My mavic 2 pro crashed into sea in California coast 2 weeks ago. It was at about 7.7m height above sea near a rock and still had 85% battery left. It was basically hovering at that moment and I was trying to climb up a bit but surprisingly it instead started drifting down and into sea. From the camera view I could see it submerged into sea and was gone! This was my first time to lost a drone having owned a phantom 3 and the original mavic pro. I looked at the flight record and surprisingly found there was a warning "Upward Obstacle Detected" near the end. It must be trying to avoid the "obstacle" above and chose to go downwards. But it was open sky. How on earth there could be "obstacle"? Could be DJI's Vision Position System get confused by the reflection of water and think going down is the safest path?

I contacted DJI and the best thing they can do is a *PAID* analysis of the cause of the flying away because the warranty already expires. I can't understand why a customer needs to pay to provide them important data that can help them improving their product! I won't pay for such a "service". But I am curious if anyone on this forum can provide any insight about

1) Why did DJI mavic 2 pro detect upward obstacles on open sea?
2) Can DJI mavic 2 pro's VPS technology see the sea below?

I am attaching the screenshot of the flight record near the end.

Huh, that's the thing I've never heard about before! The only explanation (other than extremely rare system glitch) may be an albatross or seagull hovering above. Sorry to hear bad news, bro ...

Matthew Dobrski Posted at 1-27 21:06
Huh, that's the thing I've never heard about before! The only explanation (other than extremely rare system glitch) may be an albatross or seagull hovering above. Sorry to hear bad news, bro ...

Even the "obstacle" was momentary, how could mavic 2 pro think going down to sea is safer?

I noticed at home that some sources of light, for example my desk led lamp is causing this sensor to detect obstacle, this means drone is on table on side of table there is this desk led lamp when it is on table whole time upward obstacle detected message is on display despite there is nothing above drone, when I turn the light off it will disapear. So I think it could be the same case maybe some interference from sunlight cause this.

DH_Pilot Posted at 1-27 21:49
I noticed at home that some sources of light, for example my desk led lamp is causing this sensor to detect obstacle, this means drone is on table on side of table there is this desk led lamp when it is on table whole time upward obstacle detected message is on display despite there is nothing above drone, when I turn the light off it will disapear. So I think it could be the same case maybe some interference from sunlight cause this.

Very interesting! Then could that mean DJI's detection of obstacle is purely by vision? Now I am thinking could be the clouds moving high above causing their deep learning algorithm to get the conclusion something is approaching?

Hiya plannar,

Sorry for your loss. If you want help upload your flightlog using > https://www.phantomhelp.com/LogViewer/Upload/

mayby your log will explains more what happend.

cheers
JJB

I can't understand why a customer needs to pay to provide them  important data that can help them improving their product! I won't pay  for such a "service". But I am curious if anyone on this forum can  provide any insight
Post your recorded flight data so someone can look into what actually happened instead of having people make guesses (that will mostly be irrelevant).

JJB* Posted at 1-27 23:02
Hiya plannar,

Sorry for your loss. If you want help upload your flightlog using > https://www.phantomhelp.com/LogViewer/Upload/

Thanks JJB* for the pointer. I have uploaded it to
https://www.phantomhelp.com/LogViewer/CEN397VP7PHT7WFGTM47

Playing with the record. One thing caught my eyes. Initially the IMU altitude and VPS altitude were in sync. However the VPS altitude drifted away and a lot of N/A (VPS got confused by wave movement/reflection?) . Mavic 2 pro seemed to make bad decision about its position. Eventually VPS altitude dropped to 0 and submerged in sea while IMU altitude still at 25.9ft. Now I interpret the "upward obstacles detected" as the sea water.

Doesn't this show mavic 2 pro's made BAD decision based on inconsistent IMU altitude and VPS altitude?

plannar Posted at 1-28 00:55
Playing with the record. One thing caught my eyes. Initially the IMU altitude and VPS altitude were in sync. However the VPS altitude drifted away and a lot of N/A (VPS got confused by wave movement/reflection?) . Mavic 2 pro seemed to make bad decision about its position. Eventually VPS altitude dropped to 0 and submerged in sea while IMU altitude still at 25.9ft. Now I interpret the "upward obstacles detected" as the sea water.

Doesn't this show mavic 2 pro's made BAD decision based on inconsistent IMU altitude and VPS altitude?

Hi Planner,

A difficult one, your flightlog....

At the end your MP was flying at an altitude of 23-24 feet, 64% fwd and about 40% yaw left.
The barometric height increases a little ; from 23 to 27 feet, but with no UP input.
Guess this height measuring is not correct but to keep MP at height it steers down, vertical speed minus values. (so the software thinks, going up so steering down, wich is correct ofcourse if the sensor measures correct values)

Height measuring even more up, so steering down more down.

See the VPS shows the actual distance, and as it hits the water at the end i think VPS is correct measuring/indicating.

Your RC input 100% UP too late for keeping it dry, vertical speed down at impact 1.6 feet/sec, UP need some time to stop that movement.

Stopping the pitch forward made the drone pitches up from minus 9 to 25 degrees up, this will not help the going down move.

I have no idea why the barometric sensor went wrong, but could it have been water spray ?
And yes, agree   2 times "Upward Obstacle Detected" most have been water drops.

In the chart the white vertical line = first contact with the sea.

cheers
JJB

JJB* Posted at 1-28 02:08
Hi Planner,

A difficult one, your flightlog....

Thanks JJB for the excellent analysis! It all makes sense. Only thing still in doubt is why barometric sensor went wrong. The first "upward obstacle detected" happened at VPS altitude 0 meaning it already hit the water.  However it rained for about 10 min earlier before my flight but has stopped.

So it is
1) the barometric sensor went wrong
2) DJI's software does not give much weight to VPS in determining the altitude reflecting how confident they are in their VSP technology!

plannar Posted at 1-28 09:14
Thanks JJB for the excellent analysis! It all makes sense. Only thing still in doubt is why barometric sensor went wrong. The first "upward obstacle detected" happened at VPS altitude 0 meaning it already hit the water.  However it rained for about 10 min earlier before my flight but has stopped.

So it is

Flying in rain is never a good idea, don`t know why the barometric sensor went wrong.
Guess that low near the water more thicker drops on top of drone, hence the warning.

VPS height is used if the autolanding protection is enabled, guess a MP2 pro has this feature too.
So or it did not work on time, or it was disabled or is had less priority than the steering down beacuse it was correcting the height.

cheers
JJB

plannar Posted at 1-28 09:14
Thanks JJB for the excellent analysis! It all makes sense. Only thing still in doubt is why barometric sensor went wrong. The first "upward obstacle detected" happened at VPS altitude 0 meaning it already hit the water.  However it rained for about 10 min earlier before my flight but has stopped.

So it is

There are  problems with your data that suggest it cannot be relied on.
From 2:06.1 the VPS is showing height steadily decreasing to zero at 2:09.4, while barometer data stays steady at 26 ft.

From 2:09 you went full stick forward with the left stick but neither barometer data nor VPS shows any increase in altitude.

If the sensors were working properly, the drop in VPS height would be indicating that you flew at a steady height over a steep edge of a 20 ft rock.
But if you were over the sea the whole time, that's suggesting that one or both height sensors are not reading properly.

From 2:09.4 the VPS stays at 0.3 ft while barometer height stays constant and then gradually decreases to 20 ft.

Looking deeper into the recorded flight data, reveals that the Obstacle Avoidance was disabled for the whole flight.
It also indicates a problem with the VPS sensors.
The data indicates that VPS was alternately on and off right through the flight.

Have you flown over the sea much?
Flying in rain is not a good idea and if there was salt on and in your drone, the addition of water could have caused unpredictable errors as small short circuits happened.


It might help in understanding the data to know how far above water level was your launch spot and was the sea calm or rough?

Labroides Posted at 1-28 14:50
There are  problems with your data that suggest it cannot be relied on.
From 2:06.1 the VPS is showing height steadily decreasing to zero at 2:09.4, while barometer data stays steady at 26 ft.

There is no problem with my data. There is only problem with the interpretation.

As JJB pointed out in #10, the barometric increased a little and was wrong reading. Since I did not push the up stick, mavic 2 pro decided to go down to keep the previous barometric value. then the barometric (wrong) value increased more, so mavic 2 pro decided to go down more. It was like the doomed Boeing 737 max it decided to suicide!

That day was low tide, so the home spot on the flight record was land. The crash time was 5:14pm and the lowest tide was scheduled at 5:57pm. The Sun just set down.  I pull out my drone after the short rain had stopped. I am uploading a picture closest to that time.

Several points to your doubt.
. The barometer was initially in sync with VPS height reading. then became wrong. But vps height reading was correct. However dji only rely on the barometer reading and made wrong decisions (going down by itself)

.2:09 you went full stick forward with the left stick but no height change. Take a look at the VPS height. It was 0! That was the moment it hit water.

plannar Posted at 1-28 20:19
There is no problem with my data. There is only problem with the interpretation.

As JJB pointed out in #10, the barometric increased a little and was wrong reading. Since I did not push the up stick, mavic 2 pro decided to go down to keep the previous barometric value. then the barometric (wrong) value increased more, so mavic 2 pro decided to go down more. It was like the doomed Boeing 737 max it decided to suicide!

There is no problem with my data. There is only problem with the interpretation.
Sounds like you are now an expert at analysing flight data?
So you should have no problem pointing out what you think I got wrong.

As JJB pointed out in #10, the barometric increased a little and was wrong reading. Since I did not push the up stick, mavic 2 pro decided to go down to keep the previous barometric value. then the barometric (wrong) value increased more, so mavic 2 pro decided to go down more. It was like the doomed Boeing 737 max it decided to suicide!
JJB pointed out that he was guessing.
I've read his assessment multiple times, comparing it with the actual data and I still can't make sense of it.
That's because of significant problems with your data which I explained in my earlier post.

Prob a bird crapped directly on the top sensor.

Labroides Posted at 1-28 14:50
There are  problems with your data that suggest it cannot be relied on.
From 2:06.1 the VPS is showing height steadily decreasing to zero at 2:09.4, while barometer data stays steady at 26 ft.

Hi Labroides,

From 2.06.1 the log shows not what you are saying....not steady at 26feet, see just before that at 2m4.1 when the values are going up. (not the craft)

See chart ; Baro Alt increases (23.3 ft to 27.9) without UP input, Vertical speed into minus values, VPS height shows height decreasing.

If OA was disabled, guess not. Do you get warnings from upper sensor when its disabled? Mayby yes, but that would be not logical.
Data field OA in decrypted file is not always correct for its value. (did just a check switching all the sensors off/on, changes not written in the file (for my MA))
I can disable or enable OA on my MA and the field remains FALSE.  But with disabling OA the AutoLandingProtection is not swicth off, that is an other option. (for MA, guess the same for MP etc)
If this protection is ON than drone will stop at 0.5 VPS meter, so it looks like that this setting was off or overruled by the down steering by SW.
With manually down it will stop at 0.5, don`t know if it will work when the SW dictates it down.

I do not see VPS height errors in the log, what do you read for value(s) and where?  (just being curious to learn more)

VPS height indication above moving water is not always correct in its measuring/indication, when loosing reflected signals no data is recorded ; as in this file.  But VPS height decreasing indication matches the Vertical speed values. see my chart.

cheers
JJB

AntDX316 Posted at 1-28 21:23
Prob a bird crapped directly on the top sensor.

lol  are you the stand-in for Cutis ?  

JJB* Posted at 1-29 00:22
lol  are you the stand-in for Cutis ?

What were his posts though?  I never seen it until someone called him out.  

edit: It doesn't matter.  I'm not even here to pick people or for people to pick me out.  I'm here to learn more about the DJI products.

JJB* Posted at 1-29 00:21
Hi Labroides,

From 2.06.1 the log shows not what you are saying....not steady at 26feet, see just before that at 2m4.1 when the values are going up. (not the craft)

From 2.06.1 the log shows not what you are saying....not steady at 26feet, see just before that at 2m4.1 when the values are going up. (not the craft)
I'm sorry.
Should I have said that the barometer showed height at 26 ft +/- 6 inches??
That's close enough to stable for me.
At the same time he VPS is showing height steadily decreasing to zero at 2:09.4.
Clearly both cannot be correct, but which is wrong and why?
What else is incorrect?
When did the data become unreliable?


I do not see VPS height errors in the log, what do you read for value(s) and where?  (just being curious to learn more)
You need to look at the Verbose CSV to see that (and a lot more).

Labroides Posted at 1-29 03:24
From 2.06.1 the log shows not what you are saying....not steady at 26feet, see just before that at 2m4.1 when the values are going up. (not the craft)
I'm sorry.
Should I have said that the barometer showed height at 26 ft +/- 6 inches??

23.0 to 27.9 is more than just few inches....(that`s why vertical speed into minus values)

But no answer to my last question, i know all the verbose csv fields, i understand DAT files etc ect.
So in wich field do you see VPS errrors? plain txt or true/false values? just curious as i cannot find....

I trust the VPS reading, as the drone follows the VPS readings (uh...other way around ofcourse )

cheers
JJB

JJB* Posted at 1-29 03:47
23.0 to 27.9 is more than just few inches....(that`s why vertical speed into minus values)

But no answer to my last question, i know all the verbose csv fields, i understand DAT files etc ect.

23.0 to 27.9 is more than just few inches....(that`s why vertical speed into minus values)
If you had read what I wrote in post #13 properly, I wouldn't have had to explain your misunderstanding a second time.

From 2:06.1 the VPS is showing height steadily decreasing to zero at 2:09.4, while barometer data stays steady at 26 ft.
There's no 23 feet showing between 2:06.1 and 2:09.4.
If there was, I wouldn't have said that the barometer data was steady.

But no answer to my last question, i know all the verbose csv fields, i understand DAT files etc ect.So in wich field do you see VPS errrors? plain txt or true/false values? just curious as i cannot find....
You know all the Verbose CSV fields?
But you don't know what column AD is showing?
Look at it and you see the data indicates that VPS was alternately on and off right through the flight.

I made an effort to explain clearly back in post #13.
Twice you've told me I am wrong, because you haven't read properly.
I won't bother explaining again.

Labroides Posted at 1-29 04:35
23.0 to 27.9 is more than just few inches....(that`s why vertical speed into minus values)
If you had read what I wrote in post #13 properly, I wouldn't have had to explain your misunderstanding a second time.

Well, to be precise and telling the whole story  and yes i understand all data in those files.

AD represents if the VPS is measuring values or not *
AM represent if VPS has errors or not. (why only ask me if i know AD?....)
* if out of VPS range or not measuring becasue of not receiving signals back value = FALSE

When flying low over water its 'normal' to see this VPS height measuring behaviour.
In this flight AM is completely FALSE, so no VPS errors in flight.

And ofcourse we can have different opinions what happend, nothing to do with my misunderstanding.
Just see the facts about climbing 6 feet and steering down, but nah  no reactions on my writing or data shown in charts.

BTW  another chart ; upper line = VPS error, lower line = VPS measuring. [ lime colour = oke ]

"Twice you've told me I am wrong, because you haven't read properly"Uh? did i say you were wrong? just have a different idea about what happend, that`s all.

Happy many landings,
cheers
JJB

Thanks JJB and Labroides. I learned a lot from your technical discussion.  Some points:
.JJB's interpretation agrees what I have observed. I was feeling the drone slowly lowering its height and when my final effort to pull it up was too late.
.Top sensor (AntDX316's guess) could be out of the concern because the first time "upward obstacle detected" happened was when VPS height hit 0. Of course the second time was already under water. So the obstacle was the sea water
.I did not observe any jerky action from the video camera. So probably nothing hit it while it was in the air
.I was flying over water. did not fly along the edge of the big rock.
.I believe both OA and AutoLandingProtection are on. I used to see it detecting obstacle from all directions and would hover a while above 0.5m before landing or when I put my hand below it. and I did not change setting for this flight.

Thank you guys so much! I learned a lot here.
  

plannar Posted at 1-29 09:19
Thanks JJB and Labroides. I learned a lot from your technical discussion.  Some points:
.JJB's interpretation agrees what I have observed. I was feeling the drone slowly lowering its height and when my final effort to pull it up was too late.
.Top sensor (AntDX316's guess) could be out of the concern because the first time "upward obstacle detected" happened was when VPS height hit 0. Of course the second time was already under water. So the obstacle was the sea water

Hi palnnar,

Thanks, its my pleasure to help.

If you want to learn more about analysis, use FRAP.
https://forum.dji.com/thread-203659-1-1.html

Fly a 'mission', check in FRAP and see back how the flight was performed in data.
I alway fly with screen recorder active, see what messages i have missed watching the drone and not the screen app. (and not all messages are logged in the flightrecord)

cheers
JJB

JJB* Posted at 1-29 11:10
Hi palnnar,

Thanks, its my pleasure to help.

Thanks JJB for the link. No wonder you are an expert on flight analysis, you wrote your own analysis app
I will definitely download it and play. Screen recorder is a good idea!
Really appreciate all your help here!

Your drone had probably already crashed into the sea and the "upward obstacle" was the water itself