AI Player Systems

Generally, the models can be split into two categories:

• Simple: Uses discrete states (like run, fight and idle).

• Deep: Uses continuous states (like health, distance and damage).

Data-To-Action Players (Imitation / Behavior Cloning)

• Creating Simple Data-Based AI Player Model

  • Matches with real players’ general performance.

  • Minimal implementation takes a minimum of 30 minutes using DataPredict™.

• Creating Deep Data-Based AI Player Model

  • Matches with real players’ general performance.

  • Uses real players’ environment data so that the AI players mimic real players.

Data-To-Action-To-Reaction Players (Manual Counter)

• Creating Simple Data-Based Reactionary AI Player Model

  • Minimal implementation takes a minimum of 30 minutes using DataPredict™.

• Creating Deep Data-Based Reactionary AI Player Model

Data-To-Action Optimization Players (Goal-Based Optimization)

• Creating Simple Reward-Maximization-Based AI Player Model

  • May outcompete real players.

• Creating Deep Reward-Maximization-Based AI Player Model

  • May outcompete real players.

  • May exploit bugs and glitches.

Data-To-Action-To-Reaction Optimization Players (Counter Optimization)

• Creating Simple Reward-Maximization-Based Reactionary AI Player Model

  • Breaks mathematical theoretical guarantees due to interference from game designers’ control instead of model’s own actions. Therefore, it is risky to use.

• Creating Deep Reward-Maximization-Based Reactionary AI Player Model

  • Breaks mathematical theoretical guarantees due to interference from game designers’ control instead of model’s own actions. Therefore, it is risky to use.