Air2 S started landing over water.

[Jpat]

I will have to do some testing with the gaffers tape. Having it off and knowing it wont do this again would be great. I don't mind having to manually avoid whats below. This just happened so fast not much time to react before it was too late. And yeah very interested on dji's thoughts but @msinger seems to be dead on with vps issue.

 

[RodPad]

You answered my question.

Rod ..

 

[msinger]

Your thought is reported to DJI no matter what?

Yes, as it might at least get the OP a discount on a replacement. Better than nothing.

 

[msinger]

I will have to do some testing with the gaffers tape

For sure. You should definitely do some test flights with the tape attached (ideally over land first) to see how the drone reacts.

Having it off and knowing it wont do this again would be great

Flying that close to the surface of water is always going to be risky. I guess it's a chance you gotta take if you want to get shots like that.

 

[WanderLost]

The recorded altitude suddenly increases right before the crash, while the real altitude is obviously not increasing in the video. This confusion causes the drone to descend rapidly to maintain its previous altitude:

I don't know what sensor caused this confusion. GPS satellite count remains strong, so probably not that.

Altitude measurements come from the onboard barometric sensor which can be confused by rapid changes in air pressure. There may have been wake turbulence from the boat that messed with your drone's height measurements and caused it to overcompensate. Just a possible guess. I'd like to see a chart showing the altitude recorded in the moments leading up to the crash.

 

[BudWalker]

Altitude measurements come from the onboard barometric sensor which can be confused by rapid changes in air pressure. There may have been wake turbulence from the boat that messed with your drone's height measurements and caused it to overcompensate. Just a possible guess. I'd like to see a chart showing the altitude recorded in the moments leading up to the crash.

I've been thinking this also.

The top green plot shows an increase in altitude starting at about 295.7 secs. The bottom red plot is the time integrated vertical velocity. It matches the video and shows no change in height. About 1 secs later a descent is commanded which is evident in both plots.

I suspect the disparity between the time integrated vertical velocity and the barometric pressure altitude will be seen in the .DAT if @Jpat cares to retrieve it.

 

[Jpat]

I've been thinking this also.
View attachment 166700
The top green plot shows an increase in altitude starting at about 295.7 secs. The bottom red plot is the time integrated vertical velocity. It matches the video and shows no change in height. About 1 secs later a descent is commanded which is evident in both plots.

I suspect the disparity between the time integrated vertical velocity and the barometric pressure altitude will be seen in the .DAT if @Jpat cares to retrieve it.

Is this file on the phone?

 

[BudWalker]

Is this file on the phone?

If the App is set to NOT synch with DJI then, yes, the .DAT should be on the mobile device. Take a look at
Mavic Flight Log Retrieval and Analysis Guide
Section 3 Log File Retrieval
subsection Aircraft DAT files.

The correct .DAT will have a name that ends with FLY068.DAT

 

[Jpat]

If the App is set to NOT synch with DJI then, yes, the .DAT should be on the mobile device. Take a look at
Mavic Flight Log Retrieval and Analysis Guide
Section 3 Log File Retrieval
subsection Aircraft DAT files.

The correct .DAT will have a name that ends with FLY068.DAT

Pretty sure I synced my flight logs.

 

[MS Coast]

Altitude measurements come from the onboard barometric sensor which can be confused by rapid changes in air pressure. There may have been wake turbulence from the boat that messed with your drone's height measurements and caused it to overcompensate. Just a possible guess. I'd like to see a chart showing the altitude recorded in the moments leading up to the crash.

But, the drone was ahead of the boat and would not have encountered any wake effects from the boat.

I doubt this could happen, but if the idea is that the boat was pushing a mass of air ahead of it, that would increase the air pressure which would indicate lower altitude, and cause the drone to climb.

 

[BudWalker]

But, the drone was ahead of the boat and would not have encountered any wake effects from the boat.

I doubt this could happen, but if the idea is that the boat was pushing a mass of air ahead of it, that would increase the air pressure which would indicate lower altitude, and cause the drone to climb.

I agree, initially it was a bit of a stretch. But, I'm starting to like it more. IF a high pressure builds in front of the boat then a corresponding low pressure area has to exist somewhere. And, IF it looks like an airfoil pressure distribution then that low pressure area would be just behind the high pressure area that's in front of the boat. I.e., the low pressure are would be directly over the boat.

Reviewing the video the MA2 is over the boat when the incident occurs at video time 2:38 (.txt log time 297 secs). There is just one other time where the MA2 is over the boat which occurs at video time 1:10 ( .txt log time 209 secs).

Does the data show a disagreement between the recorded altitude and vertical velocity at this previous time where the MA2 is over the boat? Theoretically, the difference between vertical velocity and time differentiated altitude should always be 0.0. In practice, excursions from 0.0 will be seen when the MA2 is changing altitude. When altitude is not changing the difference will be close to 0.0.

Got all that? The data shows the same effect, to a lesser degree, when the MA2 crosses over the boat at video time 1:10 ( .txt log time 209 secs). Here is a plot of that difference.

 

[Zbip57]

The issue could have been VPS related (the downward sensors confused by the water below)

The downward sensors often receive unwarranted blame for lots of mysterious behaviour, but never with any reasonable explanation as to how that would be possible.

The VPS is composed of two completely separate systems:

• Infrared (or ultrasonic) sensor for measuring height above surfaces, and
• Optical sensor for holding position

Neither one of those would, or even could, have been a factor in this boat crash. Yes, there are lots of warnings in the manual about the VPS possibly being confused by water, but NONE of those factors can ever cause the drone to unexpectedly drop into the water.

Height is normally sensed by the barometric altimeter. The VPS infrared height sensor only ever seems to have any active effect on the drone when detecting anything within 0.5m (2ft) of the underside of the drone, in which case Landing Protection will only ever automatically cause the drone to rise up. Otherwise if you are holding the throttle down when within 2ft of the detected surface, you are commanding the drone to engage auto-landing.

Maybe it's different with other DJI models, but my Mini never reacts in any way to changing VPS height measurements as the terrain height fluctuates below, whether while flying or hovering. So even if the infrared height sensors could somehow become "confused" by water, how could that cause the drone to crash?

I did a series of four videos in this YouTube playlist demonstrating what the infrared height sensors do.

The VPS optical sensor is only used for holding position when hovering with control sticks centred. It will only work with sufficient lighting, when seeing a discernible pattern to lock on to. If that pattern is moving (wave ripples, a leaf drifting by on the water, etc) the drone may follow that motion rather than holding a steady fixed position.

I was convinced the VPS optical sensor on my Mavic Mini would only ever track and follow moving objects in the absence of sufficient GPS lock. I did a video experiment last week intending to demonstrate that, only to be shocked to discover that's not true at all. The VPS optical sensor will track and follow moving objects even when the Mini has sufficient GPS coverage.

See the fifth video in that same playlist and in this other discussion thread, mavicpilots.com/threads/yes-landing-protection-can-be-disabled-but.134162

The optical sensor can cause the drone to wander away while hovering, hit a tree and fall into the water, but the optical sensor, even if "confused" by water, will never cause the drone to drop all by itself.

Whenever something unexpected happens over water, the VPS sensors are always the much maligned first suspect. I'd love to see anybody's video demonstration with a logical explanation of how that's actually possible.

There is no setting to disable the downward sensors in DJI Fly. You could cover them with a thin strip of gaffer's tape.

There is a configuration switch within the Litchi app to disable/enable Landing Protection, and/or Vision Positioning System altogether.

 

[msinger]

The downward sensors often receive unwarranted blame for lots of mysterious behaviour, but never with any reasonable explanation as to how that would be possible.

I guess you mean if we're solely going by what's recorded in publicly available DJI documentation. If so, then, yes, I agree that the downward sensors could not have caused the drone to auto descend.

Whenever something unexpected happens over water, the VPS sensors are always the much maligned first suspect

It's not my personal go-to suspect, but there are some abnormalities in the VPS data at the time of the incident in this case.

 

[BudWalker]

I guess you mean if we're solely going by what's recorded in publicly available DJI documentation. If so, then, yes, I agree that the downward sensors could not have caused the drone to auto descend.



It's not my personal go-to suspect, but there are some abnormalities in the VPS data at the time of the incident in this case.

In this case the vpsHeight data becomes valid only after the uncommanded descent starts and had been in effect for 1.5 secs. I.e., the uncommanded descent caused the vpsHeight data to become valid.


IMUCalcs:Height is the time integrated vertical velocity and shows that the MA2 had dropped about 1.1 meters when OSD:isSwaveWork became true (green background).

 

[Zbip57]

I guess you mean if we're solely going by what's recorded in publicly available DJI documentation. If so, then, yes, I agree that the downward sensors could not have caused the drone to auto descend.

I'm just really curious how that would work in any case. Many people seem to believe that VPS should always be disabled whenever flying over water, as that will somehow prevent a drone from auto-descending into water. I just don't understand by what mechanism VPS could ever cause a drone to auto descend into water.

I understand how the optical sensor can be fooled into tracking moving patterns instead of hovering in one fixed spot, but that won't make the drone descend. How can the VPS height sensor ever cause the drone to descend, since in all my previous video experiments the VPS height is either totally ignored or only ever makes the drone climb up if within 2ft height of an obstacle?

The DJI documentation is what had me confused. The Mavic Mini user manual says,

"The Downward Vision System and Infrared Sensing System help the aircraft maintain its current position, hover in place more precisely, and to fly indoors or in other environments where GPS is unavailable."
[...]
"When GPS is unavailable, the Downward Vision System is enabled if the surface has a discernible surface and sufficient light."

I know that without GPS, the optical sensor is used to hold position in hover. What I didn't know before, and what really surprised me because it makes no sense to me, is that the optical sensor takes priority even when GPS is available! Even with good GPS reception, my Mini will track and follow a patterned rug being pulled away under it, as shown in my most recent video. I wasn't expecting that.

[...] there are some abnormalities in the VPS data at the time of the incident in this case.

If the drone had initially been launched from water level, then the barometric height and VPS height would agree showing the same height above the water. If the drone in level flight then passes over the boat, the drone's height above water wouldn't change as it's being controlled by the barometric altimeter. But the VPS height would register the decreased ground clearance as it's passing over the boat because that sensor measures the actual distance between itself and anything passing below.

But the drone never reacts to any changes in recorded VPS height. The drone maintains level flight regardless of how much the VPS height changes as the terrain falls away or rises up underneath it, regardless or whether it's actively flying or hovering, unless it comes within 2ft height of anything below. Even then, it is programmed to respond by Landing Protection to automatically raise itself up to maintain at least 2ft clearance. It never responds by descending, unless you choose to trigger auto-landing by holding the throttle stick down from that 2ft height.

I'm just saying, neither the optical sensor nor the infrared height sensor of the VPS can ever cause the drone to descend. If anyone can explain and demonstrate a repeatable experiment showing how that could possibly make it descend, I'd love to see that.

I think the pressure change of the aerodynamic bow wave from the boat affecting the barometric sensor is a more plausible explanation. But the reaction seems backward and too dramatic. Increased pressure would be interpreted as an unwanted descent, causing the flight controller to command a climb. The drone dropped instead.

 

[MS Coast]

I agree, initially it was a bit of a stretch. But, I'm starting to like it more. IF a high pressure builds in front of the boat then a corresponding low pressure area has to exist somewhere. And, IF it looks like an airfoil pressure distribution then that low pressure area would be just behind the high pressure area that's in front of the boat. I.e., the low pressure are would be directly over the boat.

Reviewing the video the MA2 is over the boat when the incident occurs at video time 2:38 (.txt log time 297 secs). There is just one other time where the MA2 is over the boat which occurs at video time 1:10 ( .txt log time 209 secs).

It appears that the drone is still substantially ahead of the boat at 2:38. Is the camera tilt angle recorded in the log? It's definitely not near 90 degrees. Just offhand, I'd say it's less than 45 degrees

 

[BudWalker]

I think the pressure change of the aerodynamic bow wave from the boat affecting the barometric sensor is a more plausible explanation. But the reaction seems backward and too dramatic. Increased pressure would be interpreted as an unwanted descent, causing the flight controller to command a climb. The drone dropped instead.

As was said in post #31 if a high pressure region is caused by the bow wave there has to be a low pressure region somewhere. The net pressures have to sum to 0.0.

It appears that the drone is still substantially ahead of the boat at 2:38. Is the camera tilt angle recorded in the log? It's definitely not near 90 degrees. Just offhand, I'd say it's less than 45 degrees

View attachment 166827

The gimbal pitch is about -25°. That would put the low pressure region somewhere over the front of the boat, or in front of the boat, or over the bow wave, or etc. Looking at the video 2:38 secs and 1:10 secs the MA2 is in the same position relative to the boat. And, these are the only times this occurs. Looking at the .txt (times 297 secs and 209 secs) these are the only times where the altitude vs vertical velocity discrepancy is seen.

 

[Zbip57]

As was said in post #31 if a high pressure region is caused by the bow wave there has to be a low pressure region somewhere. The net pressures have to sum to 0.0.

I'd expect a low pressure area behind the boat, in the high drag area behind the transom. The drone was well ahead of the boat when it went down, where a high pressure area should be (if there is such a thing).

Looking at the stick inputs, he's pulling full rearward on the elevator stick to keep the drone flying backwards at high speed, then at 2:32 the stick is centred for a second braking the drone to a near halt, and immediately at 2:33 it's back to full rearward. A brief lift, then continued full back stick and at 2:38 the drone just drops.

It feels almost more like a voltage sag, except the voltage display in the video shows a constant 4.0v with only a momentary drop to 3.9v.

 

[BudWalker]

I'd expect a low pressure area behind the boat, in the high drag area behind the transom. The drone was well ahead of the boat when it went down, where a high pressure area should be (if there is such a thing).

Looking at the stick inputs, he's pulling full rearward on the elevator stick to keep the drone flying backwards at high speed, then at 2:32 the stick is centred for a second braking the drone to a near halt, and immediately at 2:33 it's back to full rearward. A brief lift, then continued full back stick and at 2:38 the drone just drops.

It feels almost more like a voltage sag, except the voltage display in the video shows a constant 4.0v with only a momentary drop to 3.9v.

All good points. But, I'm stuck believing the MA2 1) entered a low pressure region, 2) thought it was ascending, and 3) commanding a descent to compensate.

For me it's a choice between
1) Believing the altitude data was caused by some HW problem. It wasn't caused by a low pressure region. or
2) Accepting that I don't know enough about pressure distributions caused by turbulence to conclude where the low pressure region would be. I suspect that a lot would depend on data we don't have; e.g. the boat profile, boat and MA2 speeds, air density, etc.

It's an easy choice for me. I'm going with the latter.

 

[msinger]

I'm just really curious how that would work in any case. Many people seem to believe that VPS should always be disabled whenever flying over water, as that will somehow prevent a drone from auto-descending into water. I just don't understand by what mechanism VPS could ever cause a drone to auto descend into water.

Per DJI's documentation, the downward sensors could negatively affect the horizontal position of the drone when flying over moving/reflective surfaces (like in this case). I don't know of any advertised VPS functionality that would cause a drone to auto descend into water. However, that data is certainly available and DJI could use it to command the drone (even if they don't advertise they are doing it).

FWIW, objects close to the bottom sensors can cause the drone to automatically ascend. And I don't see that mentioned in DJI's documentation either.

I think the pressure change of the aerodynamic bow wave from the boat affecting the barometric sensor is a more plausible explanation.

That one is outside of my wheelhouse.