In this project, an AI agent is developed to survive in a simulated environment containing food, predators, and shelters. The agent uses reinforcement learning to optimize decision-making processes, managing internal states such as hunger, health, and attack level while interacting with external elements like environmental features.

Our agent is implemented using a Deep Q-Network (DQN) architecture designed to handle both spatial input and internal agent states. The model processes the grid-based environment through two convolutional layers(CNN), concatenating the output with the agent’s internal state (health, hunger, attack), and optionally incorporates temporal information using a recurrent neural network (RNN) implemented by Long Short-Term Memory, and finally feed into fully connected layers.

For more details, please check out the project report