This project is a reinforcement learning (RL) based battle tank simulation. It uses Q-learning and Neural Networks to train a tank agent to compete in a battle environment. The agent learns from its experiences and improves its performance over time.
The project contains several Java classes that implement different components of the reinforcement learning system:
- Experience.java: Manages the experience replay mechanism used to train the neural network.
- LUT.java: Implements a Lookup Table for Q-learning.
- MyAgent.java: The main agent class that interacts with the environment.
- MyAgent_NN.java: An extension of MyAgent that uses a Neural Network for decision making.
- NeuralNet.java: Implements a simple feedforward neural network.
- RobotState.java: Represents the state of the robot in the environment.
You can run a battle simulation using the main interface. The agent will use either a Lookup Table (LUT) or Neural Network (NN) to make decisions and compete in the battle environment.
src/Experience.java: Contains the logic for experience replay, a crucial part of the reinforcement learning process.src/LUT.java: Implements the Q-learning Lookup Table (LUT) used by the agent.src/MyAgent.java: Defines the agent's behavior using traditional Q-learning.src/MyAgent_NN.java: ExtendsMyAgentto use a Neural Network for decision making.src/NeuralNet.java: Contains the implementation of a simple neural network used byMyAgent_NN.src/RobotState.java: Manages the state of the robot during simulation.src/Interface/: Contains various interfaces used throughout the project, such asLUTInterfaceandNeuralNetInterface.
The project implements a reinforcement learning approach using Q-learning and Neural Networks:
- Q-learning: A model-free RL algorithm that the agent uses to learn the optimal action-value function.
- Neural Networks: Used as function approximators in place of the traditional Lookup Table in Q-learning.