs (and APIs to support this)


Currently confirmed speakers are:

  • Samad Ahmadi, De Montfort University
  • Aladdin Ayesh, De Montfort University
  • Ruth Aylett, Heriot Watt University
  • Richard Bartle, University of Essex
  • Simon Colton, Imperial College
  • Peter Cowling, University of Bradford
  • Udo Kruschwitz, University of Essex
  • John Levine, University of Strathclyde
  • Simon Lucas, University of Essex
  • Daniela Romano, University of Sheffield
  • Kostas Stathis, Royal Holloway College

In this talk we plan to discuss the similarity between developing computer games and multi-agent systems. Our discussion is based on a logic-based framework called GOLEM, which supports the development of an agent environment as a composite structure that evolves over time. We show how such a complex structure involves the interaction between three main classes of entities: agents (players), objects (game objects), and containers of agents and objects (game environments). Interactions between these entities are specified in terms of event occurrences governed by a set of physical laws describing all the possible evolutions of the game environment, including how these evolutions are perceived by players and affect game objects and containers in a game application.

To illustrate how the GOLEM platform can support interaction of a simple game, we will show how players (including humans), game objects, and the game environments can be deployed and distributed over a network for a game application. We will also report on existing experimentation that allows thousands of players and objects can be distributed over thousands of containers, as well as the implication of these experiments in networked games.

The talk concerned the use of a tried-and-trusted AI technique (Planning) as a solution to a problem for Massively Multiplayer Online Role-Playing Games (quest generation). It concluded that while it's good that the AI community is producing new tools and techniques for the games industry, there's still plenty to be done in selling older work to it.

Movement of NPC groups in realtime strategy games is often a nuisance: Units travel and battle separately, resulting in huge losses and the AI looking dumb. This applies to computer as well as human commanded factions. Unit selection for attack/defense tasks and building is currently mostly skripted, randomized, or follows very simple rules. We suggest to tackle the unit movement problem by means of a combination of flocking and influence maps, and the unit selection problem via learning self-organizing maps (SOM's). Experimental results also demonstrate that the two approaches complement each other well.

Giving NPCs more emotionally expressive behaviour is one way of increasing the engagement of players in a game. This talk argues that doing so is not just an issue of animation but requires an ability to model the NPCs affective state in context. Cognitive appraisal theories have advantages as a way of implementing such capabilities and can be linked to autonomous action selection producing expressive actions as well as expressive behaviour.

This talk briefly introduces the real-time strategy game DEFCON from Introversion Software, followed by a description of the AI created for a Masters project at Imperial College, using case-based reasoning and decision-trees. After that, an application programming interface for DEFCON is introduced, which allows researchers and hobbyists to create their own bots for the game.

AI & NPC Workshop Report

Found Here - by Andrew Armstrong