I was flying two batteries yesterday and when the second one was at 20% I flew back to the home point. 20m away and 2m above I slowly (1-2km/h) moved the drone closer pointing directly to me.
All of the sudden the Avata accelerated hard to its right and hit a wall.
In the screenshot you can see the drone pointing to me and then a line to its right.
12m37.3 and before is the forward flight with 2km/h.
12m38.5 it flew with 12km/h to the right.

Logs show no RC input, only in the last second I tried to correct with stick left but it was too late and I did not notice any change.
Sports mode.
Flight with the RC FPV Controller.
All devices on the latest firmware.
Drone was recording the video but the last frame on the MP4 file was seconds before it happened.
Never had any problems before.
Full details and flight log download: https://app.airdata.com/share/TQkmDg

It hit the surrounding, has hit marks on the outside of one propeller guard, barely visible scratch marks on the inner side of all prop guards. Props look fine. Back home I started the drone again to check and the props were turing normal - had no flight since then though.

How were the light conditions and GPS connection at that time?
I had a similar problem with the DJI FPV quad, but that was in a dark tunnel.
Nevertheless, this is not good for trust into the system.

It was during an overcast day, late afternoon. So no direct sunlight but also not dark.

In summary, the Flight Controller mistakenly thought the Avata was moving North and rolled it CW to compensate. This can be seen by looking at North velocity from two different perspectives. OSD:xSpeed is computed by the FC. ImuCalcs:velN is not computed by the FC and derived by time differentiating the Lat/Long provided by the FC. Usually, these values are in close agreement.


The incorrect velocity and position is shown in green curve and trajectory. The actual velocity and position is shown in red.

From the OP's description it seems clear that OSD:xSpeed is incorrect and the location solution provided by the FC is correct. This could be caused by incorrect magnetometer or accelerometer data.

Also possible would be incorrect GPS velocity data which is derived from the Doppler effect on the recieved GPS signals. GPS location would be correct, just the GPS velocity would be incorrect. A bit unusual; velocity is more often correct.

Finally, the vision system may be being used to determine OSD:xSpeed. This would be my guess.

More could be learned if the .DAT could be provided. If you had the mobile device connectd to the goggles and the Fly App is set to not synch with DJI that .DAT will exist on the Fly App. Take a look here
https://mavicpilots.com/threads/mavic-flight-log-retrieval-and-analysis-guide.78627/
to see how to retrieve the .DAT from the Fly App. If the mobile device was not connected to the goggles during the flight the .DAT can first be moved from the goggles to the Fly App
using the method described here
https://fpvdronepilots.com/threads/dat-available.4225/#post-35016

Hi there. Thanks for reaching out, and we're sorry to read about the accident. Since the drone crashed, we strongly recommend you send it back for a proper diagnosis or repair. Our team can perform a Data Analysis to determine the specific cause of the crash. Warranty service will be applied depending on the analysis result and the warranty status of the drone. If you would consider sending it, you may directly submit a repair case at (repair.dji.com) or contact our Customer Support for further assistance with submitting the case. We appreciate your understanding and continued support.

Thanks for your guidance, I will do as advised.

BudWalker Posted at 3-26 09:42
In summary, the Flight Controller mistakenly thought the Avata was moving North and rolled it CW to compensate. This can be seen by looking at North velocity from two different perspectives. OSD:xSpeed is computed by the FC. ImuCalcs:velN is not computed by the FC and derived by time differentiating the Lat/Long provided by the FC. Usually, these values are in close agreement. [view_image]

[view_image]

Thank you very much @BudWalker, I and many others appreciate your detailed analysis and feedback!