True gimbal attitude from a flight log

Mini-series drones (Mini 3 Pro, Mini 4 Pro without an embedded telemetry track, and similar) record no gimbal attitude anywhere on the SD card. Their SRT sidecars carry position but not where the camera pointed, so the 3D map draws the camera footprint as a labelled estimate.

The attitude does exist — inside the flight record your DJI app or RC keeps. --flight-log merges it into the map:

dji-embed flightmap ./flights --3d --flight-log my-flight.csv

Where a log matches a flight, the camera footprint becomes a measurement and the "estimated" badge disappears on its own. Where it doesn't, nothing changes — the estimate stays, honestly labelled. Repeat the flag for several flights.

Getting the TXT off your controller

Flight records are small .txt files your DJI app or RC writes per flight — named like DJIFlightRecord_2026-07-27_[17-28-49].txt or, on the DJI RC 2 with current DJI Fly, FlightRecord_2026-08-08_[15-43-20].txt (no DJI prefix). Despite the extension they are encrypted binary, not text.

DJI RC (RC 2, RC Pro, ...):

  1. Plug the RC into your computer over USB, with the RC powered on and unlocked. It appears in Explorer as a portable (MTP) device — no drive letter, which is why some copy tools cannot see it.
  2. Open the device › Internal shared storage. Where the records sit depends on the controller generation: older RCs keep them at the root of that storage; the DJI RC 2 keeps them under Android\data\dji.go.v5\files\FlightRecord.
  3. Copy the *FlightRecord_*.txt files anywhere on your computer.

Or let the helper script from the repository do it (Windows; copies only records you do not already have) — it is not shipped in the installed package, so grab it from the repo first:

powershell -ExecutionPolicy Bypass -File tools\mtp-copy.ps1

If the device does not appear: unlock the RC's screen, try another cable/port (some cables are charge-only), and accept any "allow access" prompt on the RC.

Phone as the controller: use the DJI app itself — Profile › More › Flight Data (wording varies by app version) offers a flight-record export/upload; the files it produces are the same DJIFlightRecord_*.txt.

Getting the CSV

DJI encrypts flight records and holds the decryption keys, so decrypting a flight log requires an internet connection whichever tool you use — the key comes from DJI. There is no offline decoder; that is a property of DJI's design, not of any particular product.

  1. Get the record onto your computer (see Getting the TXT off your controller above).
  2. Feed it to any decoder that exports CSV — Airdata, Flight Reader, PhantomHelp Log Viewer, and others all work. This tool is deliberately vendor-neutral: it recognises the content of the export, not one product's schema.
  3. In the decoder's export settings, make sure these are included:
    • gimbal pitch and yaw (in Flight Reader: GIMBAL.pitch and GIMBAL.yaw),
    • a UTC timestamp (in Flight Reader: the UTC option under Logs/Reports; Airdata's datetime(utc) is there by default),
    • the aircraft's latitude/longitude if available — the merge uses them to verify the log really belongs to the flight.

Fetching the CSV via the API (optional)

If you have a Flight Reader API key, fetch-log does the decode step for you:

$env:FLIGHTREADER_API_KEY = "sk_..."   # PowerShell; use  export FLIGHTREADER_API_KEY=sk_...  on macOS/Linux
dji-embed fetch-log DJIFlightRecord_2026-07-27_[17-28-49].txt
dji-embed flightmap D:\Flights --3d --flight-log DJIFlightRecord_2026-07-27_[17-28-49].flightreader.csv

Know what this does before you run it: your entire flight log is uploaded to flightreader.com to decrypt it — your exact coordinates, and everything else the log records, leave your computer (Flight Reader states uploads are deleted immediately after processing). The command asks for confirmation once per invocation before uploading new records, and each record's CSV is written beside it and reused on every later run — the same record is never uploaded twice. Delete the .flightreader.csv to refetch. You supply your own key; this tool never embeds or brokers one.

The fetcher asks the API for exactly the columns the merge needs, including an epoch timestamp — so API-fetched CSVs always get the exact join. The UTC export advice above is for hand-made exports.

How the merge behaves

  • With a UTC column the join is exact: each track point takes the nearest log sample within one second.
  • Without one the UTC offset is inferred from the start times (snapped to 15-minute steps) and the output says so — inference assumes the recording started within a few minutes of the log, so prefer the UTC export option.
  • A log whose GPS track sits far from the flight is refused even when the clocks line up.
  • Gimbal data already present in the SRT always wins; a log fills gaps, it never overwrites.
  • Numbers in the export follow the decimal separator of the machine that produced it (comma or dot) — both parse. A value that parses as neither stops the run with the column and line named, rather than being silently dropped.