Developing an Artificial Intelligence Engine

Michael van Lent and John Laird. Developing an Artificial Intelligence Engine

Book: Artificial Intelligence for Games

Ian Millington and John Funge. Artificial Intelligence for Games (2009). Morgan Kaufmann.

Game: Neuro-Evolving Robotic Operatives

Game: Xpilot-AI Genetic Algorithm

Paper: The Painful Face – Pain Expression Recognition Using Active Appearance Models

Ahmed Bilal Ashraf, Simon Lucey, Jeffrey F. Cohn, Tsuhan Chen, Zara Ambadar, Kenneth M. Prkachin, and Patricia E. Solomon. 2009. The painful face - Pain expression recognition using active appearance models. Image Vision Comput. 27, 12 (November 2009), 1788-1796.

Self-reported pain is a big problem because it is difficult to interpret and may be impaired or not even possible, as in young children or the severely ill. The authors of this paper tackle this problem by developing a computer vision system that automatically recognizes acute pain.

To achieve their goal, adult patients with rotator cuff injury were video-recorded while a physiotherapist manipulated their affected and unaffected shoulder. From these ratings, sequences were categorized as no-pain (rating of 0), pain (rating of 3, 4, or 5), and indeterminate (rating of 1 or 2). And a facial expresion detector were implemented, as can be seen in the figure below.

Paper: Human-Level AI's Killer Application: Interactive Computer Games

John E. Laird and Michael van Lent. 2000. Human-Level AI's Killer Application: Interactive Computer Games. In Proceedings of the Seventeenth National Conference on Artificial Intelligence and Twelfth Conference on Innovative Applications of Artificial Intelligence. AAAI Press 1171-1178.

What I most like about this article was the review of the major genres of video games which human-level AI is relevant and all the discussion of how AI could improve these games, and how these games provide research problems for AI.

All this information is collected together in the figure below. 

Paper: A Survey on the Need and Use of AI in Game Agents

Sule Yildirim and Sindre Berg Stene. 2008. A survey on the need and use of AI in game agents. In Proceedings of the 2008 Spring simulation multiconference (SpringSim '08). Society for Computer Simulation International, San Diego, CA, USA, 124-131.

This paper is a good survey on the use of various AI methodologies in commercial video games. They found that the most commonly used AI methodologies to achieve game AI can be stated as follows:

* Decision trees: can be realized by If-Then statements..

* Fine state machines

* Command hierarchies

* Manager task assignment

* Path finding (A*)

* Terrain analysis

* Influence mapping

* Formations

* Flocking

* Emergent behaviour

* Artificial Neural Networks 

* Genetic Algorithms

* Fuzzy logic

One of the most interesting AI approches listed above is emergrnt behavior. This kind of AI has been explored in Black & White.

Also, they present a misture of needed AI depending of the type of game. For example, In RPG games, a team is built up to reach a common goal. The intelligence required from a team or from individual characters depends on how complex it is to reach to the common goal.

Game: Armored Core: Verdict Day

Armored Core, From Software, is a mech fighting war series. The new big feature for Verdict Day is the UNAC system. Like Armored Core V, Verdict Day is about building teams of Core fighters and letting the game’s three warring factions duel it out. Each playthrough is a season, with Red, Blue, and Green teams fighting in skirmishes around the world until only one controls everything. The problem in V according to the producer, is that there weren’t enough people playing the actual game for people to form teams. The UNAC system is meant to be the remedy. Rather than forming a team of fighters with other actual players, you build an AI team.

Full article:
Armored Core: Verdict Day

Paper: Enemy NPC Design Patterns in Shooter Games

Gabriel Rivera, Kenneth Hullett, Jim Whitehead. Enemy NPC Design Patterns in Shooter Games (2012). Proceedings of the 1st Workshop on Design Patterns in Games.

This paper presents design patterns for NPCs in shooter games and explores their effects on gameplay.


Elements of an NPC
- Movement Type:
     * Flanking Intensive
     * Passive
     * Slow Push
     * Rush
     * Cautious
- Movement Range - This is how far the NPC will move during an engagement. This can be Low, Medium, or High.
- Movement Frequency - This is how often the NPC will change their position during an engagement. This can be Low, Medium,
or High.
- Attack Frequency – This describes how often the NPC will initiate an attack. This can be Low, Medium, or High.
- Weapon Type:
     * Sniping Weapon
     * Close Blast
     * Assault
     * Weapon Projectile
     * Power Weapon
     * Melee Weapon
- Weapon Damage – A general indicator on how much damage the NPC will do to the player’s Health, Shields, or Armor. This can
be Low, Medium, or High.
- Armor/Health – This denotes how much damage the NPC can take before being killed. This will typically be linked to how hard
the NPC is to defeat. This can be Low, Medium, or High.
- Motive:
     * Challenge – The degree of difficulty within a combat encounter.
     * Tension – The degree of mental stress the player experiences during a combat encounter.
     * Pacing – The degree of movement that the player will engage in during a combat encounter.

Structure Patterns of NPC Design
     Name - The name of the pattern should describe what the pattern does essentially.
     Description - A brief description of how the pattern is typically used and the effect it has on gameplay. This can be seen as the primary role of the enemy NPC.
     Affordances – This section lists the elements of the NPC and what those elements can be broken down into.
     NPC Relationships – Description of any special relationships with other patterns. This includes an Enemy NPC switching to a different pattern.
     Examples - Some examples from well-known games

Example: How one hospital is using video games to manage kids' pain

http://www.denverpost.com/digitalfirstmedia/ci_22951996/how-one-hospital-is-using-video-games-manage

Thesis: Jonathan

Future Work

Multi-Agent System
- Extend the system to support multiple teams at the same time.
- Allow the system to dynamically change learning and/or prediction agents.
- Implement dynamic unit grouping and Captain creation to allow easy scaling of total units controlled at the same time.

Learning AI
- Implement a prediction AI that can work together with the imitative one in order to create varying and interesting gameplay decisions.
- Extend the system to incorporate co-operative game styles.
- Use a multi neural network approach to allow strategy adaptation and dynamic difficulty adjustment.
- Add a dynamic factory creation/destruction factor to the learning al- gorithm.

Thesis: Pedestrian Simulations with the Social Force Model

Anders Johansson, Kristian Lindgren, and Dirk Helbing. Pedestrian Simulations with the Social Force Model (2004). Dresden University of Technology, Germany.

For this thesis, they implement different simulations using the Social Force Model (Helbing, 1991, 1995, 1997; Helbing and Moln ́ar, 1995; Helbing and Vicsek, 1999; Helbing et al., 2000; Moln ́ar, 1996a, 1996b). Model is discussed, which is suitable for both small-scale pedestrian simulations as well as computation- ally fast large-scale pedestrian simulations.

The model itself consist of some primitives:

- Agent: A pedestrian with a set of properties.

- Obstacle: A wall that will produce forces onto agents in the vicinity.

- ForceF ield: A polygon area with a constant force within. Can be used to model stairs or sloping floors.

- Goal: A location where agents want to go.

The pedestrian movement is built up by different forces:

- Obstacle forces: Repulsive forces are applied to all pedestrians in the vicinity of the obstacle with an increasing magnitude when the distance is decreasing.

- Pedestrian forces: Between all pedestrians there are repulsive forces that increase in magnitude when the distance is decreasing.

Among the applications one can find:

- evacuation of soccer arenas and theaters.

- self-organization phenomena such as lane formation, stripe formation.