Mavic Air Fly-away

[joejaxvic123]

It's starting to hit me like...a ten ton... heavy something...

 

[sar104]

Please tell me where do you see 2 compass readings in my attached screenshots of Mavic Air vs Mavic Pro?
Anyone with Mavic Air and/or Pro can check themselves and compare with my screenshot to verify whether you're right or me.

So you are basing your assertion on the compass calibration screen in the app - just because it doesn't show two compasses? That's suggestive but not really conclusive.

On the other hand, these are the magnetometer data streams in the Mavic Air DAT files:

IMU_ATTI_0:mx0
IMU_ATTI_0:my0
IMU_ATTI_0:mz0
IMU_ATTI_1:mx1
IMU_ATTI_1:my1
IMU_ATTI_1:mz1

The data are similar, but different, as would be expected from different magnetometers, the same as with the MP and P4 - hence my conclusion that there are two magnetometers. If you have an alternative explanation for those streams I'd be happy to hear it.

 

[Rnl]

Sar104 wouldn't even respond back, your work for our community is and has been a blessing. There's one in every group. He knows not what he does.

 

[ac0j]

Sar throws the big right hook aaaaand the opponent is wobbled........

 

[sar104]

For reference, these are the x-axis magnetometer data mx0 and mx1, just to illustrate that there are two sets of distinct data:

 

[sar104]

Sar104 wouldn't even respond back, your work for our community is and has been a blessing. There's one in every group. He knows not what he does.

Sar throws the big right hook aaaaand the opponent is wobbled........

I'm perfectly happy to have the discussion, even if his manner is a bit off-putting. And while the evidence looks rather clear in my opinion, there may be other explanations that I haven't thought off.

 

[ac0j]

The extra jab doesnt hurt though......

 

[Prismatic]

Interesting thought. But as you'll see above, I didn't attack anyone - merely stated the truth. Then it's him who attacked my statement with falsehood. Doesn't matter if someone is highly skilled or even the president, if they use falsehood to make incorrect claims, they are still wrong.

You're right; an untruth is an untruth, though it's funny you should use His Orangeness as your foil in this instance. Because like him, you don't seem to even know when: You. Have. Been. *ahem* .... Owned!
On a lighter note, I offer you and our fellow followers this:

 

[rfc]

Good luck. You are welcome to use anything that I've posted.

DJI should put you on their payroll. Not only do you save them dealing with a good number of erroneous "my drone flew away" situations that turned out to be pilot error, I'm convinced many of these analyses would be of great value to their engineers.

 

[joejaxvic123]

DJI should put you on their payroll. Not only do you save them dealing with a good number of erroneous "my drone flew away" situations that turned out to be pilot error, I'm convinced many of these analyses would be of great value to their engineers.

He's an undercover agent for the dji....

 

[RayOZ]

fascinating. new guy comes in guns blazing, talking like he's an expert (which just boils down to a screenshot), telling a veteran who has helped to find countless lost drones from dissecting flight logs, that he needs to learn a few things.

 

[laurens23]

Now that’s how you demolish a blowhard: calmly and with rock solid evidence.
[…cue the crickets, unlikely he’ll be back …]

well there is no need to treat Zaheed like he treated others... besides that his comments also have some merit. The Mavic air has Dual IMU , one "traditional" compass and a 'vision compass'

From the Mavic Air page on dji.com:
"Besides dual IMUs, FlightAutonomy 2.0 has more redundant modules to increase the Mavic Air’s safety level. The Vision Compass will assist the navigation system to estimate flight direction when the compass experiences interference. When only one direction of the front, bottom, or rear vision systems is functioning normally, the aircraft will still position itself successfully. There are two sets of sensor fusion algorithms based on two independent hardware platforms. When an error occurs to one set of algorithms, the aircraft will switch to the other pair seamlessly to resume normal functionality."

So @sar104 is right that there are 2 sets of compass measurements, one from the traditional compass and another from the "vision compass"
but it looks like @Zaheed is also correct in that there is only 1 traditional compass (and a "vision compass") in the Air while the pro clearly has 2 traditional compasses.

Unfortunately a quick search with google/Youtube and in skimming through some of DJI patents did not reveal what exactly this "visual compass" is or how it works.

 

[Jameslavis]

It’s ok to have a discussion, this guy came in like a troll. My air has been solid without any issues. His gripes can live in the MP forum

 

[sar104]

well there is no need to treat Zaheed like he treated others... besides that his comments also have some merit. The Mavic air has Dual IMU , one "traditional" compass and a 'vision compass'

From the Mavic Air page on dji.com:
"Besides dual IMUs, FlightAutonomy 2.0 has more redundant modules to increase the Mavic Air’s safety level. The Vision Compass will assist the navigation system to estimate flight direction when the compass experiences interference. When only one direction of the front, bottom, or rear vision systems is functioning normally, the aircraft will still position itself successfully. There are two sets of sensor fusion algorithms based on two independent hardware platforms. When an error occurs to one set of algorithms, the aircraft will switch to the other pair seamlessly to resume normal functionality."

So @sar104 is right that there are 2 sets of compass measurements, one from the traditional compass and another from the "vision compass"
but it looks like @Zaheed is also correct in that there is only 1 traditional compass (and a "vision compass") in the Air while the pro clearly has 2 traditional compasses.

Unfortunately a quick search with google/Youtube and in skimming through some of DJI patents did not reveal what exactly this "visual compass" is or how it works.

I did wonder whether one of the compasses might not be a magnetometer, but the problem is that it's easy to show the data from both are 3-axis components of the earth's magnetic field. I guess it is possible that the vision system is being used to generate synthetic magnetometer data but I cannot imagine why the firmware would want to do that since the Kalman filter only needs yaw values - it doesn't care about the magnetic field per se.

 

[sar104]

well there is no need to treat Zaheed like he treated others... besides that his comments also have some merit. The Mavic air has Dual IMU , one "traditional" compass and a 'vision compass'

From the Mavic Air page on dji.com:
"Besides dual IMUs, FlightAutonomy 2.0 has more redundant modules to increase the Mavic Air’s safety level. The Vision Compass will assist the navigation system to estimate flight direction when the compass experiences interference. When only one direction of the front, bottom, or rear vision systems is functioning normally, the aircraft will still position itself successfully. There are two sets of sensor fusion algorithms based on two independent hardware platforms. When an error occurs to one set of algorithms, the aircraft will switch to the other pair seamlessly to resume normal functionality."

So @sar104 is right that there are 2 sets of compass measurements, one from the traditional compass and another from the "vision compass"
but it looks like @Zaheed is also correct in that there is only 1 traditional compass (and a "vision compass") in the Air while the pro clearly has 2 traditional compasses.

Unfortunately a quick search with google/Youtube and in skimming through some of DJI patents did not reveal what exactly this "visual compass" is or how it works.

An interesting test, for anyone who has a MA, might be to disturb the local magnetic field with a magnet while the aircraft is powered up and on the ground, without moving the aircraft, and then check the two "magnetometer" sets of data to see if both m0 and m1, or only m0, detected the change.

 

[Prismatic]

well there is no need to treat Zaheed like he treated others

Agreed. Apologies to all. I've deleted it.

 

[Prismatic]

An interesting test, for anyone who has a MA, might be to disturb the local magnetic field with a magnet while the aircraft is powered up and on the ground, without moving the aircraft, and then check the two "magnetometer" sets of data to see if both m0 and m1, or only m0, detected the change.

Apparently some MA's generate readable DATs on the aircraft, and some do not. Mine is the obtuse kind, a point of some frustration. If the DAT from the tablet suffices, let me know. Otherwise, I hope someone with the former type will perform that test and put their DATs into the hands of the resident analysts here.
As a retired engineer, I can easily imagine a vision-based pseudo-compass, but it would a challenge to make it sufficiently reliable for the purpose. For example, imagine flying on an overcast day offshore, filming whales while shreds of fog drift by. Even lobstermen use a compass on those days. Or imagine the aircraft gets walloped by a gust of wind. Sensors in two axes briefly point at the sky; on recovery of attitude, it may be impossible to re-establish orientation due to unknown yaw changes during the event.
That said, a drone only needs a magnetic compass to anchor its frame of reference to the earth's FoR on-the-fly. Any functional and sufficiently reliable system that does that will suffice, and a non-magnetic solution would have obvious benefits.
I will say that a vision-based solution would necessarily be really computationally intensive! (My Master's leaned heavily on computability theory applied to graphics, so I'm not just guessing.) That probably means dedicated processing with associated extra power drain, and it still requires a fail-safe that saves your bacon when it loses its visual fix in difficult circumstances.
Still, it's an interesting concept, and apparently DJI has at least a partial solution. A reliable anchor to the earth's FoR that is impervious to outside influences of any kind ... that's gotta be a kind of Holy Grail for drone engineers!

 

[Kevjones777]

In sympathy with this rather interesting discussion and given that I do not understand stuff like magnetometers and meridian cut offs following a linnea curve instead of the parabolic inverse. I have removed all compasses from my MA with immediate effect. It can now no longer fly off course and never does a RTH either as it, like me has no idea anymore where home is.

These posts are stupendous and the one of the most interesting discussions while you are waiting to fly.
@sar104 thanks bud, your stuff is greatly appreciated.

 

[sar104]

Apparently some MA's generate readable DATs on the aircraft, and some do not. Mine is the obtuse kind, a point of some frustration. If the DAT from the tablet suffices, let me know. Otherwise, I hope someone with the former type will perform that test and put their DATs into the hands of the resident analysts here.
As a retired engineer, I can easily imagine a vision-based pseudo-compass, but it would a challenge to make it sufficiently reliable for the purpose. For example, imagine flying on an overcast day offshore, filming whales while shreds of fog drift by. Even lobstermen use a compass on those days. Or imagine the aircraft gets walloped by a gust of wind. Sensors in two axes briefly point at the sky; on recovery of attitude, it may be impossible to re-establish orientation due to unknown yaw changes during the event.
That said, a drone only needs a magnetic compass to anchor its frame of reference to the earth's FoR on-the-fly. Any functional and sufficiently reliable system that does that will suffice, and a non-magnetic solution would have obvious benefits.
I will say that a vision-based solution would necessarily be really computationally intensive! (My Master's leaned heavily on computability theory applied to graphics, so I'm not just guessing.) That probably means dedicated processing with associated extra power drain, and it still requires a fail-safe that saves your bacon when it loses its visual fix in difficult circumstances.
Still, it's an interesting concept, and apparently DJI has at least a partial solution. A reliable anchor to the earth's FoR that is impervious to outside influences of any kind ... that's gotta be a kind of Holy Grail for drone engineers!

I was working from mobile DATs in all the recent MA cases - they contain all the necessary data.

The vision-based compass is interesting - I guess they have traded components for added computational complexity, which suggests that processing power is not a limiting factor. But the issue remains as to the nature of the m1 data and whether or not it is from a "vision compass" or a second magnetometer - why would the IMU or FC go to the trouble of disguising vision-based yaw data as 3-D magnetic field data? I'll see if I can make some enquiries directly with DJI.

 

[OldBlueHeron]

Subscribed. This is all ridiculously fascinating, and way over my head. I am looking forward to learning from this.