Project Abstract

In this dissertation, I present a simulation-based survival environment designed for the evolution of autonomous agents competing to navigate through a track. The goal is to develop optimal autonomous vehicles capable of efficiently traversing the given environment. These environments can be created through the use of Microsoft Paint. Performance evaluation is based on the successful passage of checkpoints placed along the track. The agents navigate the track using a neural network that makes a decision and selects the next position to move to. The result of the decision-making process is altered by a physics function that adjusts the agent’s decision based on its current attributes, such as speed and direction. A genetic algorithm is used to evaluate and evolve populations of agents over successive generations. The algorithm identifies the highest-performing agents and combines their traits to create a new generation, iteratively refining the population until convergence to an optimal solution is achieved. This approach harnesses the power of machine learning to iteratively improve the performance of the autonomous vehicles.Visualization plays a crucial role in understanding the evolution of the agents. Images depicting the track environment, along with the routes taken by the agents and the location of checkpoints, are provided for each generation. Additionally, graphs illustrating the average and highest fitness levels of the agents across generations offer clear insights into the progression of agent performance over time.

Keywords: Survival Environment, Machine Learning, Microsoft Paint

 Conference Details

Session: Poster Session B at Poster Stand 114

Location: Sir Stanley Clarke Auditorium at Wednesday 8th 09:00 – 12:30

Markers: Manlio Valenti, Yuanbo Wu

Course: BSc Computer Science, 3rd Year

Future Plans: I’m looking for work