Rust Reinforcment Learning

Overview

• Simple Maze Grid
• Move the goal after collection
• Train off of the path after collecting the goal
• Nerual Network for Agent
• Memory Replay
• Train off of the memory which contains (s, a, s, r)
• Negative reward in each state to incentivize faster paths, positive points when reaching goal
• Max amount of actions within a trajectory
• Beable to train in batches for faster improvment

Neural Network

• x, y cordinate for the goal and player
• 1 hidden layer with 6 nodes
• 4 outputs representing probability of each action
• Train it based on its path and accosiated gains, more likly to do good rewards given a certain state

File Structure

• Nerual Network Struct
• Map Enum
• Memory Enum