A Reinforcement Learning Approach to Predicting Human Design Actions Using a Data-Driven Reward Formulation

DS 116: Proceedings of the DESIGN2022 17th International Design Conference

Year: 2022
Editor: Mario Štorga, Stanko Škec, Tomislav Martinec, Dorian Marjanović
Author: Molla Hafizur Rahman (1), Alparslan Emrah Bayrak (2), Zhenghui Sha (3)
Series: DESIGN
Institution: 1: University of Arkansas, United States of America; 2: Stevens Institute of Technology, United States of America; 3: The University of Texas at Austin, United States of America
Section: Artificial Intelligence and Data-Driven Design
Page(s): 1709-1718
DOI number: https://doi.org/10.1017/pds.2022.173
ISSN: 2732-527X (Online)

Abstract

In this paper, we develop a design agent based on reinforcement learning to mimic human design behaviours. A data-driven reward mechanism based on the Markov chain model is introduced so that it can reinforce prominent and beneficial design patterns. The method is implemented on a set of data collected from a solar system design problem. The result indicates that the agent provides higher prediction accuracy than the baseline Markov chain model. Several design strategies are also identified that differentiate high-performing designers from low-performing designers.

Keywords: artificial intelligence (AI), human behaviour, design thinking