This was the closing project that I produced for the Udacity Machine Learning Nanodegree and my initially entry into applying deep reinforcement mastering.
The maze condition is converted into a simplified 2nd array of values indicating walls and open up spaces, and the agent. A convolution neural internet is then educated applying reinforcement mastering to come across the swiftest route from the bottom left corner to the middle target for any randomly produced 5×5 maze.
This video reveals the agent’s functionality right after 30 hrs of schooling.
The code for the project is offered here, https://github.com/awbrown90/DeepReinforcementLearning