BeatriXR: Comprehensive and Adaptive Feedforward Support for Guidance in Virtual Reality
Our paper "BeatriXR: Comprehensive and Adaptive Feedforward Support for Guidance in Virtual Reality" (PDF) has been accepted at EICS 2026 and will appear in the EICS issue of Proceedings of the ACM on Human-Computer Interaction. This is work by Valentino Artizzu together with Gustavo Rovelo Ruiz, Lucio Davide Spano, and myself.
Virtual Reality environments are notoriously difficult to learn. The diversity of input devices, interaction modalities, and UI designs creates a steep learning curve, especially for first-time users. Feedforward addresses this directly: it informs users about the expected result of an action before they commit to it, through contextualized previews that make the interaction space legible.
BeatriXR is a an adaptive and smart toolkit for creating direct feedforward in VR. It combines a modular VR toolkit with an LLM-based adaptive decision support layer. Designers can construct, visualize, and customize feedforward using virtual avatars rendered in two forms: an in-world representation and an on-screen comparison of interaction alternatives. The toolkit covers the full feedforward design space across the Triggering, Previewing, and Exiting phases.
The LLM-based layer proposes context-sensitive configuration alternatives. At design time it helps domain experts identify appropriate configurations; at runtime it adapts to the user’s behavior, for example recommending a switch from a partial to a full ghosted avatar when a trainee shows signs of uncertainty. An exploratory review with XR domain experts evaluated both the UI and four LLM models customized for BeatriXR. No single model consistently outperformed the others, suggesting they are interchangeable for this task.
Citation
@article{artizzubeatrixr2026,
author = {Artizzu, Valentino and Luyten, Kris and {Rovelo Ruiz}, Gustavo and Spano, Lucio Davide},
title = {{BeatriXR}: Comprehensive and Adaptive Feedforward Support for Guidance in Virtual Reality},
journal = {Proc. {ACM} Hum. Comput. Interact.},
number = {{EICS}},
year = {2026},
month = {jun},
publisher = {ACM},
address = {New York, NY, USA}
}