Design history and decisions
Earlier Bela sensing experiments
The earlier work used physical piezo recordings to study vibration and surface roughness. A controlled saved-clip pilot produced useful traces and held-out results, then sustained live P60/P220 testing disagreed with the saved-clip evaluation. That result exposed the need to control mounting, gain, force and movement. Experiment evidence preserves this distinction.
Camera-free V3
V3 established twenty paired pressure/PVDF sensing locations, a soft laminate, labelled wiring and a routed camera-free electronics design. The source includes its actual native board, fabrication files, cut pattern and assembly guide. V4 keeps those sensing and protocol/button dependencies; V3 fabrication outputs identify their own earlier revision.
Camera and two motion references in V4
V4 combines tactile signals with wrist video and two IMUs. A hand IMU and a cuff/camera IMU provide separate motion references across wrist movement. A reference hand mesh is useful for showing contact locations; calibrated joint angles and absolute hidden-hand tracking require additional estimation/calibration.
The camera influences pin allocation, power rails, PSRAM ownership, recording capacity, antenna placement and enclosure size. Those constraints led to the dimensioned 112×104×45.5 mm cuff rather than the smaller appearance concept. Size options explain the tradeoffs.
Bounded acquisition and recoverable faults
Explicit core/driver/queue ownership keeps a slow browser or camera consumer from holding the acquisition buffer. Native tests exercise actual packet output, malformed camera recovery, preview loss accounting and shutdown ordering. The shutdown review records why supply loss, deadlines and acknowledgements needed corrections.
Analogue settling correction
Differential ADC simulation exposed pressure settling error. The current firmware uses 24 μs pressure and 6 μs acoustic settling. Simulation evidence and actual core fixtures connect that circuit model to decoded packets. Electrical stimulation and measured force/texture classification remain separate evidence classes.
Applied USB correction and remaining routing
The reviewed USB power-escape candidate was applied to the canonical board in 097093661f7aa7e29208f5f636a8289161d4f53e. It widens U32 OUT and connects U42 VBUS while retaining the 561-component mapping. The current board has 20 remaining unconnected items. Review preserves the resistance calculation and checks; earlier archives retain their original baseline/candidate status.
Thermal and cost proposals
The heat-path/temperature-readout proposal is separate and unapplied. The value-engineering work preserves the full sensing, camera, recording, protection and six-hour target requirements while testing possible cost reductions. Detailed thermal evidence, proposal and cost decisions retain their assumptions and unresolved bounds.