linyiLYi/street-fighter-ai

Features

• Fighting Game Agent Training - Trains an AI agent to play a fighting game by learning from raw pixel inputs via reinforcement learning.
• Convolutional Feature Extractors - Applies convolutional filters to raw pixel arrays for learning visual game patterns.
• Convolutional Q-Network Implementations - Uses a convolutional neural network to approximate optimal action-value functions from raw pixels.
• Experience Replay Buffers - Stores past game transitions to break temporal correlations and stabilize training via random sampling.