May - Jun 2019
Team:
Team:
Omer Zadok - programming, game design and AI
Idan Rooze - Art, Animations and UI
Embedded Files
POGO KIDS! is a competitive game about 2 kids having a jumping war on pogo sticks one afternoon in 1922.
It's a project in deep learning AI, all AI in the game learned how to play from real human players.
This project was completed as part of the course Machine Learning as a Tool for Interactive Products. In this course, beyond exploring standard artificial intelligence concepts, you also learn how to apply these tools in the design of interactive products.
Training data / features
Training data / features
Velocity vector, position, player state (i.e. air / ground)
Labels data
Labels data
Actions: move right, move left, contract string, no operation
A neural network of the following configuration:
A neural network of the following configuration:
Input layer size: 14
Hidden layer size: 128 / 64
# Hidden layers: 2 / 4 / 6
Output layer size: 4
Output:
Probability to each action.
Stochastic world -> multinomial evaluation with random
Activation function: Swish.
Results:
Results:
Summary / What I’ve learned
Summary / What I’ve learned
From this project, I learned that when creating an NPC for computer games, the goal of the artificial intelligence doesn't need to be about making it the strongest. Instead, design thinking emphasizes making the character feel as human as possible. By shifting the AI's goal from being the most powerful and successful to mimicking human behavior, players are more likely to feel they are competing against a fair opponent, ultimately enhancing their enjoyment of the experience.
Google Sites
Report abuse