Mavic Air Flyaway, Second Set of Eyes Requested

[atdragos]

Problem? Flyaway

Was unit in a crash? Yes

What have you tried so far? Analysing flight data, talking to DJI support.

My account of what happened:

I was flying the aircraft outdoors when it decided to fly away, crash into a cliff face and drop into a river. I was unable to retrieve the aircraft from the river without a risk to life. I spent well over an hour assessing the situation after the crash before abandoning the recovery attempt.

A few comments from the scene:
- The data was logged during the entire event, indicating to me that signal between phone and drone was not lost.
- Line of sight was maintained until shortly before the drone hit the water.
- The only warning message before the crash was with regard to the maximum altitude, which was only at 2.5m.
- No other warning messages were received before the crash.
- The aircraft was not observed to initiate obstacle avoidance before colliding with the cliff face.
- I did not initiate RTH at any time during the flight.
- After loss of control, I pressed the pause button with no effect on the drone side.
- I tried stopping the aircraft with the RC after loss of control with no effect on the drone side.
- All other warning messages were received after the crash.

I was aware of the lack of GPS signal. However, I was mistakenly confident in the drone's control and fail-safe algorithms. I have attached my own analysis and labelled the important events, which are summarised as follows:

00.0s Takeoff
20.9s Complete loss of control
27.6s Drone crashes here

DJI Support are refusing to acknowledge that the drone crashed at 27.6s, saying instead the crash occurred at the very end of the measurement, at T=48s. They first said the vision system worked until T=30s. In the most recent email they are saying the vision system was not working between T=20s and T=22s, which is why the velocity spiked. In their words:

"For flyaway cases, we usually judge the crash by whether there was abnormal abrupt change of the attitude angles.
T=00:20, the vision system stopped working as the environmental conditions was not met for its working. There was no GPS signal, the aircraft could not figure out the horizontal velocity correctly; So the velocity showed now had no reference value; Also, according to the attitude angles and the vertical velocity, we can confirm that no crash happened at this Moment."

A quick note, if the velocity value is not correct without vision positioning, shouldn't this mean the attitude (yaw, pitch, roll) values they are using to justify their argument also be incorrect..?

I remember the timeline of events during that flight quite vividly. However, I have looked at this data a lot over the last 2 weeks and would very much appreciate a second set of eyes from this community.

What device are you using ( iphone , ipad, Samsung , etc)? Huawei P8 lite

What firmware are you running ( aircraft, remote controller)? Latest at the time of the crash

What Go app version are you using? Latest at the time of the crash

Any modification? No

Did you change anything or install any apps? No

Do you have a video or pictures of the problem? Was not recording at all during that flight

My flight log is here: DJI Flight Log Viewer - PhantomHelp.com

Thank you very much in advance.

 

[sar104]

Unfortunately I think that you have misunderstood how these aircraft, and their control systems, work. I don't know what you mean by "the drone's control and fail-safe algorithms", but you mention RTH and pause as if you expected those to have some effect in the absence of GPS or some other intelligent mode which, obviously, they cannot.

Without GPS the aircraft will operate in vision positioning mode if it has good enough light and appropriate objects/terrain to use. That's really designed for indoor flight when there are structures for it to track its location by. In the absence of VPS it will default to ATTI mode, in which it holds altitude constant or as dictated by the throttle input, and pitch and roll flat or whatever is dictated by stick input. It doesn't know position or horizontal velocity and, as such, cannot execute RTH and has nothing to pause.

If you look at the data for this flight, you can see that VPS was marginal and struggling to determine aircraft position. GPS was zero throughout - P-GPS is predicated on VPS. First impact with something, on this time scale, appears at 25.8 seconds:



If you look at the velocity that the FC computed from VPS, the problem appears at around 21 seconds. The aircraft had dropped out of VPS and P-GPS and then re-entered P-GPS and, (probably) erroneously, computed a large horizontal velocity that it attempted to correct for. You then applied full reverse elevator. The apparent lack of control was just a combination of those inputs.



As for your question about pitch, roll and yaw, that does not depend on the vision system at all - those parameters are measured directly by the IMU rate gyros, with absolute attitude values provided by the accelerometers and absolute yaw values being provided by the compass.

Anyway - the bottom line is that you were expecting far too much from the FC in the absence of GPS and with marginal VPS. It is extremely unwise to fly in those conditions unless you manually select ATTI mode and are skilled at flying that mode. I would not expect DJI to regard this as a warranty issue unless they are feeling uncommonly generous.

 

[Meta4]

I remember the timeline of events during that flight quite vividly. However, I have looked at this data a lot over the last 2 weeks and would very much appreciate a second set of eyes from this community.

Here's another explanation of what happened to give you an understanding of the incident because your comments show that you don't have a very good idea of what happened and why.

I was flying the aircraft outdoors when it decided to fly away, crash into a cliff face and drop into a river.

There was no "flyaway".
Your Mavic didn't decide to fly away at all and there was no loss of control at all during the incident.

I was aware of the lack of GPS signal. However, I was mistakenly confident in the drone's control and fail-safe algorithms.

You chose to fly in an area where the view of the sky was blocked by terrain tree cover etc.
Your Mavic was not able to acquire any satellites for the whole flight so it was without the position holding ability that GPS gives it.

I was aware of the lack of GPS signal.

You got that part right

However, I was mistakenly confident in the drone's control and fail-safe algorithms.

You got that part very wrong.
Without GPS, there is no position holding, no stopping, no obstacle avoidance and controlled flight is tricky - like driving on ice without brakes.

20.9s Complete loss of control
27.6s Drone crashes here

DJI Support are refusing to acknowledge that the drone crashed at 27.6s, saying instead the crash occurred at the very end of the measurement, at T=48s. They first said the vision system worked until T=30s. In the most recent email they are saying the vision system was not working between T=20s and T=22s, which is why the velocity spiked.

There was no loss of control.
At all times you had full control but no position holding ability.
You allowed the drone to slowly drift toward the cliff and did nothing to fly it away from the cliff.
The crash occurs at 2:05.5 in your flight data.

You took your hands off the sticks and watched your Mavic drift very slowly (<2mph) from 1:46.1 till 2:05.9.
The crash happened at 2:05.5
VPS was working and showed the ground below the drone at 1:48.5 and 2:05.1
But VPS can only help at heights below 8 metres, in good lighting and where the surface has a distinct pattern or texture.

DJI's analysis appears correct and the entire incident can be attributed to user inexperience and misplaced confidence putting the drone into a situation that caused the crash.
It would have easily been avoided by a pilot of average skill.
There's no sign of malfunction.

 

[ac0j]

Sorry for your loss!

But unfortunately Mavic forums has yet another incorrect thread title.
Those new to this forum and scanning threads are seeing way too many "flyaway" and "falling from the sky" thread titles, that are NOT as titled and pilot error.
I think the mods should change these titles, or add to them "Pilot error" when it applies. so people wont wrongly decide there is a huge problem with Mavics.

 

[atdragos]

Thank you for your time and help. I apologise for any unjustified assumptions.

In reply to sar104,

From the scene, there was no interruption between the time the drone started flying towards the cliff and the collision with the cliff.

The aircraft had dropped out of VPS and P-GPS and then re-entered P-GPS and, (probably) erroneously, computed a large horizontal velocity that it attempted to correct for.

Firstly, why is this even allowed in the logic? Is there no handling of huge discontinuities like that? Is the control system really accepting any reference value whatsoever and correcting for it?

Since the velocity traces cannot be used to pinpoint the time at which the drone started flying towards the cliff, or the time of the collision, the pitch trace must be used since

As for your question about pitch, roll and yaw, that does not depend on the vision system at all - those parameters are measured directly by the IMU rate gyros, with absolute attitude values provided by the accelerometers and absolute yaw values being provided by the compass.

The first large pitch angle change happens at 25.7s, which I am suggesting was the time the drone started flying towards the wall. An even larger pitch angle change was seen at 28.2, which I am suggesting was when the collision happened.

You took your hands off the sticks and watched your Mavic drift very slowly (<2mph) from 1:46.1 till 2:05.9.
The crash happened at 2:05.5

The Mavic did not drift slowly. The manoeuvre towards the cliff was aggressive and short. It did not take 20 seconds. I remember trying to correct with the RC Aileron seconds before the collision. Not 30< seconds before.
I cannot see any large changes in the pitch angle trace between 1:46.1 till 2:05.9.

Meta4, I have apologised for my unjustified assumptions. Please refrain from making any yourself.

Your feedback would be appreciated. Thank you for being clear in your statements and pointing to the data.

 

[sar104]

Thank you for your time and help. I apologise for any unjustified assumptions.

In reply to sar104,

From the scene, there was no interruption between the time the drone started flying towards the cliff and the collision with the cliff.



Firstly, why is this even allowed in the logic? Is there no handling of huge discontinuities like that? Is the control system really accepting any reference value whatsoever and correcting for it?

Since the velocity traces cannot be used to pinpoint the time at which the drone started flying towards the cliff, or the time of the collision, the pitch trace must be used since



The first large pitch angle change happens at 25.7s, which I am suggesting was the time the drone started flying towards the wall. An even larger pitch angle change was seen at 28.2, which I am suggesting was when the collision happened.




The Mavic did not drift slowly. The manoeuvre towards the cliff was aggressive and short. It did not take 20 seconds. I remember trying to correct with the RC Aileron seconds before the collision. Not 30< seconds before.
I cannot see any large changes in the pitch angle trace between 1:46.1 till 2:05.9.

Meta4, I have apologised for my unjustified assumptions. Please refrain from making any yourself.

Your feedback would be appreciated. Thank you for being clear in your statements and pointing to the data.

Regarding the comments from @Meta4 on the velocities - the recorded velocities are unreliable, as I showed in the second graph above.

With respect to allowed logic, I don't understand your question. If it loses positional data then it has no choice but to switch. If it regains positional data are you suggesting that it should just stay in ATTI because the positional data may be untrustworthy?

In any case, the issue is simple - you flew it without GPS in difficult VPS conditions in proximity to obstacles with wildly unrealistic expectations of how it would perform and a poor understanding of how the flight control modes work. Simply complaining that it should have done better is not reasonable and unlikely to get a positive response from DJI.

 

[InvisibleName]

But unfortunately Mavic forums has yet another incorrect thread title.
Those new to this forum and scanning threads are seeing way too many "flyaway" and "falling from the sky" thread titles, that are NOT as titled and pilot error.
I think the mods should change these titles, or add to them "Pilot error" when it applies. so people wont wrongly decide there is a huge problem with Mavics.

I totally agree with you! The title of this thread is very misleading, it should read “Oops, I messed up & crashed my drone, my bad!”

 

[Meta4]

Meta4, I have apologised for my unjustified assumptions. Please refrain from making any yourself.

I think your biggest unjustified assumptions are that you can fly a drone without GPS and that you can read flight data.
These two major shortcomings are responsible for your crash and your bungled attempt to make sense of it.

Firstly, why is this even allowed in the logic? Is there no handling of huge discontinuities like that? Is the control system really accepting any reference value whatsoever and correcting for it?

Firstly do you have any idea what atti mode is and how you have to fly if you are in atti mode?
In atti mode, You are the control system.
It's up to you to handle whatever "huge discontinuities" there are and correct for it.
No GPS = atti mode = no position holding, no brakes, no obstacle avoidance.
Everything is up to the pilot - and in this incident, the pilot wasn't competent for the task.

 

[swooshdave]

It's very easy to over-estimate what these drones can do. I am so glad I had a non-GPS drone to practice with before getting an Air. I'm a lot more confident when flying without all the electronic aids. And by confident I mean still on red alert as you can't always rely on the sensors and GPS to save you.