Fields t, motor1..motor4 (5 cols). Integer PWM-domain command ~109 (idle) to 1999 (ceiling), zero valid — not tachometer RPM.
Present only on the 7 Klagenfurt runs + vibration bundles.
The vibration “RPM (Measured)” labels (v10=3560 … v100=15500) live only in the bundle READMEs; the px4_rpm.csv values there are the same normalized 0–100% scale (297…1999). A separate Wachendorff PT99 tachometer (not in the data) produced the physical RPM.
On indoor_1/indoor_2 the t field is quantized to exact 1.000 s buckets (~99–101 distinct-motor rows per bucket): it cannot time-align individual ~99 Hz samples. indoor_3 and other runs are clean.
The command predicts the IMU vibration-resonance band, through two bench-fitted maps (paper §4.1, Figures 24–25). The PX4 rate value r maps to physical RPM by the quadratic RPM = 168.5541 + 12.1870·r − 0.0023·r², and physical RPM maps to the dominant IMU vibration-resonance frequency by the linear f = 10.6666 + 0.0161·RPM (Hz). The prediction is validated against the in-flight high-rate-IMU spectrum, so the px4_rpm values are a motor-state cue for notch filtering or vibration-aware outlier gating. The high-lever-arm RealSense IMU under-samples this resonance; the ~830–900 Hz LSM IMU resolves it (see vibration coupling in Desirable Ugliness).