Crash. Now what?

[RayOZ]

Isn't it at ~7.5, there was a full reverse (rc_elevator) that explains it going backwards? The loss of GPS at 10s is when it's in the bush?

 

[sar104]

Ok, I'm still learning about compass errors and diagnosing them, so this is going to be my first real stab at it. All you experts feel free to jump in and correct and destroy this graph. I'm practicing (sorry OP ) so here we go.

The Mavic's yaw aligned on a heading of approximately 140 degrees or facing SE. In actuality, the heading was approximately 35 degrees or NE (measured on Google Earth). This caused the flight controller to be confused upon take off because its inputs resulted in completely different results than it intended. That's about as "techy" as I can get in my explanation, so here is the graph.View attachment 37090
Again, experts feel free to tear this apart! -CF

You spotted the problem, and several observations corroborate the incorrectly initialized yaw. Adding the rudder inputs we see:



So not only was the initialized yaw (135°) significantly off from the estimated initial NW orientation, but at 8.3 s, with no rudder input, the yaw value starts to spin CCW.

Another test is to compare the recorded position (lat/long) converted to meters north and east of the takeoff point with the result of integrating the recorded IMU velocities north and east with respect to time.



That comparison illustrates that the initial motion of the aircraft was not in the direction that the FC expected. Positive pitch and negative roll (see first graph) on a heading of NW should have resulted in the aircraft moving south(ish) as the IMU data with the wrong yaw value (green trace) indicates. Instead it starts moving to the north as shown by the red trace. After that it's not the classic 90° yaw error toilet bowl because the initial error was nearer 180°.

Anyway - the cause was probably magnetic interference at the takeoff point. The GPS count declined simply because the aircraft ended up upside down - that was after the crash.

 

[Simmo]

yep.

 

[MavicCF]

You spotted the problem, and several observations corroborate the incorrectly initialized yaw. Adding the rudder inputs we see:

View attachment 37093

So not only was the initialized yaw (135°) significantly off from the estimated initial NW orientation, but at 8.3 s, with no rudder input, the yaw value starts to spin CCW.

Another test is to compare the recorded position (lat/long) converted to meters north and east of the takeoff point with the result of integrating the recorded IMU velocities north and east with respect to time.

View attachment 37094

That comparison illustrates that the initial motion of the aircraft was not in the direction that the FC expected. Positive pitch and negative roll (see first graph) on a heading of NW should have resulted in the aircraft moving south(ish) as the IMU data with the wrong yaw value (green trace) indicates. Instead it starts moving to the north as shown by the red trace. After that it's not the classic 90° yaw error toilet bowl because the initial error was nearer 180°.

Anyway - the cause was probably magnetic interference at the takeoff point. The GPS count declined simply because the aircraft ended up upside down - that was after the crash.

Cool! I'll have to try the lat/long compared to IMU, looks useful. Glad I'm finally figuring out his whole compass thing!

 

[RoboticTundra]

Video of a very short flight:

 

[dbyantis]

Per earlier suggestions did a Compass, IMU and RC calibration this morning. Ran some test flights near my home and all went well. Took the MPP to a home building site about 5 miles away to take some video. All looked good at takeoff. As it was rising it got to about 2m and then suddenly began going backwards and crashed in to some brush. I found it upside down in dirt.

Airdata indicates good GPS at start and then declining as it went backwards.

View attachment 37072

I assume vacuum off good (dirt in vents) and do a visual check? Turn on and see what Go4 says?

I'd really like to know what happened. I don't think I pulled back on the right stick but may have. Download .DAT files from it and post here?

Thanks all,

Were ther lots of metal around the build site? Just thinking of reasons for compass issues. I also thought that if there is a discrepancy in the sensors and "actual", the Mavic drops into ATTI. I cannot see that in the logs, but that would make the Mavic drift.

 

[dbyantis]

Video of a very short flight:

Re-bar in the concrete where you launched from?

 

[RoboticTundra]

Yep, there's rebar in the concrete that you can see in the video. As well as in the house you can see on the next lot north and in the house to the south (that the Mavic thought looked like a nice place to crash for a while).

The launch point was on original ground level and about 3' below the tops of the concrete walls on the site.

 

[RoboticTundra]

This was likely caused by magnetic interference from the rebar then? Or something else? Combination? Mavic and RC are likely OK and safe to fly?

If rebar. How far away do I likely need to be? I've flown from that exact same spot several times without incident (other than the left yaw problem posted elsewhere) though prior there have been concrete forms (steel? alum?) up. Is the problem likely just takeoff/landing (horizontal interference) and so move a bit away for that but then flyovers (as I've been doing) should likely be OK? E.G., is the issue interference between the Mavic and magnetic north so once it's in the air it's ok?

Thanks for everyones help on this. Not the way I'd wanted to learn but quite interesting nonetheless.

 

[sar104]

This was likely caused by magnetic interference from the rebar then? Or something else? Combination? Mavic and RC are likely OK and safe to fly?

If rebar. How far away do I likely need to be? I've flown from that exact same spot several times without incident (other than the left yaw problem posted elsewhere) though prior there have been concrete forms (steel? alum?) up. Is the problem likely just takeoff/landing (horizontal interference) and so move a bit away for that but then flyovers (as I've been doing) should likely be OK? E.G., is the issue interference between the Mavic and magnetic north so once it's in the air it's ok?

Thanks for everyones help on this. Not the way I'd wanted to learn but quite interesting nonetheless.

Something ferrous at, or very near, the takeoff point caused the compass reading to be significantly incorrect. Exactly how local that effect was is hard to say - it could be just within a few inches, or it could be larger depending on what it was. How fast the magnetic distortion disappears when the aircraft takes off also depends on how extensive the distortion is. Generally the effect goes away just a few feet off the ground at most (large steel structures such as ships would be an obvious exception to that).

But it's not necessarily okay once it takes off (as you saw), because if it starts with an incorrectly initialized heading then it does not simply reset that once the magnetometer readings change. In many cases the data show the aircraft gradually adjusting IMU yaw in an attempt to align it with the magnetometer readings, but if the error is too large it doesn't succeed. At that point it should switch to ATTI, but that may be too late. In your case it crashed before switching.

The only safe solution is to take off from an undistorted location, and the best way to check that is to ensure that the aircraft heading arrow on the map is pointing in the same direction as the actual orientation on the ground. In the case of this flight, you would have noticed that the aircraft was facing NW, but the arrow was pointing SE. That tells you that you have a magnetic problem and need to find a different launch location.

 

[RoboticTundra]

Thank you. And for the tip on checking the arrow direction for mag interference (wish I'd had screen record on). If aircraft direction and arrow are in agreement then can I be relatively assured that there is no interference?

Also, am I correct that it was attempting to keep itself in the same spot (over the landing pad in this case) but that since it thought it was facing a different direction that its actions had the opposite effect so it attempted more of the action in order to correct what it thought was happening?

 

[sar104]

Thank you. And for the tip on checking the arrow direction for mag interference (wish I'd had screen record on). If aircraft direction and arrow are in agreement then can I be relatively assured that there is no interference?

Yes - it would be unlikely, though not impossible, that the distortion simply reinforced the earth's magnetic field vector, but at least, even if that happened, the initialized yaw value would be approximately correct for the rest of the flight.

For that flight, even without seeing the screen recording, we know that the arrow was pointing SE because that was the initialized yaw value.

 

[sar104]

Also, am I correct that it was attempting to keep itself in the same spot (over the landing pad in this case) but that since it thought it was facing a different direction that its actions had the opposite effect so it attempted more of the action in order to correct what it thought was happening?

That's exactly how this happens.

 

[RoboticTundra]

Would the rear facing obstacle avoidance in the P4 have likely prevented it from going in to the brush? It would likely have stopped and hovered? Or at least long enough to go in to ATTI mode?

Would it be possible for DJI to do a better job of trapping errors like this and reacting quicker or quite difficult? I'd think that the down facing camera could be used to give it a bit more info on what's happening?

GPS had 15 sats. I assume it would not have been affected? How quickly can it process GPS data and determine that what it thinks its doing is not reality and so stop doing it? From the log we know that the information was available at some point because we have a quite accurate flight path.

 

[Sprocket]

Would the rear facing obstacle avoidance in the P4 have likely prevented it from going in to the brush? It would likely have stopped and hovered? Or at least long enough to go in to ATTI mode?

Would it be possible for DJI to do a better job of trapping errors like this and reacting quicker or quite difficult? I'd think that the down facing camera could be used to give it a bit more info on what's happening?

GPS had 15 sats. I assume it would not have been affected? How quickly can it process GPS data and determine that what it thinks its doing is not reality and so stop doing it? From the log we know that the information was available at some point because we have a quite accurate flight path.

Looks like a typical liftoff from very close to a big ol' Ford truck with a metal box bed 5' away. Happens every time.

 

[sar104]

Would the rear facing obstacle avoidance in the P4 have likely prevented it from going in to the brush? It would likely have stopped and hovered? Or at least long enough to go in to ATTI mode?

Would it be possible for DJI to do a better job of trapping errors like this and reacting quicker or quite difficult? I'd think that the down facing camera could be used to give it a bit more info on what's happening?

GPS had 15 sats. I assume it would not have been affected? How quickly can it process GPS data and determine that what it thinks its doing is not reality and so stop doing it? From the log we know that the information was available at some point because we have a quite accurate flight path.

The flight control system is unstable at that point. Obstacle avoidance is not going to help.

 

[MavicCF]

The flight control system is unstable at that point. Obstacle avoidance is not going to help.

So in this case would OA have made it worse? It would try to brake but would end up just agrevating the problem? Interesting to think about.

 

[MavicCF]

Looks like a typical liftoff from very close to a big ol' Ford truck with a metal box bed 5' away. Happens every time.

Now if it had been a Toyota...

 

[sar104]

So in this case would OA have made it worse? It would try to brake but would end up just agrevating the problem? Interesting to think about.

Presumably it would know from the OA sensors where the obstacle was relative to the aircraft, rather than in the earth's frame of reference, so in principle it could apply appropriate pitch and roll. However, I can't don't think it is going to override the FC's existing attempts to hold position.

 

[RoboticTundra]

Is there a good site/document you'd recommend that talks about mag interference a bit? Is it mostly stuff north of the takeoff point? All directions? How far? How high? Can interference bounce and so even though a craft may be 10' above it still be impacted? Strategies for dealing with it in tough situations (like at building sites)?

Thanks,