Mavic 3 flies forward (on its own) crashing into a house wall

[IronSky1]

I have had simliar occurance with one of my Mavic 2s. Flying close to ground 4ft and a-5ft from wall (OA on) . When I went to roof level and try to fly back to location M2 went the other way. When I stay back from wall 10 ft fine very strange. Even stranger when I repeated with my M2E Thermal it worked fine I could get to 2-3ft no issue (thermal map of air leaks).
I beleive it was just a firmware/hardware issue on that particular M2. If you can repeat you may want to contact DJI.

My though GPS isnt accurate enough to lock at those distances with objects all around.

All would contruction no metal

 

[HoozierDroneDaddy]

Don't have all the expertise of those above, but could the makeup of the wall have something to do with it. i.e., concrete with rebar affecting the compass, etc.? Or some electronics on the other side of the wall?

 

[Phantomrain.org]

I also have never seen that Screen Before so that was New to me , Thank you

As well as loosing GPS signal there could have been some Magnetic Interference where the Compass got hammered and pointing in the Wrong Direction , I do think that is possible taking off from a cement drive.

Phantomrain.org
Gear to fly in the Rain. Land on the Water

 

[MavicMover]

Uhm ... by moving the slide-bar for "Obstacle Break Distance1' all the way to the left under "Obstacle Sensing Settings"
Was this some kind of secret?
Or did I call it the wrong thing?

Not to hijack but how do you access these settings?

 

[waynorth]

I'm confused. This screenshot is of the DJI Fly app?

The screenshot is from the Pilot app, the drone is a Mavic 3T Enterprise.

 

[3rdof5]

The screenshot is from the Pilot app, the drone is a Mavic 3T Enterprise.

I wish the Fly App had that setting!

 

[Mordin]

I'm confused. This screenshot is of the DJI Fly app?

It is a screenshot from the DJI Pilot 2 app on my controller.
I don't think i specified yet what controller i am using. It is a "DJI RC Pro".

 

[Mordin]

I also have never seen that Screen Before so that was New to me , Thank you

As well as loosing GPS signal there could have been some Magnetic Interference where the Compass got hammered and pointing in the Wrong Direction , I do think that is possible taking off from a cement drive.

Phantomrain.org
Gear to fly in the Rain. Land on the Water

Hmmm ... interesting thought about the interference. I cannot think of any big interference in the area (also behind the wall on the other side is no electric/ electronic equipment - besides the normal wires for power and rebars in the walls).

I also do not understand why a "hammered" compass would cause the drone to fly by itself in any direction. As far as I could tell, it was flying in a fairly straight line (toward/ into the wall). With a "freaking out" compass, I would expect it to fly (if at all) a curve or even more erratically.

BTW: Whenever I land my drone on the metal transformer case in my yard, it shows an alert to calibrate the compass, but landing and taking off there was also never an issue.

 

[Meta4]

I also do not understand why a "hammered" compass would cause the drone to fly by itself in any direction. As far as I could tell, it was flying in a fairly straight line (toward/ into the wall). With a "freaking out" compass, I would expect it to fly (if at all) a curve or even more erratically.

It wouldn't and didn't.
This is what you get when people start making random guesses.
You only get good explanations for drone incidents from actual data.

 

[Mordin]

Something log-related I wanted to get verification on.
While I was looking at the log of my flight (which ended in the crash) I see the entries for OSD.xSpeed and OSD.ySpeed. As I found a link online these should specify the speed of the drone in the North-South and East-West direction, correct?
In the last 8 entries in my log it shows mostly negative numbers in the OSD.xSpeed column (around -0.9). The ySpeed shows only close to 0 values. This would mean the drone was flying almost directly South (according to the log) ... in those 2 seconds it would mean roughly 1.8m going South. (as positive values would be North)
I would just like to understand if my thinking about the values is correct.

If my understanding of those log-values is correct, then it unfortunately means the log has incorrect values in those fields, as I saw the drone facing North (maybe a little to the NW), and flying forward/ North against the wall.

But how can the drone be so "confused" about what direction it is going?
(BTW: I looked at my GPS-coordinates as well, but it only changed on the 6th digit after the decimal point - by 3 or 4 - which should be past the accuracy of the GPS signal - so no help there)

 

[Meta4]

Something log-related I wanted to get verification on.
While I was looking at the log of my flight (which ended in the crash) I see the entries for OSD.xSpeed and OSD.ySpeed. As I found a link online these should specify the speed of the drone in the North-South and East-West direction, correct?

They are showing the forward/backward components of the drone's horizontal speed, not compass directions.

 

[Mordin]

They are showing the forward/backward components of the drone's horizontal speed, not compass directions.

so you are saying this description is wrong?

Flight log retrieval and analysis guide

I posted this guide on the MavicPilots forum after the introduction of the Mavic Mini led to a significant increase in pilots asking for help diagnosing events and understanding log files. It may also be of some interest over here since most of it applies equally to other DJI equipment. While...

phantompilots.com

where it shows:
15. OSD.xSpeed, OSD.ySpeed, OSD.zSpeed – velocity north, east and down, respectively;

I know nothing about these logs ... so I do not know yet.
But even if it shows the forward/backward component ... Do negative values for xSpeed mean it is going forward or backward. I mean I know the drone went forward, but it is just strange to have forward being negative numbers ...

 

[Meta4]

so you are saying this description is wrong?

Thanks .. I've learned something new.
We don't often have need to look at the x & y, just the z component is what's usually important.
Or we can visually see the direction when we have full data
The x & y speeds are the north & east components of the horizontal speed.
So that if the drone was to fly at 13 mph on a heading of 138° (SE) it would show -9.4 mph for the x component and 8.9 mph for the north component.

 

[GadgetGuy]

Thanks .. I've learned something new.
We don't often have need to look at the x & y, just the z component is what's usually important.
Or we can visually see the direction when we have full data
The x & y speeds are the north & east components of the horizontal speed.
So that if the drone was to fly at 13 mph on a heading of 138° (SE) it would show -9.4 mph for the x component and 8.9 mph for the north component.

Aren't the x components and the north components one and the same, based upon the post above?

"15. OSD.xSpeed, OSD.ySpeed, OSD.zSpeed – velocity north, east and down, respectively;”

 

[Meta4]

so you are saying this description is wrong?

Flight log retrieval and analysis guide

I posted this guide on the MavicPilots forum after the introduction of the Mavic Mini led to a significant increase in pilots asking for help diagnosing events and understanding log files. It may also be of some interest over here since most of it applies equally to other DJI equipment. While...

phantompilots.com

where it shows:
15. OSD.xSpeed, OSD.ySpeed, OSD.zSpeed – velocity north, east and down, respectively;

I know nothing about these logs ... so I do not know yet.
But even if it shows the forward/backward component ... Do negative values for xSpeed mean it is going forward or backward. I mean I know the drone went forward, but it is just strange to have forward being negative numbers ...

On the right, the blue highlighting shows you pushing the right stick forwards.
On the left the blue highlighting shows the speed of the drone as it responds to the joystick input.
Note that the x component is positive.

The pink highlighting shows the right stick pulled fully backwards and slightly left, and the corresponding speed of the drone.
This speed is negative - the opposite of what happened when the stick was pushed forward.
The purple highlighting shows the crash impact in the pitch, roll and yaw data.

Speeds and distances involved were very slow/small.

 

[Mordin]

On the right, the blue highlighting shows you pushing the right stick forwards.
On the left the blue highlighting shows the speed of the drone as it responds to the joystick input.
Note that the x component is positive.

The pink highlighting shows the right stick pulled fully backwards and slightly left, and the corresponding speed of the drone.
This speed is negative - the opposite of what happened when the stick was pushed forward.
The purple highlighting shows the crash impact in the pitch, roll and yaw data.

Speeds and distances involved were very slow/small.

Thanks for looking into this ... and showing the different areas in the log. It helps a lot to see the log separated in the different activities.

But I am not sure what you are trying to point out ...
I know the drone flew (in the last 5-6 seconds of its flight) only forward. Around timestamp 9:14-9:15 I probably pushed the throttle lever forward for the drone to get closer to the house-wall, but I let go right after ... and when I saw it got too close I pushed it all the way back to reverse direction. That part I completely agree with your assessment of what the log shows. As everything happened rather slowly I had time to react and push the lever the other way.

During the same time frame the drone was on a constant forward-path toward the wall ... so the negative xSpeed numbers at the end do not make sense to me. I do not know how the drone comes up with the xSpeed, ySpeed values ... so I do not know how those could get "confused" or calculated incorrectly.

I was also able to "Catch" the drone flight on one of my Security-cameras. It is a bit grainy as it is further back from that camera, but it shows nicely the last 10 seconds of the drone flight. First the drone reducing altitude, then going forward (probably when I pushed the levers slightly forward) ... and then continuing forward (a little jerky) until it crashes.
I hope the upload of the GIF worked ... showing the important section of the security-feed.

 

[Yorkshire_Pud]

Might on explanation for the apparent OA failure be that the house wall is in shadow whilst everywhere else is in bright sunlight? I.e. might it be that given the ambient light the house was to poorly lit for the OA to detect? Does the Mavic have IR OA sensors?
What does "set to the minimum of 3.3ft horizontally" mean?
If you can adjust the severity of the drone's braking how severe do you have it set?

 

[Mordin]

Might on explanation for the apparent OA failure be that the house wall is in shadow whilst everywhere else is in bright sunlight? I.e. might it be that given the ambient light the house was to poorly lit for the OA to detect? Does the Mavic have IR OA sensors?
What does "set to the minimum of 3.3ft horizontally" mean?
If you can adjust the severity of the drone's braking how severe do you have it set?

I am still trying to test/ figure out how and where the OA works vs. not,
The light conditions could be a factor as this drone only has optical OA (with the exception for downward - there it has an IR sensor too). But I would think even on the shadow side of the house it should have enough contrast ... something to test a little further.

The following is how I understand (And also what I verified partially) of how the settings work:
The drone has different "brake" and "warning" settings for each direction (upward, downward, horizontally)
The "warning" setting specifies when foreign objects show up as a warning on the display. The "brake" setting specifies when the drone actually applies its break to stop before an object. On "normal" objects this works surprisingly well ... like around trees, even branches, garden-tables, stools, ...)
For the most part it recognizes house-walls as well pretty reliably (until it decided to crash into one with OA enabled).

So even with me pushing the throttle lever forward, the drone refuses to go any further if there is an obstacle within the set "brake" distance.
The range for the distances is from around 3.3 to 16 or 20ft (I don't remember exactly ... it also various between the different directions)
At this time I am flying mostly very slowly - I set the distances to their minimum. If I set it any higher I would also have trouble taking off (and landing) on my covered porch, as it "detects" foreign objects all around.

 

[Yorkshire_Pud]

I am still trying to test/ figure out how and where the OA works vs. not,
The light conditions could be a factor as this drone only has optical OA (with the exception for downward - there it has an IR sensor too). But I would think even on the shadow side of the house it should have enough contrast ... something to test a little further.

The following is how I understand (And also what I verified partially) of how the settings work:
The drone has different "brake" and "warning" settings for each direction (upward, downward, horizontally)
The "warning" setting specifies when foreign objects show up as a warning on the display. The "brake" setting specifies when the drone actually applies its break to stop before an object. On "normal" objects this works surprisingly well ... like around trees, even branches, garden-tables, stools, ...)
For the most part it recognizes house-walls as well pretty reliably (until it decided to crash into one with OA enabled).

So even with me pushing the throttle lever forward, the drone refuses to go any further if there is an obstacle within the set "brake" distance.
The range for the distances is from around 3.3 to 16 or 20ft (I don't remember exactly ... it also various between the different directions)
At this time I am flying mostly very slowly - I set the distances to their minimum. If I set it any higher I would also have trouble taking off (and landing) on my covered porch, as it "detects" foreign objects all around.

I'd be very wary about using minimum distances until you have verified what the braking distances are. I have no idea if the Mavic 3 adjusts those settings depending on flight speed but if it does not and you were flying at full speed in N mode, 3.3 ft may not be enough distance for the OA to prevent the collision.

 

[Yorkshire_Pud]

I am puzzled as to why your video seems to show the drone moving more or less consistently forwards whilst the logs seem to suggest it stopped and reversed.

Attached is a slightly longer version of duration Meta4's csv, it shows more of the descent seen in your video, with a slightly different arrangement of columns. Unfortunately I do not know how to partially shade cells when using a Mac.