My Role
: Producer and UX Designer
Collaborators
: UX Designer (1 No.s), 3D Artist (1 No.s), Developer (1 No.s), Sound Artist (1 No.s)
My Responsibility
Organization
: Carnegie Mellon University
Duration
: 4 Weeks
Imitation game is a VR game that blends classic Pong gameplay with the latest AI advancements. It’s an exciting exploration of how we can incorporate machine learning into game development and create unique AI behaviors. The game gives both players and AI equal control, and the AI development process used imitation learning with reinforcement learning techniques to create non-traditional NPC behaviors. The game showcases the potential of AI in game design and how it can be used for creating NPC behaviors, generating art assets, and even unique game features.
Introduction
Imitation Game is virtual reality game challenges players to rally a ball across a hallway and score against a computer opponent. The purpose of the game is to hit the wall behind the opponent while defending one’s own side.
Agent Design
The agents in the game are designed using a demonstration dataset provided by the game. Utilizing Machine Learning, the agents develop their own playstyles and improve their performance through a rewards and punishment system.
Gameplay Features
Target Platform
The game is designed to be compatible with VR Headsets such as Quest and Oculus.
VISUAL INSPIRATION
The game pays homage to the classic 2D Pong games of the past, offering players the opportunity to test their skills against a formidable AI opponent.
The game's visual aesthetics are inspired by the gritty atmosphere of a training gym, evoking the iconic look and feel of the classic Street Fighter game, Street Fighter. This serves as the backdrop for the AI agents to hone their skills and develop their own unique play styles.
It aims to immerse the player in a nostalgic and challenging virtual reality experience.
Start game:
Lose a life:
5. Starts to play > 6. Loses a life > 5. Start play position.
Loses 5 lives:
5. Starts to play > 6. Loses a life > 7. Loses all 5 lives.
Beats the agent:
5. Start to play > 8. Beats agent > 9. Beats agent 5 times > 3. Select Agent.
The repetitive pattern of vertical and horizontal elements in the hallway design creates an exaggerated sense of depth, adding to the immersive experience in VR environments.
The initial sketches illustrate the concept of repetitive elements in the design.
A stadium environment has been added as the outside world, creating a feeling of being in a Pong arena. The audience are depicted as agents who dynamically cheer and support the opponent NPC, adding to the immersive experience.
The game’s UI incorporates both non-diegetic and diegetic elements to enhance the player experience. Our research shows that players are familiar with non-diegetic UI commonly used in previous games, therefore critical information such as lives, health, and score are displayed in a clear non-diegetic manner. On the other hand, elements such as the entry and main menu have been designed as diegetic UI, integrating them into the in-game environment. This approach creates a seamless and immersive experience for the player.
The design of the agents is intentionally kept simple, with a focus on retro colors in dark hues. The agents have three distinct expressions, representing their default state, when they lose a life, and when they score a point.
"AImee" is designed as the easiest agent, trained with limited rotation and low speeds. Her design reflects this with a large, blocky gray square shape.
“GAIl” is intended as a medium-difficulty agent, trained with less restricted rotation and moderate speeds. This training is reflected in his round shape and expressive features.
“DAImIAn” is the hardest agent, trained with fast speeds and limited rotation, rewarded for hitting corners. His design reflects this challenge with a sharp, red color and angular features.
This feature was added in response to user testers' desire for more interaction and customization. To vary the difficulty, three different obstacle types have been added:
TRAINING SESSSIONS – training the agents & obstacles
Parallel ML training sessions for the three agents