CR/RL-based Simulated Users with Unity and ML-Agents

Philipp Wintersberger

Abstract: 3D game engines are well-suited to train cognitive user models in realistic and dynamic simulation environments. In this course, we will address the design of bounded internal environments for computationally rational cognitive models and discuss the pros and cons of different belief state approximations, such as discrete or continuous. Together, we will implement and train models of “distracted drivers” that overtake lead vehicles while occasionally interacting with a simulated in-vehicle infotainment system. The course offers hands-on C# and Python programming experience, enabling students to implement, train, and test constrained cognitive models directly in the Unity 3D game engine.

 All speakers  Program