Volitiq · methodology

The science behind Volitiq.

How ordinary webcam video becomes interpretable motor phenotyping: markerless pose capture, kinematic features grounded in sensorimotor-control research, and transparent statistics you can trace to a formula.

← Back to Volitiq

The signal

Motor signatures in autism

Motor coordination differences in autism are well established: a synthesis and meta-analysis found a large, consistent effect across studies.^{[Fournier et al., 2010]} The "micro-movement" perspective goes further, arguing that fine-grained fluctuations in movement — the kind standard observation discards as noise — carry real information about the nervous system that produces them.^{[Torres et al., 2013]}

Reading that signal is a sensorimotor-control problem. The features that matter — how a movement initiates, how smooth it is, how it varies from trial to trial — are exactly the quantities that motor-control research treats as windows onto how the brain plans and corrects movement.^{[Wolpert et al., 2011]}

Capture

Markerless pose, in the browser

Markerless pose estimation — recovering the positions of body joints directly from ordinary video — is the research direction opened up by systems such as OpenPose.^{[Cao et al., 2019]} Volitiq builds on MediaPipe Tasks Vision,^{[Lugaresi et al., 2019]} whose face-mesh, body-pose and hand-tracking models are efficient enough to run live in a web page, on the device's own GPU, at around 30 frames per second.

Running on-device has two consequences that matter here. First, raw video need never leave the device — only the derived landmark stream and features are stored by default. Second, there is no specialist hardware and no installation, so a caregiver, teacher or researcher can capture usable data with the camera already in front of them.

Features

Seven domains, 40+ features

From the landmark stream, Volitiq computes velocity, acceleration and jerk with central-difference derivatives and Savitzky-Golay smoothing, then derives 40+ named features organised into seven motor-control domains: initiation, smoothness, variability, coordination, rhythm, complexity and efficiency.

Each feature is defined and interpretable rather than latent. Smoothness, for instance, uses established measures including spectral arc length,^{[Balasubramanian et al., 2015]} coordination uses Hilbert-transform phase coupling, and complexity uses approximate entropy and fractal dimension. A researcher can read what every number means and how it was computed.

Analysis

Transparent statistics

Group comparisons use classical, inspectable statistics: Welch's t-test for unequal-variance differences, Cohen's d so effects are reported by magnitude and not only by significance, a transparent feature-importance ranking that combines the two, and FFT-based spectral analysis for rhythm and longitudinal change. Nothing here is a black box.

On machine learning. Volitiq does not ship trained diagnostic classifiers. It is a capture-and-analysis platform; a research team that has validated its own model can apply it to new recordings, but the model, its training data, and its claims remain the researcher's own. Any predictive accuracy figures belong to that team's peer-reviewed work — not to the platform.

Studies

Cohorts and change over time

Studies are organised as user → studies → participants → sessions, with per-study data isolation enforced at the storage layer, on a serverless AWS backend built for population-scale collection. Three dashboards read from that one secure store: longitudinal (per-participant change across sessions), inter-group (cohort comparisons with effect sizes), and intra-group (distribution and correlation within a group).

From this work

Publications

Decoding Motor Signatures in Autism from Markerless Video

Scientific Reports

Under review
Multi-Modal Deep Learning for Automated Detection and Classification of Stimming Behaviors

In preparation
E-MotionSpec: A Web-Based Platform for Real-Time Body Motion Analysis

In preparation

See the full publications list →

Sources

References

Fournier, K.A., Hass, C.J., Naik, S.K., Lodha, N., & Cauraugh, J.H. (2010). Motor Coordination in Autism Spectrum Disorders: A Synthesis and Meta-Analysis. Journal of Autism and Developmental Disorders 40(10), 1227–1240. doi:10.1007/s10803-010-0981-3
Torres, E.B., Brincker, M., Isenhower, R.W., Yanovich, P., Stigler, K.A., Nurnberger, J.I., et al. (2013). Autism: The Micro-Movement Perspective. Frontiers in Integrative Neuroscience 7, 32. doi:10.3389/fnint.2013.00032
Wolpert, D.M., Diedrichsen, J., & Flanagan, J.R. (2011). Principles of Sensorimotor Learning. Nature Reviews Neuroscience 12(12), 739–751. doi:10.1038/nrn3112
Balasubramanian, S., Melendez-Calderon, A., Roby-Brami, A., & Burdet, E. (2015). On the Analysis of Movement Smoothness. Journal of NeuroEngineering and Rehabilitation 12, 112. doi:10.1186/s12984-015-0090-9
Lugaresi, C., Tang, J., Nash, H., McClanahan, C., Uboweja, E., Hays, M., et al. (2019). MediaPipe: A Framework for Building Perception Pipelines. arXiv:1906.08172
Cao, Z., Hidalgo, G., Simon, T., Wei, S.-E., & Sheikh, Y. (2019). OpenPose: Realtime Multi-Person 2D Pose Estimation Using Part Affinity Fields. IEEE Transactions on Pattern Analysis and Machine Intelligence 43(1), 172–186. doi:10.1109/TPAMI.2019.2929257