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

[Mordin]

I am glad I could get that video showing the flight path ... I am sure a lot of readers of this thread were thinking "he was probably Hi and did not know where up/ down, left/ right was" ;-)

So yeah ... the drone was on a constant forward motion during the last 7 seconds.

I have a Theory ...
What if the signal from one of the GPS-satellites was reflected?
Let me explain what I mean:
GPS works in getting the distances from each of the satellites (in current view) and calculating the position of the receiver by triangulating the different satellite distances.
For my though-experiment I am assuming that only one satellite signal is being reflected. Let's assume it is that GPS-satellite that happened to be South (toward the back) of the drone.

If the drone receives a stronger reflected signal from that satellite in the South and moves North (As I was pushing the lever to go that way) it would see the distance to the South-Satellite become less. As the drone "thinks" it is too far South, it should even push the motors to go further North to correct the position. It will continue to do that as long as the reflected signal is stronger than the direct signal from the satellite ... until it crashes into the wall.

Unfortunately I have no idea how to prove (or disprove) this theory as these satellites zip all over the place, and you never know where one or more of them are (in view).
Just from my day-work dealing with WiFi-"issues" all the time, I know that "Low-E" windows with the metal substrate do absorb and reflect RF signals pretty good. With the GPS signal of around 1.2-1.6GHz it is fairly close to the 2.4GHz and 5GHz used by WiFi, so I think it is possible those RF-signals can get reflected in a similar way.
The location where the drone crashed is also between my house and the neighbors house ... about 8 meters (26ft) apart. They are both 2 story houses. My neighbors house could have blocked the signal, and on my house there was a window right where the drone crashed ... and also on the second floor is one.

If the drone calculates the xSpeed from the GPS-signals, this would explain why the log shows the movement toward South/ backwards ... as the drone "thought" it went that way.

I am not even asking for "plausible", but does anyone think this is a "possible" explanation?

When I look at the latitude information of the log, I do see the value decrease constantly in the last 3-4 seconds (except for the last 2 entries), but it only happened on the 6th digit after the decimal ... that's all below the accuracy of GPS (which should be around 1m).

 

[Yorkshire_Pud]

Those 'questions' are probably @sar104 domain.

 

[MS Coast]

I had the exact thing happen to me, but I know why. I was low on battery. Extremely low. I had cancel go home a couple times. I went forward to land the drone, but it must have went into an auto land sequence. It proceded to just go forward. It didn't respond to my stick movements. Hit the wall and fell to the ground. No harm done luckily.

Take a close look at all your props and check for nicks and cracks. It's hard to imagine that the drone would collide with a solid surface and not do some damage to the props.

 

[Mordin]

Take a close look at all your props and check for nicks and cracks. It's hard to imagine that the drone would collide with a solid surface and not do some damage to the props.

Thanks for the suggestion.
At the same time of the drone-purchase I also got 3 additional sets of replacement props, as I had a few object-disagreements with my previous drone.
This crash did not cause any significant damage though. I guess because it was VERY low speed? (on the video you actually see the drone almost stop right before the wall, but then inched just a little further ... into the wall)
There are some scuff marks, and the edge of one prop is not exactly straight ... but just a little "rough".
I cannot imagine these minor bruises would cause any change in flight characteristics.

 

[Yorkshire_Pud]

I'd twist each blade slightly at the tip, I have found cracks in blades that closed so perfect they couldn't be seen unless the light caught them 'just right', twisting the blade helps to open such cracks up and makes them easier to see.
I've had an M2P catch a solid object with a prop, I'd have thought it would be the leading edge or tip of the blade that got damaged, wrong, it was the trailing edge that got damaged. Eventually I came to the conclusion that the striking blade more or less stopped dead whilst the motor continued to turn, the other blade then came around and whacked into the rear of the stuck blade.
Check the propeller hubs and their hooks/latches as well as the blades.

 

[MS Coast]

Thanks for the suggestion.
At the same time of the drone-purchase I also got 3 additional sets of replacement props, as I had a few object-disagreements with my previous drone.
This crash did not cause any significant damage though. I guess because it was VERY low speed? (on the video you actually see the drone almost stop right before the wall, but then inched just a little further ... into the wall)
There are some scuff marks, and the edge of one prop is not exactly straight ... but just a little "rough".
I cannot imagine these minor bruises would cause any change in flight characteristics.

It's not the speed of the drone that's significant, but rather the speed of the prop tips. A Mavic 3's props are about 4.25" long. With a motor speed of about 10,000 rpm, the tips of the props are moving at about 250 miles per hour.

I'd use some of those replacement props if it were my decision. The worry would be that one of those little nicks in the edge of a propeller propagates into a crack and the entire end of a blade breaks free, leading to some really bad results.

 

[Meta4]

But I am not sure what you are trying to point out ...
That part I completely agree with your assessment of what the log shows.

I wasn't "trying" to do anything or making any assessment.
I simply showed what data the drone's sensors recorded about the incident.

You will also note that the drone did pitch backwards when you pulled the left stick back.
But despite pulling it fully back for two seconds, the pitch angle never exceeded + 5.5°.

 

[pentatony]

Take a close look at all your props and check for nicks and cracks. It's hard to imagine that the drone would collide with a solid surface and not do some damage to the props.

Oh, believe me , I checked. I always check for cracks or damage anywhere after even a minor crash. The props especially. I learned my lesson , that's why I do what I do. Prop broke 2 days after , what I thought was a minor crash.

 

[tlyons]

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.
...

At the very end of the video there is a full stop. Then a short reverse. Then a fast descent. And another abrupt stop.
Then the drone is very quiet and still. (grin)

 

[Yorkshire_Pud]

At the very end of the video there is a full stop. Then a short reverse. Then a fast descent. And another abrupt stop.
Then the drone is very quiet and still. (grin)

Might you be seeing a bounce off the wall isn't it? I see no free flight reverse in the security cam footage.

 

[Drbobk]

Take a close look at all your props and check for nicks and cracks. It's hard to imagine that the drone would collide with a solid surface and not do some damage to the props.

Remove your props and check the tabs that hold them in place. It’s possible to break one even though the blades are not cracked. I had one break off in a tangle with a bush that left the props visibly undamaged.

 

[tlyons]

Might you be seeing a bounce off the wall isn't it? I see no free flight reverse in the security cam footage.

That was intended as humor; I was pointing out the incredibly obvious. Of course the log shows a stop and reverse; the drone flew into a wall and bounced off. I don't have a Mavic 3 but every drone I have had keeps the log running until it powers off or lands and I would expect the log to show the crash and at least a few moments after.

It is also worth remembering that GPS positioning is very coarse relative to these small movements. The locations appear precise because the algorithm is averaging and predicting and smoothing but there are constant, random +/- 1ft changes in location and velocity in the raw data. As likely as not, the averaged/predicted/smoothed position became in error with the sudden unexpected 0 velocity and the algorithm just inserted some negatives to fix the mistake and line up to the actual GPS position.

My guess on the crash? probably a random mechanical or electrical failure. Sticky pot on the controller left it pushing forward after pulled back; maybe a mismatched or loose prop recently changed; maybe a neighbor's microwave oven briefly interrupted the control signal.

 

[Filmarik]

It's not the speed of the drone that's significant, but rather the speed of the prop tips. A Mavic 3's props are about 4.25" long. With a motor speed of about 10,000 rpm, the tips of the props are moving at about 250 miles per hour.

I'd use some of those replacement props if it were my decision. The worry would be that one of those little nicks in the edge of a propeller propagates into a crack and the entire end of a blade breaks free, leading to some really bad results.

10,000 rpm? Where did you get that info from?

 

[Meta4]

I would expect the log to show the crash and at least a few moments after.

See post #35.
The recorded data shows the crash and another 0.6 seconds after that.

As likely as not, the averaged/predicted/smoothed position became in error with the sudden unexpected 0 velocity and the

As well as speed data derived from GPS, the recorded data has data from gyro sensors that shows the drone pitching backwards to reverse.

My guess on the crash? probably a random mechanical or electrical failure. Sticky pot on the controller left it pushing forward after pulled back; maybe a mismatched or loose prop recently changed; maybe a neighbor's microwave oven briefly interrupted the control signal.

None of those.
The drone did reverse, just not enough to avoid collision.

 

[MS Coast]

10,000 rpm? Where did you get that info from?

I looked around on several drone forums and found several seemingly realistic sources that suggested figures ranging from 8,000 to 10,000 rpm for various DJI models.

 

[Mordin]

In the last few days I was flying my drone in the same area numerous times, and even though it sometimes behaved a little erratic, it never decided to accelerate so far to cause another crash.

I also looked into the GPS-system and how the satellites orbit the earth. Just as a summary: GPS-satellites have a period of one half of a sidereal day (almost the same as a stellar day) which is the 360 degree turn of the earth with a distant star as reference point. The sidereal day is roughly 4 min. shorter than a solar day.

Like this GPS-satellites go over the same spot on the surface of the earth once a day (shifting each day by 4 min.).
This should mean the position of all the GPS satellites is the same every 23 hours 56 minutes.

I am not sure what exactly caused the malfunction of my drone a few days ago, but I am assuming it is related to the GPS signals being blocked/ attenuated. So I was trying to fly in the same spot (between my house and the neighbor's house) and with my newly discovered "rhythm" with the GPS satellite positions I chose a time where the satellites are in roughly the same spot as on May 15th (at 17:41) when the drone crashed into the house the first time.

So I decided to fly yesterday (May 19th) and moving toward the same spot between the houses around 17:20 (as according to my calculations the same GPS-satellite locations should occur around 17:25 that day). After a few flights back and forth (and having some erratic behavior) it did not take long until the drone decided again to accelerate toward my house-wall (this time also toward the left) and it crashed into the house in an angle ...

Again (just like with the first crash) I verified in the log that I pushed the forward/ backward stick all the way for the drone to go back, but it kept going forward & slightly left into the house.

Now I had 2 crashes with my new drone that were not even my fault.

Just to test a little more, I took my Mavic 2 and flew in the same spot ... back and forth. Eventually the Mavic 2 also decided to go directly toward the house. (sometimes it was toward my neighbor's house as well) Just like before I pushed the stick for the drone to go backwards (away from the wall), and the drone actually moved away from the wall.

So why does the Mavic 2 react so differently from the Mavic 3? Why does the Mavic 3 rely so much on its GPS data that even a manual intervention on the controls is being ignored?

Should not the manual controls always take precedence?

Even though I agree with some of tlyons’ descriptions about GPS, I do not believe in random/ unexplained failures that occur strangely enough in a specific area and not anywhere else ... I have no idea why some “sticking” pot should cause the crash, where the log entries confirm my stick movement, but the video (and my own visual observation) shows the drone’s actual movement (contradicting the xSpeed in the same log)

... but maybe I should just start yelling at my neighbor for using his microwave (even though they can cause interference in the WiFi bands, with the spread-spectrum usage and the dual-band connection of the drone which works over 10 miles, it is highly unlikely this could cause an interruption in the controller signal with me being less than 20ft away)

so ... what now?
Does anyone here have experience with opening a case with DJI?
Is there any chance they actually "do" anything about it?

 

[Meta4]

I am assuming it is related to the GPS signals being blocked/ attenuated.

Why?
There's nothing to indicate you had any GPS issue.

so ... what now?

Learn not to fly so close and toward obstacles.
Find a nice clear area, well away from obstacles to do your flying.

Does anyone here have experience with opening a case with DJI?
Is there any chance they actually "do" anything about it?

You would have to have a case where you could show that there was an actual issue DJI was responsible for.

 

[Meta4]

So why does the Mavic 2 react so differently from the Mavic 3? Why does the Mavic 3 rely so much on its GPS data that even a manual intervention on the controls is being ignored?

Should not the manual controls always take precedence?

You make a lot of assumptions.
Following them isn't productive unless they are accurate (and I'm not sure they are).

 

[Mordin]

You make a lot of assumptions.
Following them isn't productive unless they are accurate (and I'm not sure they are).

Well ... that's how I usually find the cause for "issues" at my day job.
You need a good theory ... and for every theory you need to make assumptions ... and then try to prove parts of it (or the whole theory)

I don't know much about the technical details on how a drone controls its movement, but I intend to find out enough details ... hopefully to a point where the options of what could cause the malfunctions is limited enough.

If you have a better theory, I would love to hear it ... otherwise I will do more tests in regards to a reflected and/ or attenuated GPS signal. (hopefully without crashing my drone again)

 

[BudWalker]

In the last few days I was flying my drone in the same area numerous times, and even though it sometimes behaved a little erratic, it never decided to accelerate so far to cause another crash.

I also looked into the GPS-system and how the satellites orbit the earth. Just as a summary: GPS-satellites have a period of one half of a sidereal day (almost the same as a stellar day) which is the 360 degree turn of the earth with a distant star as reference point. The sidereal day is roughly 4 min. shorter than a solar day.

Like this GPS-satellites go over the same spot on the surface of the earth once a day (shifting each day by 4 min.).
This should mean the position of all the GPS satellites is the same every 23 hours 56 minutes.

I am not sure what exactly caused the malfunction of my drone a few days ago, but I am assuming it is related to the GPS signals being blocked/ attenuated. So I was trying to fly in the same spot (between my house and the neighbor's house) and with my newly discovered "rhythm" with the GPS satellite positions I chose a time where the satellites are in roughly the same spot as on May 15th (at 17:41) when the drone crashed into the house the first time.

So I decided to fly yesterday (May 19th) and moving toward the same spot between the houses around 17:20 (as according to my calculations the same GPS-satellite locations should occur around 17:25 that day). After a few flights back and forth (and having some erratic behavior) it did not take long until the drone decided again to accelerate toward my house-wall (this time also toward the left) and it crashed into the house in an angle ...

Again (just like with the first crash) I verified in the log that I pushed the forward/ backward stick all the way for the drone to go back, but it kept going forward & slightly left into the house.

Now I had 2 crashes with my new drone that were not even my fault.

Just to test a little more, I took my Mavic 2 and flew in the same spot ... back and forth. Eventually the Mavic 2 also decided to go directly toward the house. (sometimes it was toward my neighbor's house as well) Just like before I pushed the stick for the drone to go backwards (away from the wall), and the drone actually moved away from the wall.

So why does the Mavic 2 react so differently from the Mavic 3? Why does the Mavic 3 rely so much on its GPS data that even a manual intervention on the controls is being ignored?

Should not the manual controls always take precedence?

Even though I agree with some of tlyons’ descriptions about GPS, I do not believe in random/ unexplained failures that occur strangely enough in a specific area and not anywhere else ... I have no idea why some “sticking” pot should cause the crash, where the log entries confirm my stick movement, but the video (and my own visual observation) shows the drone’s actual movement (contradicting the xSpeed in the same log)

... but maybe I should just start yelling at my neighbor for using his microwave (even though they can cause interference in the WiFi bands, with the spread-spectrum usage and the dual-band connection of the drone which works over 10 miles, it is highly unlikely this could cause an interruption in the controller signal with me being less than 20ft away)

so ... what now?
Does anyone here have experience with opening a case with DJI?
Is there any chance they actually "do" anything about it?

Really interesting theories. Too bad you won't share the data so that others can validate your theories.