What is it about?
Evaluating Augmented Reality (AR) training interfaces usually requires testing them with real users, which can be time-consuming, expensive, and difficult to organise. But what if we could identify potential design problems before involving human participants? In this study, we explored the use of simulated users to evaluate AR training interface designs. Using a paramedic training scenario, we developed computer-generated users that follow different interaction strategies inspired by how novice and expert professionals make decisions. We then tested these simulated users across three different interface layouts to understand how the arrangement and amount of information affect their interactions. Our findings showed that simulated novices experienced greater difficulty with information-dense interfaces, while simulated expert-like users were more affected by ambiguous arrangements of information. These results suggest that simulated users could help designers identify potential usability issues and compare interface alternatives before conducting studies with real people. However, this is an initial proof of concept, and further research is needed to validate the simulated behaviours against human users.
Featured Image
Photo by Martin Splitt on Unsplash
Why is it important?
Designing effective Augmented Reality (AR) training interfaces is particularly challenging in specialised fields such as healthcare, where access to experienced professionals and trainees for repeated testing can be limited. As a result, identifying interface design problems through traditional user studies can be costly and time-consuming. What makes this research different is its approach to simulated users. Rather than simply programming virtual users to complete tasks, we developed agents with different interaction strategies informed by research on how novice and experienced paramedics recognise and prioritise information when making decisions. Our findings suggest that interface designs can affect these simulated users differently depending on their interaction strategies. Information-dense layouts created greater difficulties for novice-like agents, while ambiguous arrangements of information were more challenging for expert-like agents. This highlights why evaluating interfaces using only one type of user behaviour may overlook important design trade-offs. The potential impact is a more efficient approach to early-stage interface evaluation, allowing designers to identify potential problems and refine training interfaces before involving human participants. Although further validation with real users is necessary, this research offers an initial step towards developing more accessible and systematic methods for evaluating AR training interfaces.
Perspectives
This research represents one of the early steps in my PhD journey, where I am exploring how Augmented Reality can support the development of clinical decision-making skills among paramedic students. One of the challenges I encountered early in my research was understanding how to evaluate training interfaces when access to domain experts and students for repeated testing is limited. This led me to explore whether simulated users could help us investigate interface design decisions before involving real participants. What I found particularly interesting was that even relatively simple differences in simulated decision-making strategies could produce noticeably different interaction patterns across interface layouts. It made me think more deeply about an important question: Are we designing interfaces that simply make information accessible, or are we designing interfaces that support how people actually recognise, prioritise, and use information? I see this work as an early exploration rather than a complete solution. There is still much to learn about how closely simulated behaviour reflects real human interaction. Nevertheless, I hope this research encourages others to explore how theories of human cognition and decision-making can inform the design and evaluation of emerging technologies. For me, the most valuable part of this work was the opportunity to bring together human-computer interaction, cognitive science, and artificial intelligence to investigate a practical design challenge.
Thusithanjana Thilakarathna
Auckland University of Technology
Read the Original
This page is a summary of: RPD-Grounded Simulated Users for Pre-screening AR Training Interface Designs: A Unity ML-Agents Approach, January 2026, BCS Learning and Development Limited,
DOI: 10.14236/ewic/britchi26.13.
You can read the full text:
Contributors
The following have contributed to this page







