Beyond heartbeats: re-thinking interoception in autism through virtual reality and computational modelling.

Interoception, the sensing and interpretation of internal bodily signals, is increasingly invoked to explain heterogeneity in autism, yet empirical findings remain inconsistent. A major reason is methodological: the field remains cardio-centric and relies heavily on heartbeat counting tasks and trait questionnaires, which are vulnerable to cognitive strategies, attention, expectancy effects, and confounds such as alexithymia and anxiety. Autism research further indicates that interoception is multidimensional (accuracy, sensibility, metacognitive awareness) and developmentally dynamic, with evidence for dissociations across components and compensation across the lifespan. We argue that progress requires reframing interoception as an inference problem: the brain estimates hidden physiological states from noisy sensory evidence under context-dependent precision weighting (predictive processing/active inference). Building on this framework, we propose a participatory, multimodal research program for autism that integrates (i) lived experience as hypothesis-shaping data, (ii) virtual reality (VR) as an experimental testbed to manipulate embodiment, sensory load, agency cues, and feedback, and (iii) computational physiological modelling to estimate latent cardiovascular-respiratory and autonomic-control parameters, which could subsequently constrain separately specified models of perceptual inference.