Unleashing the Power of Genetic AI to Solve Poly Bridge Puzzles

Unleashing the Power of Genetic AI to Solve Poly Bridge Puzzles

Table of Contents

• Introduction
• Building the Bridge: Using Unity Game Engine
• Implementing the Genetic Algorithm
• Programmatically Placing Points and Creating Bridge Pieces
• Implementing the Population Manager
• Genetic Selection Process
• Using Genomes to Create the Bridge
• Calculating Fitness for the Bridge
• Tweaking The Simulation to Match Poly Bridge
• Testing and Result Analysis
• Pros and Cons of Using Genetic Algorithms in Problem Solving

Introduction

In recent times, Poly Bridge 3 has gained significant popularity on YouTube and has become an enjoyable and engaging game. However, given my busy schedule, I found it challenging to invest time in playing and solving the puzzles. Instead, I decided to explore a different approach by creating a program that uses a genetic algorithm to automatically solve the game's challenges and achieve the lowest cost possible. In this article, we will Delve into the process of replicating the bridge building and simulation aspects of Poly Bridge within the Unity game engine. We will also explore the implementation of a genetic algorithm to train an artificial intelligence (AI) agent to solve the game's various puzzles. So, let's get started!

Building the Bridge: Using Unity Game Engine

To begin, we need to replicate the most critical aspect of Poly Bridge - the bridge building and simulation. This can be achieved by leveraging the capabilities of the Unity game engine, which was also used to develop Poly Bridge. The process involves two distinct steps: the paused mode, where You can freely place points and connect roads or supports, and the simulation mode, where you can no longer place anything, and the bridge is simulated with cars driving across its surface.

To accomplish this, we will create a point prefab and a road prefab in Unity. When the left mouse button is pressed, a road will start being made and connect to another point at the end of the Cursor. In order to ensure predictability, we snap the points to the nearest whole number and use code to rotate and Scale the road pieces correctly. Additionally, the simulation aspect involves automatically generating hinge joint 2D components and configuring anchors and connected bodies for each piece. By pressing play, the bridge will be simulated, with the script generating the necessary components and Unity's physics system handling the rest.

Implementing the Genetic Algorithm

Now that we have replicated the bridge building and simulation aspect, we can move on to implementing the genetic algorithm. The genetic algorithm is a methodology that involves a population of agents in each generation. These agents are tested and scored Based on their performance using a fitness function. The best-performing agents are selected as parents to create offspring for the next generation. The genes, represented as a list of numbers, are combined and mutated using different crossover methods.

The genetic algorithm allows us to train an agent to efficiently solve a variety of problems. However, before integrating it with the algorithm, we need to have a way to programmatically place points and create bridge pieces from coordinates. This can be achieved by creating a function that takes two points in Unity's vector2 format and generates a bridge piece accordingly. We also create a test function to build a predefined bridge for easy integration.

Programmatically Placing Points and Creating Bridge Pieces

In order to make the genetic algorithm work seamlessly, we need to programmatically place points and generate bridge pieces based on coordinates. To achieve this, we create a function that takes two points in Unity's vector2 format as input and generates a bridge piece accordingly. Starting from the left anchor, each coordinate determines the position of the next point, connecting the last and Current points with a bridge piece. We also perform checks to ensure that the length of the bridge pieces does not exceed two units. Additionally, we calculate the cost and other fitness-related parameters during the pre-physic simulation phase.

Implementing the Population Manager

To manage the population of agents and facilitate the genetic algorithm, we need to create a population manager. The population manager keeps track of around 100 agents and performs all the necessary genetic algorithm operations to improve their performance. Initially, we instantiate all the agents and their accompanying genome scripts. The genome script, adapted from an existing neural network class, holds the genetic information for each agent. In the fixed update loop, the Elapse functions of all agents are called simultaneously. The population manager monitors the failure state of each agent. Once the loop is completed for a specified number of iterations, the agents are sorted based on their fitness scores. The agent with the best fitness score has its genome saved to a file, allowing for training continuation or future replays.

Genetic Selection Process

The genetic selection process involves selecting the best-performing agents as parents for the next generation. In this project, we employ the tournament selection method. A random number of agents, usually between two and four, are selected for each tournament. The agent with the best fitness score out of the selected group becomes a parent. This process is repeated to find a different group of agents, while the highest-scoring agent becomes the Second parent.

Using Genomes to Create the Bridge

Utilizing the previously established genome script, we can generate bridges based on the genetic information. In this implementation, we use the values from the genome as XY coordinate pairs. Starting from the left anchor, each coordinate dictates the position of the next point, with bridge pieces connecting the last and current points. Additional checks are performed, such as ensuring the length of the bridge components does not exceed two units. We also calculate and accumulate the cost during this process.

Calculating Fitness for the Bridge

To assess the quality of the bridges created by the genetic algorithm, we need to calculate their fitness. Determining the fitness of a bridge is not straightforward, as it requires algorithmic analysis and consideration of various factors. Multiple fitness calculation methods and factors have been developed for this purpose. These factors are then combined to arrive at a final fitness value. Aspects considered for fitness calculation include the distance from the car (represented by a ball) to the destination, the number of unconnected anchors, time-based rewards for staying alive, and the affordability of the bridge based on its component costs.

Tweaking the Simulation to Match Poly Bridge

To ensure that the simulation closely matches the behavior of Poly Bridge, we fine-tune various parameters and values. This involves making adjustments based on the information gathered from Poly Bridge's sandbox mode. By examining the specifications of materials and vehicles used in Poly Bridge, such as mass, cost per unit, and other Relevant characteristics, we can adjust the simulation in Unity to closely Resemble the game. This involves manually tweaking values and performing stress tests to achieve a more accurate simulation.

Testing and Result Analysis

After implementing the genetic algorithm and fine-tuning the simulation, it is time to conduct testing and analyze the results. The program will run simulations based on the genetic algorithm to solve the bridge-building challenges. The output bridges, created by the trained AI agent, will be evaluated based on their fitness and ability to accomplish the task. The results will be compared against the expected performance, and any discrepancies or shortcomings will be noted. This analysis will help determine the effectiveness and limitations of using genetic algorithms in problem solving.

Pros and Cons of Using Genetic Algorithms in Problem Solving

While genetic algorithms offer a unique approach to problem solving, they come with their own set of advantages and disadvantages. Let's explore these:

Pros:

• Genetic algorithms provide an automated and efficient approach to problem solving.
• They have the potential to find creative and Novel solutions by exploring a wide range of possibilities.
• Genetic algorithms can be applied to various problems and domains, making them versatile.

Cons:

• Genetic algorithms often require significant computational resources and time to achieve desirable results.
• The quality of solutions can heavily depend on the fitness function and genetic operators implemented.
• Genetic algorithms may sometimes overlook optimal solutions or exploit loopholes to achieve higher fitness scores rather than desired outcomes.

It's important to consider these pros and cons when deciding to utilize genetic algorithms for problem-solving tasks.

FAQ

Q: Can genetic algorithms be applied to any problem? A: Genetic algorithms can be adapted to a wide range of problems. However, their efficiency and applicability may vary depending on the nature of the problem and the quality of the fitness function and genetic operators used.

Q: How long does it take to train an AI agent using genetic algorithms? A: The training time for an AI agent using genetic algorithms can vary significantly depending on factors such as population size, mutation rate, and the complexity of the problem at hand. Some problems may require hours, while others may take days or even weeks to achieve desirable results.

Q: Are genetic algorithms capable of finding the optimal solution? A: Genetic algorithms are not guaranteed to find the globally optimal solution for a problem. They explore a search space and optimize the fitness of solutions based on the defined criteria. There is always a possibility of getting stuck in local optima or suboptimal solutions.

Q: How can the genetic algorithm be further improved for solving bridge-building challenges? A: The genetic algorithm's performance can be enhanced by refining the fitness function to better evaluate the quality of the bridges. Additionally, exploring different crossover and mutation techniques can lead to improved diversity among the solutions generated by the algorithm.

Q: Can genetic algorithms be combined with other optimization techniques? A: Yes, genetic algorithms can be combined with other optimization techniques to create hybrid approaches. This integration allows for leveraging the strengths of different algorithms to improve overall performance and solution quality.