This project is a Breakout Game created using the Asynchronous Advantage Actor-Critic (A3C) reinforcement learning algorithm developed by Google DeepMind. The game is trained by an agent that chooses the next state, while the critic evaluates the agent's performance.
A3C stands for Asynchronous Advantage Actor-Critic, which is a deep reinforcement learning algorithm. In this algorithm, a neural network is used to approximate the policy and value functions. The policy function decides which action to take at each step, while the value function estimates the expected return of the current state. N number of Asynchronous Agents has been created to share their experience. The advantage is calculated by the difference between value loss and policy loss to backpropagate it to find the maximum advantage. Then, the maximum advantage is shared with the ball to break the walls and get the maximum score.
Breakout Game is a classic game where the objective is to break the wall of bricks using a ball and a paddle. The game gets more challenging as the ball moves faster and the bricks become harder to break.
- Programming Language: Python
- Frameworks: PyTorch, Gym (Frontend)
This project demonstrates the effectiveness of the A3C reinforcement learning algorithm in training an agent to play a classic video game like Breakout. The agent learns to break the bricks efficiently and achieve the maximum score. The project can be extended to other classic video games or real-world scenarios where reinforcement learning can be applied.