Designing Environments for Machine Learning Agents

This class will examine designing environments for visually capable Machine Learning Agents to train in. Looking at both modular, procedurally generated environments and non-generative spaces, the class will cover constructing environments to promote generalization and learning. Additionally, the class will give an overview of the design constraints imposed when training AI, including vision sizing, computing power, and graphic fidelity. The class will dissect the Unity Obstacle Tower Challenge environments, including their construction in Autodesk Maya, the modular structure, and procedural generation. The class will also survey the possible issues in adapting an existing digital environment to use with AI.