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

[Meta4]

If you have a better theory, I would love to hear it .

I don't have any theory.
I looked at your flight data and told you what it showed.
Your drone did respond to your control inputs, just not enough to save teh drone from crashing into the obstacle tou pushed it towards.

.. otherwise I will do more tests in regards to a reflected and/ or attenuated GPS signal. (hopefully without crashing my drone again)

You are talking about issues over 1 or 2 feet.
Forget it and go fly somewhere out in the open where 1 or 2 feet make no difference..

 

[Yorkshire_Pud]

Perhaps a compromise on privacy and if the log-examiners are willing, PM the intact log to the log-examiner?
Alternatively raise the problem on the DJI forum, their Mods frequently ask for logs to be sent to themselves or a DJI address, presumably for forwarding and or examination.

 

[GadgetGuy]

I don't have any theory.
I looked at your flight data and told you what it showed.
Your drone did respond to your control inputs, just not enough to save the drone from crashing into the obstacle you pushed it towards.

You are talking about issues over 1 or 2 feet.
Forget it and go fly somewhere out in the open where 1 or 2 feet make no difference..

More to the point, according to bladestrike/Ken Booth, DJI considers obstacle avoidance a feature and not a guarantee. When it fails, it is not a warranty issue, but pilot error! Ask me how I know! LOL!

 

[Rich Sugden M.D.]

I am curious if anything like this happened to someone else. (I did search, but could only find crashes when Active track was in use).

Yesterday I had the following issue occur:
I fly my drone outside of my house and it was hovering about 3ft above ground, and about 7-8ft from the house wall (facing the wall).
Obstacle avoidance was enabled (set to the minimum of 3.3ft horizontally).
I tap the stick to go forward slightly and it starts moving forward toward the wall ... Usually it stops about 3-5ft from the object in the path). This time it kept going forward (fairly slowly, but steadily - even with me letting go of the sticks completely).
When it got "too close for comfort" I pushed the right stick all the way backwards ... to make the drone fly backwards. Unfortunately the drone ignored this input and kept on flying forward ... a little jerky, but kept flying forward into the house wall (and crashed).

I even downloaded the flight via airdata and verified my stick-movement to see if I "dreamt" it and maybe pushed the lever in the wrong direction ... but it shows very nicely in the last 1-2 seconds before the crash that I pushed the right stick down, and the drone kept going forward ... into the wall.

There was no active-track enabled ... I was flying the drone with manual control only.

I understand that the drone can get confused if there is not enough contrast (if it wants to follow something in a specific direction), but is it normal for the drone to go in a direction even though the controls direct it to go the exact opposite way?
I would also expect that a house-wall has enough contrast ... but maybe not? There was also a window a little higher from where the drone crashed ...

Any thoughts would be welcome ...
Thanks!

EXACTLY same problem w. first flight of BRAND NEW Mini-3, but into tree branches, damaging the drone. Returned drone to DJI and they confirmed that drone moved forward in ATTI mode w/o any command, and drone did not respond to rearward stick command. I hover at about 6’ waiting for GPS lock in my yard w. all my drones. Never had this happen before. To add insult to injury the drone was delivered to my home while I was traveling overseas for a month, about 3 wks before I returned. Opened pkg., charged batteries, Registered/activated drone w. DJI, updated firmware, went outside and calibrated compass and took off to low hover … and drone wanders away. DJI would not honor warranty because drone was “3 weeks old”!!! Wanted to charge me to repair!!

 

[Yorkshire_Pud]

@Mordin I have just had another idea concerning hiding your location

Doesn't matter this idea was up a gum tree, as was pointed out to me by someone else, whoops.

 

[Mordin]

I don't have any theory.
I looked at your flight data and told you what it showed.
Your drone did respond to your control inputs, just not enough to save teh drone from crashing into the obstacle tou pushed it towards.

You are talking about issues over 1 or 2 feet.
Forget it and go fly somewhere out in the open where 1 or 2 feet make no difference..

So you do not have a theory ... and everything else in your post is simply wrong/ incorrect.

Just to summarize the events I tried to describe a week ago:
1) My drone was descending and then hovering between the houses (about 13ft. away from my house).
2) I push the lever slightly forward (to about 30%) for less than a second ... to make it go slowly toward my house.
3) The drone starts the expected slow approach (1ft. per second or so) to my house.
4) At the same time when I let go of the sticks (here the drone is about 8-9ft away from the house) the drone starts to accelerate toward the house.
5) I see this acceleration and try to control the drone to fly backwards (holding the stick at -100%)
6) The drone ignores this input and flies forward into the house-wall.

So why do you say this an issue over 1 or 2 ft? Have you not seen the video?
What obstacle did I "pushed it towards"???
You mean me pushing the stick 30% forward when the drone was 10ft. away from the wall? (and you completely ignore the fact that I pulled the stick all the way back for a full 2 seconds ... during which the drone was flying forward and crashing)

If you don't have another theory or idea that is fine, but please don't write things which are completely wrong ... that just confuses others reading through this thread.

I did not rely on OA to prevent a crash!
The drone itself accelerated toward the house (with me telling the drone at the same time to go backwards)

 

[Mordin]

@Mordin[ I have just had another idea concerning hiding your location.
Download the log's csv from Phantomhelp, open the csv in a spreadsheet program and apply a "change xyz to abc" to the integer part of the gps coordinates.
For example, if a latitude ordinate is "99.349956789" change "99." to "79." etc. and apply the same change to each entire latitude column. You'd need to include the decimal point in the 'change' so that the change does not pick up on the second "99".
Using "99" I assume your flight did not cross the boundary between "99" and either "98" or "100" but just check each entire column.

Similarly change the longitude ordinate columns.
Save the csv and upload that csv to a file hosting site etc. etc.

Thanks for the idea Yorkshire ... I did think of changing the actual coordinates and can already hear others scream "you changed the log to show whatever you want" ...

I will go ahead and add/subtract a constant value to each row (easy enough in Excel). Let's hope this will help others to "see" the GPS data too. The only changes during the significant timeframe are beyond the 5th digit anyways.

 

[MS Coast]

So you do not have a theory ... and everything else in your post is simply wrong/ incorrect.

Just to summarize the events I tried to describe a week ago:
1) My drone was descending and then hovering between the houses (about 13ft. away from my house).
2) I push the lever slightly forward (to about 30%) for less than a second ... to make it go slowly toward my house.
3) The drone starts the expected slow approach (1ft. per second or so) to my house.
4) At the same time when I let go of the sticks (here the drone is about 8-9ft away from the house) the drone starts to accelerate toward the house.
5) I see this acceleration and try to control the drone to fly backwards (holding the stick at -100%)
6) The drone ignores this input and flies forward into the house-wall.

So why do you say this an issue over 1 or 2 ft? Have you not seen the video?
What obstacle did I "pushed it towards"???
You mean me pushing the stick 30% forward when the drone was 10ft. away from the wall? (and you completely ignore the fact that I pulled the stick all the way back for a full 2 seconds ... during which the drone was flying forward and crashing)

If you don't have another theory or idea that is fine, but please don't write things which are completely wrong ... that just confuses others reading through this thread.

I did not rely on OA to prevent a crash!
The drone itself accelerated toward the house (with me telling the drone at the same time to go backwards)

Or, you might say, "Thanks for your help, but I disagree with some of your conclusions."

 

[Mordin]

EXACTLY same problem w. first flight of BRAND NEW Mini-3, but into tree branches, damaging the drone. Returned drone to DJI and they confirmed that drone moved forward in ATTI mode w/o any command, and drone did not respond to rearward stick command. I hover at about 6’ waiting for GPS lock in my yard w. all my drones. Never had this happen before. To add insult to injury the drone was delivered to my home while I was traveling overseas for a month, about 3 wks before I returned. Opened pkg., charged batteries, Registered/activated drone w. DJI, updated firmware, went outside and calibrated compass and took off to low hover … and drone wanders away. DJI would not honor warranty because drone was “3 weeks old”!!! Wanted to charge me to repair!!

Thanks for sharing your experience here!

I am not sure how DJI can say it is not their fault (after confirming the drone acted on its own). That's why I am concerned to go to DJI with my complaint ... I am sure they can wiggle themselves out of whatever I think should be their fault.

This issue seems a bit different from the malfunction of my drone as it works perfectly fine most of the time. I have been flying this drone for about 2 or 3 weeks without any issues. During this time I also tested the OA feature ... with me slowly approaching an obstacle until the drone does not go any further.
Toward trees, branches or the house it always stops a few feet before and just does not go any closer regardless how I push the levers.

That's why I think this malfunction my drone experienced is not an OA failure per se. In my case the drone did not only ignore OA - it also ignored the inputs from the sticks and flew in a direction completely on its own.

 

[Mordin]

Or, you might say, "Thanks for your help, but I disagree with some of your conclusions."

I definitely like your approach!
I always try to keep an open mind and accept other people's opinions ...

BUT in that post from Meta4 everything was wrong. Accusing me of pushing my drone into the wall ... even though I have video and logs of the drone flight showing it was not so. I just have to push back on completely incorrect statements.

 

[BudWalker]

@Mordin
I noticed this bit of oddness in the data you submitted. The xSpeed, ySpeed and zSpeed have no recorded values when they are in the interval [-0.2, +0.2]. Never seen that before.

I'm assuming that was not done by you? If not, then I would guess the FC opted not to compute or record these values.

BTW, if the .csv is modified by offfsetting the latitude and longitude values then it's possible to obtain the original values. That is, if anyone cared enough to go to the trouble.

 

[Yorkshire_Pud]

Thanks for the idea Yorkshire ... I did think of changing the actual coordinates and can already hear others scream "you changed the log to show whatever you want" ...

I will go ahead and add/subtract a constant value to each row (easy enough in Excel). Let's hope this will help others to "see" the GPS data too. The only changes during the significant timeframe are beyond the 5th digit anyways.

My idea was up a gum tree, 1 deg of arc at latitude x does not equate to the same distance as 1 deg of arc at latitude y, whoops. Sorry.

 

[MS Coast]

I definitely like your approach!
I always try to keep an open mind and accept other people's opinions ...

BUT in that post from Meta4 everything was wrong. Accusing me of pushing my drone into the wall ... even though I have video and logs of the drone flight showing it was not so. I just have to push back on completely incorrect statements.

Well, I see no accusation. However, you were the PIC and you chose to fly the drone between bushes toward a wall. The ultimate responsibility is yours. Perhaps there was an unexpected response from the drone, but you as the PIC, put the drone into a situation where there was little margin for error, allowing that unexpected response to cause a collision.

As for everything in that post being wrong, there are five sentences. I can't see that any is completely wrong. Is the statement "I don't have any theory." completely wrong?

Here's what the folks who provide GPS service in the US say about position accuracy.

GPS.gov: GPS Accuracy

Information about GPS accuracy

www.gps.gov

"For example, GPS-enabled smartphones are typically accurate to within a 4.9 m (16 ft.) radius under open sky (VIEW SOURCE AT ION.ORG). However, their accuracy worsens near buildings, bridges, and trees.""

"...the government commits to broadcasting the GPS signal in space with a daily global average user range error (URE) of ≤2.0 m (6.6 ft.), with 95% probability,"

DJI drones may may provide better position accuracy than smartphones, but the standard for the GPS signal itself is only 6.6 feet and that only for 95% of the time!

Considering that we're told not to expect accuracy better than a 32-foot circle under open sky and less near buildings and trees for consumer-grade devices, it seems unwise to intentionally fly a drone into a situation where a 3- to 5 -foot position error will lead to a collision.

To get down to basics and avoid further collisions, I offer three suggestions: 1) recognize that there's a lag between control inputs and aircraft motion, 2) don't rely excessively on OA features, and 3) don't fly so close to things you don't want to hit.

 

[Mordin]

@Mordin
I noticed this bit of oddness in the data you submitted. The xSpeed, ySpeed and zSpeed have no recorded values when they are in the interval [-0.2, +0.2]. Never seen that before.
View attachment 164441
I'm assuming that was not done by you? If not, then I would guess the FC opted not to compute or record these values.

BTW, if the .csv is modified by offfsetting the latitude and longitude values then it's possible to obtain the original values. That is, if anyone cared enough to go to the trouble.

Hello BudWalker,
I am still learning about what is in the log vs not. The visual representation of the Speed-values looks interesting ... but I have no idea what is "normal"
Do you have a similar graph of one of your flights? How it is supposed to look like?

Do you have a Mavic 3 to compare my flight data? Maybe this model of drone records things differently ...
Just to confirm: I removed serial numbers and GPS-coordinates - otherwise I did not change anything in the log.

 

[BudWalker]

Hello BudWalker,
I am still learning about what is in the log vs not. The visual representation of the Speed-values looks interesting ... but I have no idea what is "normal"
Do you have a similar graph of one of your flights? How it is supposed to look like?

Do you have a Mavic 3 to compare my flight data? Maybe this model of drone records things differently ...
Just to confirm: I removed serial numbers and GPS-coordinates - otherwise I did not change anything in the log.

Here is a the xSpeed, ySpeed and zSpeed data from another Mavic 3 flight.

All the DJI platforms will look like this; i.e., all data recorded.

 

[0Kajuna0]

GPS signal reflection is a thing. I'm not saying that's what happened here, and DJI drones should normally be able to mistrust a sudden change in coordinates. But if a surface reflects the signal, the position calculations will be wrong, while the recorded number of satellites, dilution of precision, level of GPS fix, etc. might not reflect the issue.

The example below looks something like that. Although I do not know if this was a reflection or just the brief drop in GPS satellites number

To view this content we will need your consent to set third party cookies.
For more detailed information, see our cookies page.

Accept third party cookies

 

[BudWalker]

Well, I see no accusation. However, you were the PIC and you chose to fly the drone between bushes toward a wall. The ultimate responsibility is yours. Perhaps there was an unexpected response from the drone, but you as the PIC, put the drone into a situation where there was little margin for error, allowing that unexpected response to cause a collision.

As for everything in that post being wrong, there are five sentences. I can't see that any is completely wrong. Is the statement "I don't have any theory." completely wrong?

Here's what the folks who provide GPS service in the US say about position accuracy.

GPS.gov: GPS Accuracy

Information about GPS accuracy

www.gps.gov

DJI drones may may provide better position accuracy than smartphones, but the standard for the GPS signal itself is only 6.6 feet and that only for 95% of the time!

Considering that we're told not to expect accuracy better than a 32-foot circle under open sky and less near buildings and trees for consumer-grade devices, it seems unwise to intentionally fly a drone into a situation where a 3- to 5 -foot position error will lead to a collision.

To get down to basics and avoid further collisions, I offer three suggestions: 1) recognize that there's a lag between control inputs and aircraft motion, 2) don't rely excessively on OA features, and 3) don't fly so close to things you don't want to hit.

The location solution provided by the Flight Controller is considerably better than the underlying GPS only solution. That's because the FC uses a Kalman filter that employs not only GPS location data but also accelerometer, gyro and the GPS velocity data. The GPS velocity data depends on the doppler shift in the received signals; it's not the derivative of the position data.

 

[MS Coast]

The location solution provided by the Flight Controller is considerably better than the underlying GPS only solution. That's because the FC uses a Kalman filter that employs not only GPS location data but also accelerometer, gyro and the GPS velocity data. The GPS velocity data depends on the doppler shift in the received signals; it's not an integral of the position data.

Not just a derivative of position data, rather than integral?

I'll have to look up Kalman filters and do some reading.

Thanks.

 

[Mordin]

The ultimate responsibility is yours

I completely agree with you about the PIC (me) being at fault for whatever the drone is doing. Even if the drone has a malfunction and decides to crash into a house (against the input on the control-sticks) - it is my fault. That's why I want to find out what exactly went wrong to prevent it from happening again - and to say "just stay away from houses" without knowing the exact reason for the malfunction is just a "guess". As the malfunction happened to me when I was 10ft away from the wall ... does that mean I should not ever get closer than 10ft from any object?
To be safe I should probably stay 20ft away ...

I plan to use the drone for house-inspections, so to just "not fly" between houses or close to one is not an acceptable solution. For the last few years I have been flying close to houses like 3-4ft. distance (with my previous drones) without any issues.

For now I cannot use my new Mavic 3 for anything around houses ... until I understand what causes these malfunctions. Then I can avoid them.

The accuracy of GPS is an interesting subject ... and different people/ websites have many different opinions about it. Even the URL you suggested (on gps.gov) shows right after the mentioned accuracy of 6.6ft that on Apr.21 2021 the accuracy was 2.1ft (across all satellites for 95% of the time).
So what should I do with this? The accuracy is anywhere between 6ft and 2ft? What about the remaining 5% of the time? What is the accuracy then? 10ft? or more?
Of course in addition to the system-inaccuracy at a certain time, there can be compounding factors (a tree or building in just the right spot) to reduce the accuracy even further.
How can DJI guarantee a GPS accuracy of +-0.5m? (https://dl.djicdn.com/downloads/DJI_Mavic_3/DJI_Mavic_3_User_Manual_v2.0_enI.pdf)

All of that is even more reason that the GPS signals should never be used as the main reason for the drone to fly in a certain direction (which I believe happened for the incidents with my drone)

 

[Mordin]

The location solution provided by the Flight Controller is considerably better than the underlying GPS only solution. That's because the FC uses a Kalman filter that employs not only GPS location data but also accelerometer, gyro and the GPS velocity data. The GPS velocity data depends on the doppler shift in the received signals; it's not an integral of the position data.

I knew about using the doppler-shift for calculating velocity (which can be highly accurate) I never heard about the Kalman filter ... from a quick google search it sounds like a very interesting approach to provide estimates. I will have to read more.