3g1l-vaktros/flight-20260227_100022

Flight 20260227_100022

Drone telemetry dataset captured on a Raspberry Pi 5 with stereo camera, IMU, barometer, and Betaflight flight controller.

Overview

Property	Value
Duration	9.5 minutes
Total frames	12,803
Video chunks	21
Camera	1600x600 @ 60 fps
IMU	LSM6DSL + LIS3MDL @ 100 Hz
Barometer	BMP388 @ 25 Hz
MSP Telemetry	Betaflight @ 5 Hz (9 commands)
Platform	aarch64 / Python 3.13.5

Configs

train (default)

Each row is one video frame with stereo left/right JPEG images and all sensor data aligned to the frame timestamp.

Images: Stereo pair split from 1600x600 into two 800x600 images.

Sensor alignment:

• IMU: Linear interpolation between bracketing 100Hz samples (threshold: 10ms)
• Barometer: Nearest neighbor from 25Hz samples (threshold: 80ms)
• MSP: Nearest neighbor from ~5Hz polling (threshold: 250ms)

Quality flags: imu_ok, baro_ok, msp_ok indicate whether alignment was within threshold.

imu

Raw Pi IMU data at 100Hz: gyroscope (rad/s), accelerometer (m/s^2), magnetometer (uT).

baro

Raw Pi barometer data at 25Hz: pressure (hPa), temperature (C), altitude (m).

msp

Raw Betaflight MSP telemetry at ~45Hz: attitude, GPS, battery, motors, RC channels.

Calibration

{
  "imu": {
    "gyro_bias": [
      0.0,
      0.0,
      0.0
    ],
    "accel_bias": [
      0.0,
      0.0,
      0.0
    ],
    "accel_scale": [
      1.0,
      1.0,
      1.0
    ],
    "mag_bias": [
      0.0,
      0.0,
      0.0
    ],
    "mag_scale": [
      1.0,
      1.0,
      1.0
    ]
  },
  "baro": {
    "sea_level_pressure_hpa": 1013.25
  }
}

Usage

from datasets import load_dataset

# Load aligned frame data (default config)
ds = load_dataset("3g1l-vaktros/flight-20260227_100022")

# Access a sample
sample = ds["train"][0]
left_img = sample["left_image"]   # PIL Image 800x600
right_img = sample["right_image"] # PIL Image 800x600
roll = sample["msp_roll_deg"]
gyro_x = sample["imu_gyro_x_rads"]

# Load raw IMU data
imu = load_dataset("3g1l-vaktros/flight-20260227_100022", "imu")