ModalAI Starling 2 / VOXL 2 Prototyping

Tools for capturing and post-processing IMU + GPS logs from a ModalAI Starling 2 (VOXL 2 compute), with both an on-drone capture path and a host-side parser that produces clean CSVs suitable for analysis.

The video-to-QGC path is intentionally not part of this repo: the stock VOXL SDK already serves an RTSP stream via voxl-streamer and advertises it to QGroundControl through voxl-mavcam-manager. See ModalAI's docs for setup.

Layout

voxl/                 # runs on the drone (VOXL 2)
  scripts/            # B1: bash wrapper around voxl-logger
  logger/             # B2: custom MPA -> CSV subscriber (C)
host/                 # runs on your PC
  pull_logs.sh        # rsync logs off the drone
  parse_logs.py       # parse a flight folder into clean CSVs
  requirements.txt
mock/                 # offline-testing data generator (no drone needed)
flights/              # captured / pulled flights (gitignored)
docs/                 # architecture notes

Two capture paths

B1 — voxl-logger wrapper (works today, IMU only is clean)

Runs voxl-logger on the drone with sensible defaults for IMU + GPS pipes. The IMU output lands as a tidy CSV; the GPS pipe is captured as raw MAVLink and is best decoded by B2.

B2 — Custom MPA subscriber (clean IMU + GPS in one program)

Small C program built against libmodal_pipe that subscribes to imu_apps, imu_px4, and mavlink_gps_raw_int, decodes the GPS MAVLink inline, and writes one clean CSV per stream with controlled schema.

Quick start

On the drone

# B1 — bash wrapper around voxl-logger
voxl/scripts/start_logging.sh -t 60 -n bench_test
# logs land in /data/voxl-logger/logNNNN/

On your PC

# pull a flight
host/pull_logs.sh <drone-ip>

# parse the most recent pulled flight to clean CSVs
python -m pip install -r host/requirements.txt
python host/parse_logs.py flights/log0001
# -> flights/log0001/clean/imu_apps.csv, imu_px4.csv, unified.csv

Without a drone (mock data)

python mock/generate_mock_flight.py flights/mock_log
python host/parse_logs.py flights/mock_log

Output schema

clean/imu_apps.csv, clean/imu_px4.csv:

timestamp_ns, ax_ms2, ay_ms2, az_ms2, gx_rads, gy_rads, gz_rads, temp_c

clean/gps.csv (B2 only, or mock):

timestamp_ns, time_usec, fix_type, lat_deg, lon_deg, alt_m, eph_m, epv_m,
vel_ms, cog_deg, satellites_visible

clean/unified.csv — IMU samples with the latest GPS fix joined per row.

Status

Component	Status
B1 wrapper script	ready
B1 host puller	ready
B1 Python parser (IMU + unified)	ready
Mock data generator	ready
B2 custom C subscriber	ready (needs to be built on the drone)