yenchenlin/DeepLearningFlappyBird
DeepLearningFlappyBird
This project is a deep reinforcement learning agent and AI game automation tool designed to master game objectives by analyzing visual input. It implements a Deep Q-Network to train an autonomous bot that learns to play a video game by maximizing rewards through deep Q-learning.
The system utilizes a convolutional neural network to process raw pixel data from game frames, identifying patterns to determine optimal real-time actions. Training is stabilized through the use of an experience replay buffer and an epsilon-greedy action selection strategy to balance exploration and exploitation.
The implementation covers a full reinforcement learning workflow, including frame-stacking preprocessing to provide temporal motion data, pixel-based state representation, and reward-penalty signal mapping.
Features
- Autonomous Game Agents - Creates a self-learning system that plays a video game by analyzing visual input in real-time.
- Deep Learning Game AIs - Implements a self-learning agent that masters game objectives through deep neural network training and visual input analysis.
- Game-Playing Agents - Implements a self-learning system that masters game objectives by processing visual inputs.
- Convolutional Neural Networks - Uses a convolutional neural network to process pixel data and identify patterns for action selection.
- Convolutional Q-Network Implementations - Implements a convolutional Q-network to approximate action-value functions from raw pixel inputs.
- Deep Q-Learning Implementations - Implements a full deep Q-learning workflow including experience replay and action selection.
- Deep Reinforcement Learning Implementations - Implements a deep reinforcement learning agent that masters game objectives through neural networks.
- Pixel-Based RL Training - Trains a convolutional neural network to select optimal actions based on raw image observations.
- Pixel-Based Game Agent Training - Processes raw image frames into downsampled grayscale arrays for pixel-based agent training.
- Pixel-Processing Agents - Ships a system that processes raw pixel data from game frames to determine optimal actions.
- Epsilon-Greedy Exploration Strategies - Uses an epsilon-greedy strategy to balance exploration of new moves and exploitation of known rewards.
- Experience Replay Buffers - Implements an experience replay buffer to store transitions and break temporal correlations during training.
- Game AI - Develops game AI logic that enables a bot to make real-time decisions based on visual input.
- Reinforcement Learning Reward Systems - Maps game events to numerical reward and penalty signals to guide the agent's learning.
- Game Automation Bots - Implements a game automation bot that uses convolutional layers to determine actions from raw pixels.
- Frame Stacking Preprocessors - Stacks multiple consecutive frames into a single input tensor to capture temporal motion data.
