The default move in sensory-friendly design is subtraction. Dim the lights, cut the audio, remove the haptics, slow everything down. It is a reasonable instinct and it is not quite what the evidence here supports.
A paper accepted to ACM ISS 2026 (PACMHCI, Article ISS030) by Yibo Meng, Xiujun Chen, Bingyi Liu and Xiaolan Ding asks whether multisensory XR can support informal learning for autistic adults — and tests it at the level of whole configurations rather than individual channels.
The design
The study is mixed-methods, counterbalanced, within-subjects, with 20 Chinese autistic adults. Each participant experienced five complete XR configurations presenting the same popular-science content, then rated them.
Holding content constant while varying the sensory configuration is the design choice that makes the result readable. Any difference in reported comfort is attributable to how the material was delivered, not to what it was about.
The result
Comfort differed significantly across the five configurations:
| Configuration | Mean comfort (of 5) |
|---|---|
| L3 | 2.97 — lowest |
| L5 | 4.43 — highest |
That is a spread of nearly a point and a half on a five-point scale, from the same content.
The crucial point is the direction. L5 was the highest-rated, and L4 and L5 were both credited positively — these are not the minimal configurations. If comfort were simply a function of reduced stimulation, the ranking would run monotonically downward as sensory channels were added. It does not.
What participants attributed it to
The qualitative data is where the design guidance is, and the reasons participants gave are structural rather than quantitative.
L3’s discomfort was attributed to:
- abrupt changes — transitions arriving without lead-in
- simultaneous demands — multiple channels requiring attention at once
- insufficient warning — no signal that something was about to change
- limited ability to disengage — no way out while it was happening
L4 and L5 were credited with:
- gradual transitions
- opportunities to disengage
- coordinated narrative cues
Every item on the first list is about timing, concurrency and control. Not one is about intensity. And the third bullet on the second list — coordinated narrative cues — suggests that multiple sensory channels are comfortable when they are telling the same story at the same time, and uncomfortable when they compete.
That reframes the design problem. The question is not “how much stimulation?” but “is the stimulation predictable, co-ordinated, and escapable?”
What this means for building things
Three implications follow fairly directly for anyone building an installation or an XR experience:
- Warn before you transition. The absence of a pre-signal was named explicitly as a cause of discomfort. A lead-in cue costs almost nothing.
- Give people an exit that is not removing the headset. “Limited ability to disengage” was a complaint about the experience, not the hardware. A reliable, discoverable way to step back changes the rating.
- Co-ordinate channels around the narrative. Audio, visuals and haptics pointing at the same moment appear to be additive; the same channels arriving independently appear to be a cost.
The limits
The authors position this as a configuration-level account of sensory comfort intended to guide future controlled studies — not as a finished causal model. Twenty participants, all recruited in China, rating five configurations of one piece of content: the comfort ordering is specific to those configurations, and the paper does not claim it generalises to arbitrary XR.
What it does usefully puncture is the assumption that accessible means attenuated. The most-stimulating configuration tested was also the most comfortable, because it was the best organised.
The paper is listed as a revised, accepted manuscript with 4 figures, submitted to arXiv on 16 October 2026, with a related DOI at 10.1145/3859342.
Sources
- Sensory Comfort Across Multisensory XR Prototypes for Autistic Adults — arXiv 2610.23833
- DOI 10.1145/3859342 — ACM
- ACM Interactive Surfaces and Spaces (ISS)
- Proceedings of the ACM on Human-Computer Interaction