AI for caregiving · Research-grade · Clinically-informed

Rigorous AI for caregiving, built for autism and generalizing across populations.

AutismAI Lab builds AI tools for the people doing the caring. We start with autism, where the research need is deepest, and generalize our methods to elderly care, disability, and any population that benefits from low-burden, ambient support.

Explore our tools Browse research

Lab at a glance

Live research tools

Grovalin and Volitiq, both in active deployment

Proposed: HEARTH program

WAVE (WiFi sensing) + PULSE (rPPG) for in-home health

Zero

Specialist hardware required

Browser-native tools that run on any camera, phone, or laptop

Multimodal

Sensing signals unified

Video, WiFi channel state, and rPPG combined into a single caregiver record

1 in 36

Children diagnosed with ASD in the US

CDC prevalence data (2023), up from 1 in 44 in 2021, underscoring the urgency of scalable tools.^{[Maenner et al., 2023]}

34–80%

Of autistic individuals show motor differences

Motor coordination differences are documented and clinically significant, yet under-measured in ASD research.^{[Fournier et al., 2010]}

Early

Intervention drives outcomes

AI-assisted early screening tools can accelerate diagnosis timelines, enabling earlier targeted intervention.

Our tools

Research-grade apps,
built for the real world

Each tool is grounded in peer-reviewed science, designed to work without specialist hardware, and built around documented methodology that cites the research underpinning it.

Active

Grovalin

grovalin.com · concept-generalization learning

A concept-generalization learning app for children with ASD. Grovalin turns familiar objects from a child's own home into personalized learning across four modules, so a word like "chair" becomes a category, not a single picture — with caregivers verifying every step.

Concept generalization Home objects Caregiver-verified Open-vocabulary CV

Targets the generalization gap — a documented barrier to durable language gains in ASD intervention.

Tager-Flusberg & Kasari, 2013

Learn more →

Active

Volitiq

volitiq.com · motor phenotyping

Objective motor phenotyping in the browser. Volitiq captures webcam video with on-device pose estimation and computes 40+ interpretable kinematic features across seven motor-control domains. Autism-validated methods that generalise to fall risk, rehabilitation, and elderly mobility.

Motor phenotyping 7 domains Browser-native MediaPipe

Addresses the documented rate of motor differences in autism, with methods extensible to fall risk and rehabilitation research.

Fournier et al., 2010 Torres et al., 2013

Learn more →

Proposed

WAVE

HEARTH · ambient monitoring

WiFi Ambient Vitality and Event-detection. Recovers motion, gait, falls, and coarse respiratory rate from the perturbations a body imposes on ordinary WiFi signals, with no camera, no wearable, no microphone. For caregivers across autism and elderly populations.

Ambient sensing Falls & anomalies Device-free 802.11bf-ready

Built on the cross-modal supervision paradigm behind Person-in-WiFi, DensePose-from-WiFi, and recent multi-home CSI validation in older adults (Alzaabi et al., 2025).

Proposed

PULSE

HEARTH · contactless vitals

Passive Unobtrusive Light-based Signal Extraction. Recovers heart rate, HRV, and respiratory rate from an ordinary RGB camera, with no contact sensors. Designed for longitudinal caregiving across both autism (stress, regulation) and aging (cardiorespiratory trends).

Contactless vitals HRV Longitudinal Edge inference

State-of-the-art rPPG methods (PhysFormer, CodePhys, Spiking-PhysFormer) adapted for longitudinal in-home use across the lab's cohorts.

Cross-population research

HEARTH: ambient AI for caregivers, across populations.

Home Environment Ambient sensing for Real-Time Health. A research program for passive, contact-free monitoring that uses the signals already present in the home (WiFi channel perturbations and ambient light reflected from skin) to track the movement, vitals, and behaviour patterns that matter most to caregivers.

The same methods serve two populations the lab cares about: children with autism, where caregivers benefit from longitudinal motor and behaviour signals, and older adults and people with disabilities, where fall detection and gait decline are the leading clinical risks. One technology stack, two populations, shared statistical layer.

In autism caregiving

→Motor development trajectories tracked continuously, surfacing the developmental window where intervention works best.
→Behaviour and regulation signals from HRV and ambient activity, providing early indicators of stress, dysregulation, or escalation.
→Sleep & routine disruption detected without putting a wearable on a child who would not tolerate one.
→Caregiver burden reduction: fewer clinic visits for routine assessment, more attention for what only a human can do.

In elderly & disability care

→Fall detection and gait decline: continuous gait speed and stride variability, the leading clinical risk indicators for older adults.
→Cardiorespiratory trends via PULSE: heart rate and HRV without the cognitive load of remembering a wearable.
→Activities-of-daily-living trends: occupancy and movement patterns that reveal functional decline before a single bad day would.
→Dignity-preserving design: no camera in the bedroom or bathroom, no wearable to forget, no behaviour change asked of the resident.

What our movement analysis can detect

One pipeline, six signals the caregiver needs.

Falls & sudden events

Hard falls, near-falls, and impact events identified in real time, across video and WiFi modalities.

Accidents & urgent attention

Prolonged immobility, atypical postures, or distress signatures that warrant a caregiver looking in.

Unusual & stereotyped movement

Repetitive motor patterns (relevant in autism) and gait irregularities (relevant in aging) flagged from the same kinematic features.

Movement-pattern anomalies

Deviations from the individual's own baseline. The clinical signal that something has changed, not just that it is unusual.

Longitudinal trend change

Gait, motor, and vitals tracked over weeks. A continuous z-score against age-matched norms surfaces decline early.

Need-state signalling

Patterns suggesting hunger, agitation, or unmet need, particularly useful for non-verbal or low-verbal individuals.

Research program in active development

Read the full HEARTH program → Partner with us →

Our principles

Science-first,
evidence-grounded.

Every design decision at AutismAI Lab traces back to peer-reviewed evidence. We publish our methodology, cite the research underpinning every tool, and actively seek validation from the research community.

Browse research

Evidence-grounded methodology

Every tool cites and implements validated clinical research. We document our scientific basis publicly so peers can scrutinise and extend it.

Methodology transparency

Every method we use is documented publicly with citations to the peer-reviewed research that underpins it. Researchers can scrutinise the techniques even where the implementation stays proprietary.

Privacy by design

No PII leaves the device unless explicitly consented. Volitiq runs vision inference locally in the browser. Grovalin anonymises all session data.

Equitable access

Browser-native, hardware-free tools so that families in under-resourced settings access the same quality of assessment as specialist clinics.

Why this lab exists

A small lab, built by a parent and a circle of academic collaborators.

AutismAI Lab was started because the tools families needed (for adaptive learning at home, for objective motor screening, for low-friction in-home monitoring) were not yet deployable from the published literature. It runs as a small team: a lead researcher (PhD candidate in AI at the University of Kansas, and a parent of a child on the spectrum) joined by volunteer collaborators in motor neuroscience, special education, and applied ML.

Meet the people →

Get involved

Build the future of autism AI with us

We are an open collaboration. Researchers, clinicians, developers, and families all contribute meaningfully.

Join the mission

Developers

Contribute features, improve models, and help shape the engineering behind the lab's tools.

Research partners

Co-author studies, validate tools against clinical datasets, shape direction.

Data contributors

Help gather anonymised, ethically-approved research data following IRB guidelines.

Clinicians & families

Test tools, surface real-world needs, and ground our development in practice.

Stay ahead of autism AI research

Browse the active and proposed research lines at the lab: the science behind every shipped tool, plus what we're working toward next.

Explore research →