Game Programming - Class Three

Notes on AI

• Intelligence exhibited by an artificial (non-natural, man-made) entity

• An algorithm or set of algorithms that can make decisions in a logical way.

• The branch of computer science dealing with the reproduction or mimicking of human-level thought in computers

• The study and design of intelligent agents i.e systems that perceive their environment and take actions which maximize the chances of success

• Usually involves sensing, reasoning and acting (and in some cases communicating)

• Such factors should be in levels similar to the player's (no all-knowing, super-mighty gods)

• Can be partitioned in different roles or moods, that can change for the same player

• Main goal - make it look human-like, believable: powerful, yet flawed.

Some flaw examples:

• Sensing may have limitations such as occlusion, low range, uncertainty
• Reasoning may have errors e.g inaccurate calculations
• Communicating may suffer from missing or garbled messages
• Acting may be limited/influenced by factors such as low energy, injuries, etc.

• Sensing: A player can see the ball and the players around him in a relatively close range in the direction he is facing
• Reasoning: Think of a series of rules to guide the reasoning
  • “If my team does not have the ball and I am close to it, I should get it”
  • “If I have the ball and a clear shot for the goal, kick”
  • “If I have the ball, but I do not have a clear shot, I should pass to another player in better conditions”
  • …
• Acting: Kick the ball. May be limited by the running speed and direction, the energy/fatigue, etc.

• More advanced: different roles to different players, inter-player communication, etc.

The AI player is not on the field, but controlling the knobs

• Sensing: It can see the whole field and the ball
• Reasoning:
  • “If I don't have the ball, I should defend”
  • “If I have the ball and a clear shot to the goal, kick”
  • “If I have the ball, but not a clear shot, pass the ball to a player in better position”
  • …
• Acting: Rotating the knobs is limited e.g. by the time to move hands from one knob to another.

• Wrap AI in an UpdateAI method in the relevant component(s), that is called from the Update method
• Sensing requires knowledge of the whole world and the generation of an (altered) subset of it.
  • The Game class should help in providing this to specific components
  • Avoid making tightly-coupled components
• Reasoning should take into account if the character is already performing an action. Don't jump between actions
• Actions may take some time. Make actions smooth whenever possible, even when you have to interrupt them.

• Be aware of the whole
• Decompose problems
• Think of alternative solutions of different complexity
• Prioritize and select what to implement and in which order
• Set milestones