Treadmill fatigue

This dataset contains wearable physiological signals, treadmill data and self-reported perceived exertion recorded from six healthy adult volunteers in two laboratory protocols that induce progressive physical fatigue. It was collected to develop and validate estimators of perceived exertion, exertion state and time to exhaustion for fatigue-aware task allocation in human-robot collaboration (Industry 5.0).

Participants and protocols. Six male adults (25–42 years; pseudonymised IDs 1–6) each completed one session of two protocols (12 sessions). Protocol p1 (stress test) was a treadmill ramp at 2, 4, 5, 6 and 7 km/h in 4-min stages, followed by 1 km/h increments every 2 min until volitional failure. Protocol p2 (load movement) was treadmill walking, without jogging or running, at increasing speed (two cycles of 2, 4 and 5 km/h, or 2, 4, 5 and 6 km/h). Each speed stage was divided into three sub-stages of about 3 min carrying 0, 5 and 8 kg. Sessions include a pre-exercise baseline, the exercise bout, a cool-down and a recovery period.

Sensors. Polar H10 chest strap (heart rate at 1 Hz and beat-to-beat RR intervals); Seeedstudio GSR sensor (skin conductance in raw analogue-to-digital units, about 1 Hz; lower values indicate higher conductance); a secondary photoplethysmographic heart-rate sensor with quality flags; and the treadmill log (speed, distance, elapsed time, energy). Participants rated perceived exertion on the Borg CR10 scale about once per minute; the last rating is held until the next one.

Files and format. 12 semicolon-separated UTF-8 CSV files, one per session, named by protocol and participant. Each row is a sensor event on a common UTC time base; cells are empty when a device did not report at that instant. Variables (original Spanish names):

_time: timestamp, UTC

t_elapsed: elapsed time, s

velocidad: speed, km/h

distancia: distance, m

calorias: treadmill energy estimate, kcal

pulso: treadmill pulse, bpm [confirm]

[load column name]: carried load, kg

polar_hr: heart rate, bpm

polar_rr: RR interval, ms

hr, ibi, qhr, qibi: PPG heart rate, inter-beat interval and quality flags

GSR: skin conductance, raw ADC units

borg: Borg CR10, 0–10

sujeto: participant ID

protocolo: 1 = p1, 2 = p2

[inicio, fin: describe or remove]

The signals are raw and unfiltered and contain artefacts (for example chest-strap contact losses and GSR saturation). The pre-processing, feature extraction and models are described in the related article.

Ethics. All participants gave written informed consent, including consent to the sharing of de-identified data. The study was approved by CSIC Ethics Committee..

Related publication. Haber, R., Alfaro, D., Castaño, F., Villalonga, A., Cruz, Y.J., Mayorga, J. In-context learning and conformal prediction for perceived exertion estimation toward human-robot collaboration. Manuscript under review. The article DOI will be added on publication.

Funding. Horizon Europe project FORTIS (grant agreement 101135707) and CHIPS JU project Moore4Power (grant agreement 101252572).

Terms of use. Released under CC BY-NC 4.0 for academic and other non-commercial research and teaching. Users must cite this dataset and the related article. Users must not attempt to re-identify participants. For commercial use, contact the corresponding author (rodolfo.haber@car.upm-csic.es).

Verbatim fromZenodo· Licence: cc-by-4.0