AI vs Human in Super Auto Pets: Who Wins?

AI vs Human in Super Auto Pets: Who Wins?

Table of Contents:

1. Introduction
2. The Concept of Super Auto Pets
3. The Strategic Depth of Super Auto Pets
4. Training an AI to Play Super Auto Pets
5. The Goal: Beating the Average Player
6. Training a Neural Network to Play the Game
7. Rollout, Learning, and Evaluation Phases
8. The Bellman Equation and Deep Q-Learning
9. Tweaky: The AI's Personal Mechanic
10. Model Evaluation and Performance Metrics
11. The Results: Playing the Game with the AI
12. Conclusion

Introduction

Super Auto Pets has become a popular online game that combines cute aesthetics with strategic gameplay. In this article, we will explore the concept of Super Auto Pets and Delve into the strategic depth that makes it so engaging. We will also discuss the process of training an AI system to play the game and aim to surpass the average human player. Join us as we take a deep dive into the world of Super Auto Pets and the fascinating realm of AI gaming.

The Concept of Super Auto Pets

Super Auto Pets is a game where players build teams of pets and send them into battle. The goal is to reduce the opponent's health to zero by strategically deploying pets with different stats and abilities. The game starts with a shop where players can buy pets and food using gold. Pets engage in battles, using their attack stats to deal damage to opponents. Winning battles rewards players with crowns while losing battles results in hearts being lost.

The Strategic Depth of Super Auto Pets

While Super Auto Pets may seem simple at first glance, it actually possesses a surprising amount of strategic depth. The key to success in the game lies in creating a team of pets with high health, strong attack stats, and powerful abilities. Finding powerful combinations of pets and balancing their strengths and weaknesses is crucial for victory. Additionally, players must make sharp decisions that consider both the Current state of the game and potential future developments.

Training an AI to Play Super Auto Pets

The idea of training an AI system to play Super Auto Pets stemmed from the game's strategic depth. The goal was to Create an AI that could outperform the average human player. To achieve this, a deep Q reinforcement learning AI was trained over the course of four months. The AI was trained to make optimal moves Based on the game's state by using a neural network.

The Goal: Beating the Average Player

The ultimate goal of training the AI was to surpass the performance of the average human player in Super Auto Pets. The measure of success was winning more games than losing. If the AI could reach five crowns before losing five hearts, it would prove that excellence in Super Auto Pets could be achieved through AI.

Training a Neural Network to Play the Game

To train the AI, a Python Package developed specifically for Super Auto Pets was utilized. This package allowed for the creation of a game environment in Python, enabling the AI to play thousands of games and perform millions of actions without slowing down the training process. The neural network used for training was a Transformer encoding section, an architecture known for its ability to find connections between information.

Rollout, Learning, and Evaluation Phases

The training process consisted of three key phases: rollout, learning, and evaluation. During the rollout phase, the AI played the game and recorded the results of its actions in an action replay stack. The learning phase combined the information from exploration with the model's predictions to adjust and improve its performance. The evaluation phase tested the AI's proficiency by pitting it against older versions of itself in sparring matches.

The Bellman Equation and Deep Q-Learning

Deep Q-learning played a crucial role in training the AI. The Bellman equation, a simple yet powerful equation, guided the AI in its decision-making process. The equation calculated the Q value for a given state-action pair and incorporated rewards and future Q values to determine the optimal move. Adjustments to the neural network's gears, known as tweaking, were made to ensure the model's predictions aligned with the true outcomes.

Tweaky: The AI's Personal Mechanic

Tweaky was introduced as the AI's personal mechanic, responsible for adjusting the neural network's gears during training. Tweaky compared the before and after Q values and made small adjustments to the network to bring the predictions closer to the true outcomes. Tweaky played a vital role in enhancing the model's prediction capabilities and optimizing its performance.

Model Evaluation and Performance Metrics

Regular model evaluation was conducted to assess the AI's performance. Sparring matches between the current version and older versions of the AI provided insights into its progress and growth. The evaluation process considered win-loss ratios, crowns earned, and other metrics to gauge the AI's proficiency in playing Super Auto Pets.

The Results: Playing the Game with the AI

The culmination of months of training led to the AI's ability to play Super Auto Pets at a competitive level. Ten games were played, aiming for at least five wins to surpass the average human player. The AI displayed impressive strategies, utilizing powerful pet combinations and adapting its gameplay based on the opponents. The challenge was successfully completed, showcasing the potential of AI in gaming.

Conclusion

The Journey of training an AI to play Super Auto Pets has proven the depth and complexity of the game. Through deep Q-learning and continuous improvement, the AI showcased strategies that even experienced human players might overlook. The successful completion of the challenge signifies the potential for AI to excel in gaming, providing a new perspective on strategic gameplay. With the advancement of AI technology, the future of gaming looks promising, offering exciting opportunities for both players and developers.