SmartBall PCB

An Instrumented Baseball for Per-Finger Grip Force Measurement

Reproducibility materials for the MS thesis An Instrumented Baseball for Per-Finger Grip Force Measurement: Finger Force Tracks Delivery Effort Rather Than Release Velocity (Keaton Ruthardt, MS Computer Science — Data Science, Oral Roberts University, 2026).

A regulation baseball instrumented with four thin-film force sensors under the leather cover, sampling all channels at 480 Hz and streaming over Bluetooth Low Energy, paired per pitch with TrackMan radar release velocity.

What the study found

Measured the way the field currently measures it — with a graded submaximal effort ladder — grip force predicts release velocity strongly and broadly: index + middle force at release reaches r = 0.59, and 15 of 20 grip features remain significant after Holm correction.

Holding commanded effort fixed removes that association. Index + middle force at release falls to r = 0.33 (n.s.), and its unique contribution to velocity variance is ΔR² = 0.000, against 0.743 for commanded effort alone. A mixed-effects model agrees (χ²(1) = 0.47, p = 0.49).

Three thumb-dominated features survive Benjamini–Hochberg correction (q = 0.035); none clears family-wise correction, and all are reported as exploratory.

The conclusion is that finger grip force, measured under a submaximal effort protocol, is largely a readout of delivery effort rather than a determinant of release velocity.

Layout

Firmware

• pressure_logger_480/ — deployed acquisition firmware. 480 Hz, four samples per packet.
• pressure_logger/ — earlier 120 Hz single-sample version; the host logger supports both.
• isolation_test/ — channel-isolation gate.
• sensor_test/, single_sensor/, calibrate_sensors/ — bench-check sketches.

Built with the Arduino IDE and the Seeed nRF52 mbed-enabled board package. The non-mbed package fails to link.

Wire protocol

The pressure characteristic carries a 36-byte packet: a 32-bit unsigned microsecond timestamp for the first sample in the batch, followed by four sample groups of four 16-bit unsigned ADC values each, little-endian, in channel order S1 through S4. Sample k within a batch occurs at the batch timestamp plus k × 2083 µs.

The earlier firmware sends a 12-byte packet: a 32-bit millisecond timestamp followed by four 16-bit ADC values. ble_logger.py derives the batch size from the packet length and works against either version without modification.

Data dictionary

Each logged CSV row contains:

Raw counts are the archived source of truth. Calibration is applied downstream in analysis, never on the ball, so it can be revised without re-collecting data.

Force conversion

For a sensor of resistance Rₛ in series with a fixed 10 kΩ resistor, the divider node read by the 12-bit ADC gives a conductance ratio

G = ADC / (4095 - ADC)
F = c * G

where c is the per-sensor calibration factor in calibration/calfactors_v3.csv, fitted through the origin by least squares over twenty applied-mass trials per sensor (R² = 0.961 to 0.995).

Reproducing the analysis

pip install numpy pandas scipy statsmodels matplotlib
cd analysis
python thesis_analysis_v2.py     # feature extraction -> results/
python advanced_stats.py         # corrections, variance decomposition, bootstrap
python make_thesis_figures.py    # figures

A fixed random seed (20260805) makes every reported p-value reproducible: 20,000 permutations and 10,000 bootstrap resamples throughout.

Note that thesis_analysis_v2.py reads raw session captures, which are not distributed here (see below). The derived feature table it produces, analysis/results/per_pitch_features.csv, is included, and advanced_stats.py and make_thesis_figures.py run from it directly.

Participant data

Raw per-pitch session captures are not included, in accordance with the participant-privacy commitments described in the thesis. Participants consented to their data being used in the study; that consent did not extend to public redistribution of raw records.

Derived, de-identified summary data sufficient to reproduce the reported figures and models are provided in analysis/results/. Pitchers are identified only as P1, P2, P3.

Licensing

• Code and firmware — MIT License
• Documentation, figures, and build records — CC BY 4.0

Both are permissive and allow reuse with attribution, which is the intent for a reproducibility release: you should be able to rebuild the instrument and re-run the analysis without seeking further permission.

Third-party software is used in accordance with its own licenses, none of which impose copyleft obligations on this project's code. TrackMan, PitchLogic, Kookaburra, FlexiForce, Tekscan, Seeed, and Nordic Semiconductor are trademarks of their respective owners, used nominatively to identify the products actually used or discussed. No endorsement is implied.

Safety

The instrument places a lithium-polymer cell and rigid electronics inside a ball thrown at speed. If you rebuild it, read docs/BALL_INTEGRATION_GUIDE.md first. Epoxy is poured in ~15 g layers with each cured before the next, because the exotherm of a single large pour is a genuine thermal hazard to an enclosed LiPo cell, and the cell is wrapped before potting. The ball is retired — not repaired — if the cover is breached, the potted block loosens, the cell swells or fails to hold charge, or the channel-isolation test fails after an impact. The cell cannot be removed once potted, so retirement means disposal of the whole assembly through appropriate lithium-battery disposal.

Citation

@mastersthesis{ruthardt2026instrumented,
  author = {Ruthardt, Keaton},
  title  = {An Instrumented Baseball for Per-Finger Grip Force Measurement:
            Finger Force Tracks Delivery Effort Rather Than Release Velocity},
  school = {Oral Roberts University},
  year   = {2026},
  type   = {MS thesis}
}