Mavic 2 Pro crash in urban area between tall glass buildings

[ajlholt]

I was flying my Mavic 2 Pro in an urban area with very tall glass buildings for a client. I had no problems when flying from the roof but when I went to fly over some gardens lower down, the drone could not pick up enough satellites and told me it was in Opti mode before launch. I took off to fly forward at about 6m height and noticed that the drone was not holding its position visually but was moving to the left. I decided to bring it back to land it but it suddenly started flying off to sideways the right towards a building. I tried to fly it to the left but it ignored my stick commands, impacted the building and fell causing extensive damage to the drone. Some questions:
1. Why did it drift left in the first place?
2. Why did it fly off to the right afterwards?
3. Why did it ignore my stick commands?
4. Why did the side obstacle sensors not prevent the crash into the building?
My suspicion is that the drone was initially in Opti Mode but then got some satellites which confused it. Perhaps the satellite signals were reflecting off the glass. Either way it is a nuisance that it is not possible to turn the GPS off entirely to fly in atti mode so that there is no risk of this happening.

 

[MAvic_South_Oz]

Flight log should reveal the cause.

Mavic Flight Log Retrieval and Analysis Guide

With the introduction of the Mavic Mini there has been a noticeable uptick in the number of posts requesting help with lost or crashed aircraft. While there is plenty of help and advice available on this forum, and detailed guidance and options on several other websites, I thought that perhaps a...

mavicpilots.com

Section 3 to find that file . . . and upload txt or the phantomhelp (or airdata) report link.

It really sounds like you had a magnetic interference from the ground take off point, which caused a compass - IMU disagreement, but the log will reveal that.

 

[slup]

Yeah, share the flight log ...

...4. Why did the side obstacle sensors not prevent the crash into the building?

The OD sensors doesn't work if you don't have a GPS lock ...
& furthermore regarding the lateral sensors:

 

[ajlholt]

Here is the txt file in case anyone can come up with answers. It was bright daylight and there with a light breeze.

 

[slup]

Here is the txt file in case anyone can come up with answers...

That was a very sketchy place to fly ... I would attribute this incident to a lack of knowledge how the drone works & what makes it keep horizontal hold & how difficult it can be to maneuver it in a situation like this.

A possible yaw error due to a power on in a magnetic interfered spot wouldn't have played a role here ... when the incident occurs in the later half of the flight the drone was effectively in ATTI ... or in a ATTI like state.

You never got a GPSLevel above 2 of 5, meaning that no HP was recorded & that the flight controller mainly disregarded the position data the few moments it reports to be in GPS mode even though the VPS sensor didn't lock your drone horizontally.

No flight path is recorded in the log & no Lat or Lon data exists from the drone ... it's only your mobile device that reports a position as seen below in the sat. picture.



The first half of the flight (up to 63,4sec, have placed the chart marker there in the chart below) you had assistance from the VPS sensor ... but after that the log reports VisionUsed=False, and with a GPSLevel as low as 2 your drone should have gone to ATTI but it stayed in some pseudo GPS/ATTI state leaving the drone without a proper horizontal positional hold. During some short periods the log reports ATTI (blue background in the chart). The effect of this would be a drone behaving as it was in ATTI but with hesitant & delayed stick commands.

Red graph= GPSLevel (HP is recorded & FC use positional data when reaching 4)
Green graph= # of locked satellites
Blue graph= Barometric height above take-off location
Light green background= Ordinary GPS mode with support from the VPS sensor
Pink background= No VPS sensor support for the horizontal positional hold
Blue background= ATTI mode

(Click on the charts below to make them larger)


In the second half of the flight you had to handle a drone pretty much in ATTI mode (as the GPSLevel only was 2 & the VPS didn't assist) ... the FC occasionally still didn't report full ATTI though, your drone was in some kind of "middle mode", seen this before & it usually means that the stick commands become mushy, imprecise & comes in with delay.

Have below placed the chart marker at 134,7sec ... Both the green dashed graphs shows you applying max negative stick input both with Aileron & Elevator ... but there neither pitch (red) or roll (green) change at all during 3sec ... but after that the movements shoots off. The crash occurs then at 137,2sec with major deflections in all axis ... you didn't release the sticks there, instead you had them nailed at max to 140sec.

Blue graph= Barometric height above take-off location
Red graph= Pitch
Green graph= Roll
Purple graph= Yaw
Black graph= Heading speed (take graph value times 10 to get it correct)
The 2 green dashed graphs= Aileron & Elevator stick commands

 

[eEridani]

Another observation is visual systems and mirrors do not play well together. Even if the systems were on and working as expected, glass buildings are essentially mirrors; warnings about glass and water abound, even in the drone manuals.

 

[vindibona1]

I'm sorry for the damage to you drone. I know how easy one could end up there. I had a very similar situation last year and I was too busy to see what mode I got switched into, but am sure I got thrown in the ATTI mode as I had near zero control. I did the best I could to just keep my M2P between the buildings and just bring it down safely. These drones are not built to fly well in such conditions because there is innate electrical interference with dozens of strong wifi signals present and with limited satellites the condition is a ticked to disaster. I'll never attempt to fly in such a situation again.

 

[ajlholt]

That was a very sketchy place to fly ... I would attribute this incident to a lack of knowledge how the drone works & what makes it keep horizontal hold & how difficult it can be to maneuver it in a situation like this.

A possible yaw error due to a power on in a magnetic interfered spot wouldn't have played a role here ... when the incident occurs in the later half of the flight the drone was effectively in ATTI ... or in a ATTI like state.

You never got a GPSLevel above 2 of 5, meaning that no HP was recorded & that the flight controller mainly disregarded the position data the few moments it reports to be in GPS mode even though the VPS sensor didn't lock your drone horizontally.

No flight path is recorded in the log & no Lat or Lon data exists from the drone ... it's only your mobile device that reports a position as seen below in the sat. picture.

View attachment 145740

The first half of the flight (up to 63,4sec, have placed the chart marker there in the chart below) you had assistance from the VPS sensor ... but after that the log reports VisionUsed=False, and with a GPSLevel as low as 2 your drone should have gone to ATTI but it stayed in some pseudo GPS/ATTI state leaving the drone without a proper horizontal positional hold. During some short periods the log reports ATTI (blue background in the chart). The effect of this would be a drone behaving as it was in ATTI but with hesitant & delayed stick commands.

Red graph= GPSLevel (HP is recorded & FC use positional data when reaching 4)
Green graph= # of locked satellites
Blue graph= Barometric height above take-off location
Light green background= Ordinary GPS mode with support from the VPS sensor
Pink background= No VPS sensor support for the horizontal positional hold
Blue background= ATTI mode

(Click on the charts below to make them larger)
View attachment 145738

In the second half of the flight you had to handle a drone pretty much in ATTI mode (as the GPSLevel only was 2 & the VPS didn't assist) ... the FC occasionally still didn't report full ATTI though, your drone was in some kind of "middle mode", seen this before & it usually means that the stick commands become mushy, imprecise & comes in with delay.

Have below placed the chart marker at 134,7sec ... Both the green dashed graphs shows you applying max negative stick input both with Aileron & Elevator ... but there neither pitch (red) or roll (green) change at all during 3sec ... but after that the movements shoots off. The crash occurs then at 137,2sec with major deflections in all axis ... you didn't release the sticks there, instead you had them nailed at max to 140sec.

Blue graph= Barometric height above take-off location
Red graph= Pitch
Green graph= Roll
Purple graph= Yaw
Black graph= Heading speed (take graph value times 10 to get it correct)
The 2 green dashed graphs= Aileron & Elevator stick commands
View attachment 145741

Thanks very much for this extensive and helpful reply. As I mentioned in my description I had the roll stick fully to the left while the drone flew steadily to the right towards the tall building. Is there any way to avoid this kind of problem in future or is it only possible to fly with GPS?

 

[ajlholt]

Another observation is visual systems and mirrors do not play well together. Even if the systems were on and working as expected, glass buildings are essentially mirrors; warnings about glass and water abound, even in the drone manuals.

There are certainly warnings about flying over water and the difficulty of the vertical positioning sensors being able to recognise height over water. There are no warnings about flying in the vicinity of glass clad buildings however and I wonder if GPS signals reflect off them altering the effective distance to satellites.

 

[ajlholt]

I'm sorry for the damage to you drone. I know how easy one could end up there. I had a very similar situation last year and I was too busy to see what mode I got switched into, but am sure I got thrown in the ATTI mode as I had near zero control. I did the best I could to just keep my M2P between the buildings and just bring it down safely. These drones are not built to fly well in such conditions because there is innate electrical interference with dozens of strong wifi signals present and with limited satellites the condition is a ticked to disaster. I'll never attempt to fly in such a situation again.

If it was possible to choose Atti mode that would be fine. The problem seems to be when the drone alternates between Atti and GPS. I don't understand why DJI do not allow Atti mode to be chosen.

 

[eEridani]

There are certainly warnings about flying over water and the difficulty of the vertical positioning sensors being able to recognise height over water. There are no warnings about flying in the vicinity of glass clad buildings however and I wonder if GPS signals reflect off them altering the effective distance to satellites.

Here's a clip from my DJI User Manual...

 

[Meta4]

There are certainly warnings about flying over water and the difficulty of the vertical positioning sensors being able to recognise height over water. There are no warnings about flying in the vicinity of glass clad buildings however

The worry about large expanses of glass is that obstacle avoidance can be fooled.
That wouldn't have been an issue as your drone never had GPS, so obstacle avoidance was unable to operate.

and I wonder if GPS signals reflect off them altering the effective distance to satellites.

Perhaps, but it wouldn't have made any difference.
For the entire flight GPS reliability was either 0/5 or 2/5.
The flight controller had no confidence in the location data.
That was because the buildings blocked large areas of the drone's skyview and the satellites in those parts of the sky.
For GPS to work properly, you need a good spread of sats.
When you lose most of the sats behind buildings and the only ones your drone can see are clustered in a small part of the sky, you cannot get the geometry that's needed for a good fix.

 

[ajlholt]

Here's a clip from my DJI User Manual...

View attachment 145777

Thanks. Does this mean that turning on APAS would have helped? Or would it still have been confused by getting and losing the GPS signal? The tall glass buildings were to each side of the aircraft and it was only ever facing forwards. There was vegetation below it but no water. The aircraft decided to fly sideways of its own accord firstly drifting to the left towards the building to the west and then flying to the right before the impact. In both cases I tried to correct those movements. The biggest issue was that if I let go of the sticks it did not hover in position like it should have using its vision systems. Perhaps the GPS was not telling it what stationary meant. I attach a screen shot from the video to show the location.

 

[Meta4]

Thanks. Does this mean that turning on APAS would have helped? Or would it still have been confused by getting and losing the GPS signal?

Not at all.
The drone never had GPS data that the flight controller considered reliable.

Perhaps the GPS was not telling it what stationary meant.

GPS was never good enough to give horizontal position holding.

 

[ajlholt]

So is there any way that a location like this could be flown safely in future?

 

[slup]

So is there any way that a location like this could be flown safely in future?

In general ... no.

Your drone is depending on either a good & consistent GPS accuracy or the correct conditions for the VPS sensors for a good horizontal hold.

When flying in a location where none of that is consistent, will always give you surprises & always changing control response & ways of flying the drone ... & adding to that a tight space not giving you enough time to figure out what's happening before the drone comes near obstacles ... & DJI's delay before the drone falls back to a true ATTI mode, instead leaving the drone in this pseudo mode where the drone isn't in ATTI instead in GPS mode despite that the GPSLevel is well below 4 & the VPS sensor can't lock on to the surface below... all together makes everything very sketchy.

On top of all this ... a belief from the pilot that the OD sensors... or APAS will fix it, being ignorant about that a good GPS accuracy is needed for them to work & during which flight modes the different sensors can be active.

All is set for disaster unfortunately ...

The only means of getting away from above described uncertainties ... is to have the ability to manually switch to ATTI mode & have the knowledge to master it, but that's not possible for a stock M2. In the past I believe that you could parameter hack the firmware & replace for instance Sport mode with ATTI ... don't know if that's possible anymore.

 

[BigAl07]

Here's a clip from my DJI User Manual...

View attachment 145777

Is this out of the DJI Mavic 2 Pro manual?

 

[eEridani]

Is this out of the DJI Mavic 2 Pro manual?

I can't find a Mavic 2 Pro manual at DJI. I did find Mavic Pro and Mavic 2 Enterprise. All have that same clause.

 

[eEridani]

Wait! I found the Mavic 2 Pro/Zoom manual! The same avoid glass warning is in the Tap to Fly and APAS sections of the manual.

 

[GrantMini3Pro]

I was flying my Mavic 2 Pro in an urban area with very tall glass buildings for a client. I had no problems when flying from the roof but when I went to fly over some gardens lower down, the drone could not pick up enough satellites and told me it was in Opti mode before launch. I took off to fly forward at about 6m height and noticed that the drone was not holding its position visually but was moving to the left. I decided to bring it back to land it but it suddenly started flying off to sideways the right towards a building. I tried to fly it to the left but it ignored my stick commands, impacted the building and fell causing extensive damage to the drone. Some questions:
1. Why did it drift left in the first place?
2. Why did it fly off to the right afterwards?
3. Why did it ignore my stick commands?
4. Why did the side obstacle sensors not prevent the crash into the building?
My suspicion is that the drone was initially in Opti Mode but then got some satellites which confused it. Perhaps the satellite signals were reflecting off the glass. Either way it is a nuisance that it is not possible to turn the GPS off entirely to fly in atti mode so that there is no risk of this happening.

Hi Ajlholt

I just had ‘exactly’ the same experience with a Mini 3 Pro at 4.30am filming a city scape in Vancouver - drone took off fine but within a minute whilst facing one of the mirrored glass buildings it started flying itself erratically & stopped responding to the RC. It then flew at the building & dropped to the ground, braking the casing, losing an arm, destroying the gimble & camera & effectively writing off the drone.

It seems either, as you say, the glass of buildings confuses the satellite signal & / or the sensors to the point its CPU is so confused it won’t accept instructions from the RC.

A warning for all pilots looking to capture cityscape footage - if you are above the buildings you should be fine but flying between them risks losing control of your drone.

Thanks for posting your note - pleased to know mine was not an isolated incident.

Fly safe, Grant