An AutismAI Lab application
Grow through what they know.
Grovalin is a concept-generalization learning app for children with autism spectrum disorder — which affects about 1 in 36 children in the US.[Maenner et al., 2023] It turns familiar objects from a child's own home into personalized learning material, so that concepts like "chair" or "cup" stick because they are anchored in what the child already recognises.
Scan to get Grovalin on your iPhone.
Add a second "chair" that looks nothing like the first, and Grovalin links them under one name — teaching, concretely, that all of these different things are still a chair.
The child sees each object exactly as it sits in their home — partial views, odd angles and all — because that is how they will encounter it in real life. No synthetic 3D, no stock photos.
Nothing reaches the child without human approval. Caregivers curate the object library and review every generated variation before it appears in a game.
Why concept generalization
A word can wear many faces.
Children with ASD frequently learn a label tied to a single image. They can identify their cup, but not a different cup; they learn "dog" from one picture book and do not recognise the dog across the street. This breakdown in concept generalization is one of the most documented and most stubborn barriers in early ASD intervention.[Tager-Flusberg & Kasari, 2013]
Grovalin attacks the problem at the material layer. Instead of generic clip-art, the child practises on real objects from their own home, plus caregiver-approved colour and structural variations of each one — so "chair" becomes a category, not a photograph. The approach is grounded in established, evidence-based autism intervention practice.[Wong et al., 2015]
A complete learning ecosystem
Four modules, from first words to daily independence.
Each module is gated to the child's current PECS phase[Bondy & Frost, 1994] and aligned with VB-MAPP milestone domains[Sundberg, 2008] — so the app meets the child where they are and grows with them.
Tap an object from the child's own home to hear its label spoken aloud. Colour and structural variations of the same object build a flexible tact repertoire.
"Tap the chair!" — pick the right object from distractors. Difficulty adapts as the child improves, so the target is generalisation, not memorising one picture.
AI-illustrated stories starring the child's own objects, with narration. Scenes are composed from their personal object library.
Task analyses for daily routines, in three tiers — Watch, Practice, Mastery — again using the objects the child actually lives with.
How personalization works
From a home video to a child's learning library.
A caregiver records short videos of everyday objects. An open-vocabulary computer-vision pipeline detects, segments, verifies and varies them — with the caregiver approving what the child sees. It runs on scale-to-zero cloud GPU inference, so light use costs only a few dollars a month.
Open-vocabulary detection finds objects in a caregiver's photo or video without a fixed class list.
Each object is cleanly cut from its background into a transparent asset the child can interact with.
A vision-language check confirms the crop matches the intended label and flags truncated or ambiguous objects for review.
Colour variations (HSV) and structural variations (diffusion) create many faces of the same word — the heart of generalisation.
Built on Grounding DINO[Liu et al., 2023], SAM 2[Ravi et al., 2024], CLIP[Radford et al., 2021] and rectified-flow image synthesis[Esser et al., 2024]. Read the full methodology →
Data privacy and parental consent
Built for children, controlled by caregivers.
The platform is designed around COPPA-style principles: verifiable parental consent before any data is collected, full caregiver control over the data once it exists, and clean audit trails for every consent action. The very first screen of onboarding is a consent step; a child profile cannot exist without a matching consent record.
One-click export of a complete, machine-readable archive: profile, sessions, uploaded media, extracted objects and variations, and full consent history.
One-click account deletion permanently removes every database record and every stored file tied to the account, and returns a receipt of exactly what was removed.
Withdrawing consent immediately blocks new sessions and stops logging, while preserving existing data so it can be restored later without losing progress.
Consent is checked on the server at every write boundary — session start, activity logging — so a withdrawn caregiver cannot start a child session even if a client tried.
What Grovalin does not claim
A learning tool that aligns with clinical frameworks — not a substitute for them.
Grovalin is a learning and practice tool. It is not a diagnostic instrument and does not replace clinical assessment.
Grovalin is PECS-phase-gated. The card exchange, the human partner, and protocol fidelity stay with the trained therapist or caregiver, not the app.
Grovalin is VB-MAPP aligned. Its indicators are signals from practice data; they are not scored milestones and do not substitute for a licensed VB-MAPP administration.
Every artefact exists to inform the human in the loop. Caregivers approve what the child sees; clinicians interpret what the practice data suggests.
Sources
References
External references are peer-reviewed. The lab's own manuscripts on this work are listed by status on the science page and the Publications page.
- Maenner, M.J., Warren, Z., Williams, A.R., Amoakohene, E., Bakian, A.V., Bilder, D.A., et al. (2023). Prevalence and Characteristics of Autism Spectrum Disorder Among Children Aged 8 Years — Autism and Developmental Disabilities Monitoring Network, 11 Sites, United States, 2020. MMWR Surveillance Summaries 72(2), 1–14. doi:10.15585/mmwr.ss7202a1
- Tager-Flusberg, H., & Kasari, C. (2013). Minimally Verbal School-Aged Children with Autism Spectrum Disorder: The Neglected End of the Spectrum. Autism Research 6(6), 468–478. doi:10.1002/aur.1329
- Wong, C., Odom, S.L., Hume, K.A., Cox, A.W., Fettig, A., Kucharczyk, S., et al. (2015). Evidence-Based Practices for Children, Youth, and Young Adults with Autism Spectrum Disorder: A Comprehensive Review. Journal of Autism and Developmental Disorders 45(7), 1951–1966. doi:10.1007/s10803-014-2351-z
- Bondy, A.S., & Frost, L.A. (1994). The Picture Exchange Communication System. Focus on Autistic Behavior 9(3), 1–19. doi:10.1177/108835769400900301
- Sundberg, M.L. (2008). VB-MAPP: Verbal Behavior Milestones Assessment and Placement Program. AVB Press. https://marksundberg.com/vb-mapp/
- Liu, S., Zeng, Z., Ren, T., Li, F., Zhang, H., Yang, J., et al. (2023). Grounding DINO: Marrying DINO with Grounded Pre-Training for Open-Set Object Detection. arXiv:2303.05499
- Ravi, N., Gabeur, V., Hu, Y.-T., Hu, R., Ryali, C., Ma, T., et al. (2024). SAM 2: Segment Anything in Images and Videos. arXiv:2408.00714
- Radford, A., Kim, J.W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., et al. (2021). Learning Transferable Visual Models From Natural Language Supervision (CLIP). International Conference on Machine Learning (ICML). arXiv:2103.00020
- Esser, P., Kulal, S., Blattmann, A., Entezari, R., Müller, J., Saini, H., et al. (2024). Scaling Rectified Flow Transformers for High-Resolution Image Synthesis. International Conference on Machine Learning (ICML). arXiv:2403.03206
Try Grovalin, or partner with the lab.
Caregivers can sign up and onboard a child in minutes. Clinicians and researchers interested in evaluating Grovalin can reach the team through the collaboration page.