ºÚÁÏÍø

Title of the thesis: Learning to interact: simulating users with reinforcement learning in human-computer interaction

Thesis defender: Aleksi Ikkala 
Opponent: Professor Per Ola Kristensson, University of Cambridge, UK
Custos: Professor Perttu Hämäläinen 

Understanding how people interact with devices and interfaces is central to designing technology that is efficient, intuitive, and accessible. Traditionally, researchers have relied on user studies to observe and measure human behavior, but these studies can be time-consuming, expensive, and difficult to scale. User simulation offers a powerful alternative: by creating virtual users that behave like real people, researchers can explore and improve interface designs more efficiently.

This dissertation develops a flexible approach to simulating human behavior in interactive systems. It brings together advanced models of human movement, perception, and cognition with reinforcement learning, a form of artificial intelligence that enables virtual users to learn through interaction. Unlike earlier simulations that often depend on pre-recorded data or rigid rules, these learning-based models can be adapted to a wider range of different types of environments and tasks, providing richer insights into how users perceive, decide, and act.

The research introduces methods for modeling realistic human movement, creating simulated users that can perceive and act within virtual environments, and extending these simulations to virtual reality applications for automated testing. It also explores how simulated users manage multitasking and attention-switching, offering new perspectives on cognitive processes in demanding situations.

By demonstrating how intelligent, learning-driven simulations can replicate key aspects of human behavior, this dissertation opens new opportunities for studying and improving interactive technology. Even though the underlying models of movement, perception, and cognition are still evolving, the results demonstrate that learning-driven simulations could become a practical and valuable substitute for many traditional user studies.

Thesis available for public display 7 days prior to the defence at .