Hakilix Care: contactless mobility intelligence

Research problem and rationale

Mobility changes can be difficult to describe between occasional assessments. Hakilix Care is a proposed investigation of whether fixed millimetre-wave radar can characterise movement in everyday indoor spaces without optical cameras or body-worn devices. This website documents research; it is not a monitoring service or clinical decision tool.

Literature background and research gap

Published radar studies demonstrate useful measurement approaches, but staged activities, small samples, room layout and population differences limit generalisation. Fall-event classification and predicting a future fall are different tasks. Neither can be assumed from a gait measurement alone.

Research questions

• Which radar features remain stable across sensor positions, room layouts and movement directions?
• How accurately can candidate models distinguish routine transitions from staged fall-like events in an ethically approved protocol?
• Can within-person mobility summaries remain interpretable when observations are sparse or interrupted?
• How should uncertainty and missing observations be communicated without creating false reassurance?

Proposed architecture

A fixed FMCW mmWave radar would provide range, radial velocity and, where supported, angle estimates. Local processing would suppress static clutter, identify candidate movement segments and extract temporal features. The proposed sensing system is camera-free and wearable-free. No raw signal implies anonymity: motion data still requires access controls and retention limits.

Experimental methodology

Begin with bench measurements and synthetic signals to test implementation. Next, design a controlled protocol for consenting adults, subject to appropriate ethics review. Separate participants and recording sessions across training, validation and held-out testing; random frame splitting could leak participant-specific patterns. Compare simple statistical baselines with compact temporal models before considering more complex architectures.

Data collection and annotation

Document placement, field of view, occlusion, environmental changes and missingness. Any reference measurement or observer annotation requires a separately approved protocol and clear consent. Do not collect real falls from vulnerable people merely to create training data. Annotation disagreement should be reported, not silently resolved. A dataset or study has not yet been established by the evidence available for this website.

Algorithm development and evaluation

Investigate range-Doppler features, transition durations and trajectory consistency. Report event sensitivity, precision, false alarms per observation hour, latency, calibration and confidence intervals. Evaluate unseen participants and rooms, class imbalance, atypical motion and device failure. For mobility estimates, report agreement and error against an appropriate reference; a correlation alone is insufficient.

Ethics, privacy and uncertainty

Participation, withdrawal, data access and retention require explicit governance. Local feature extraction may reduce data transfer but does not remove privacy risk. Consider consent capacity, visitors, multiple occupants, inaccessible interfaces and unintended inference. No diagnosis, treatment recommendation or predicted clinical benefit is asserted.

Potential contributions

A reproducible evaluation protocol, documented failure modes and interpretable mobility features could support future academic discussions. These are intended contributions, not achieved results.

Planned milestones

1. Review literature and define measurable questions.
2. Characterise sensor and synthetic processing behaviour.
3. Seek protocol and ethics review before recruiting participants.
4. Develop baselines and a preregistered evaluation plan.
5. Evaluate, report uncertainty and release appropriate documentation.

Milestones are sequential proposals without committed dates, funding awards or university affiliation.