This project presents an innovative application of the NeuroEvolution of Augmenting Topologies (NEAT) algorithm to enable autonomous gameplay in the popular Flappy Bird game. Leveraging the Pygame library, a game environment is developed, complete with a bird, pipes, and a base. The core contribution of this research lies in the development of a NEAT based AI system that can autonomously control the bird’s actions to achieve optimal gameplay performance.
The NEAT algorithm is used to create and evolve a population of neural networks, each representing an AI-controlled bird. These neural networks receive inputs related to the bird’s position and its proximity to obstacles, and they learn to make decisions on when the bird should jump. The fitness of each bird is continuously evaluated, encouraging behaviours that lead to higher progression in the game.
In the gameplay, we observe that the AI agents exhibit adaptive learning and evolve their strategies over multiple generations. The NEAT algorithm successfully optimizes the neural networks, leading to improved gameplay performance as measured by the distance travelled and the avoidance of obstacles. The results demonstrate the capability of NEAT-based AI to master complex tasks, even in the context of a challenging and dynamic game environment.
This project contributes to the growing field of autonomous reinforcement learning and showcases the potential of NEAT in enabling intelligent decision-making in video games. The findings hold promise for the development of AI systems capable of autonomously mastering real-world tasks and applications, beyond the realm of gaming.