Mavic 3 crashes into a big wall with obstacle avoidance on. Video and logs

[mobilehomer]

Oh, you are exactly right about the Body's Automatic Functions, like breathing, heart beating, food digesting, standing, etc… and the body takes pretty good care of this without any interference on our part.

And in the case of a computer (the one core CPU), it also has its own "Body's Automatic Functions" like performing its priority subroutines like: updating the GPS, directing the power to each motor, receiving operator inputs (even when flying autonomously), interpreting inputs from its sensors (even when it sees nothing, it has to make that determination), and so much more.

The programming team for DJI assigns the Order of Presidency to each subroutine and whether it was the Drone was not paying "attention" to obstacle avoidance at the time of the crash or the OA was fooled by the lack of Contrast due to the white walls, we can only guess.

Of course, as I mentioned in another response on this topic, the Drone might have just got too close to the wall and got "sucked" in with no escape…

You need to watch a really good drummer. All four limbs doing something completely different at the same timed. Just like a multi-core, multi thread cpu.

 

[mobilehomer]

The answer to the crash has been given already, it's in the manual. OA does not detect monochrome obstacles.

 

[FlySkyJumper]

I am not an expert on obstacle avoidance, nor am I privy to DJI's obstacle avoidance firmware's algorithms, but I am a computer programmer and as much as we say that a computer can multi-task, it can't. It can do a lot of things at one time, but in reality, it really just does one thing at a time. It works on one task for a bit, and then it works on the next task, and so forth…

For example, I can make big breakfast: eggs, bacon toast, and orange juice… I turn on the stove, while the pan is heating, I put the bread into the toaster, I put the bacon into the pan, I check the toast, I put the eggs into the pan too. I flip the bacon, but I smell the toast burning, I run to the toaster, but it's too late, I jerk the burnt toast out, and put in fresh bread, then I smell the bacon burning, I run to take the bacon out, but it's really crispy. SO I pour the juice but now I smell the toast again. While taking the burnt toast out, I remember the eggs, but they too are over cooked, hard fried… So, you see, I also multi-tasked breakfast but only the juice gets served…

I would venture a guess that all those variable shapes and structures just overwhelmed the software. Did you notice the stuttering at time 0:42 and again at 0:49. When it crashed into the wall, the Drone was trying to follow you after it lost you as you walked past the wall and its processor running the software to find you and track you probably just took precedence over the obstacle avoidance routine.

View attachment 144782

I to am a seasoned programmer 25 years and going and I'm a little confused on what you're saying.

Isn't that the very definition of multi tasking when you say "It can do alot of things at one time..."

My point is you're some what right, but there is far more to this topic than programming. You're also trying to explain something most folks don't know about. The answer was already given above, which can be complex, but not when it's speculation. That just creates more confusion.

Especially when computers can literally perform millions of tasks at once aka multi tasking.

For simplicity the correct answer would be -

The drone does not detect monochrome obstacles.

This is a common problem with cameras and lenses with most any tech out at the moment.

 

[eaglesurfing]

I love post like this so others can learn.
This also helps the people who make prop guards and Gimbal guards make more money.
Then we also get to hear how smart other people think they are.
Also how great is that our "Staff Members" have to jump in and stop people who cannot control themself.
Thanks for the free entertainment even though it is from someone else expense.

 

[LoudThunder]

You need to watch a really good drummer. All four limbs doing something completely different at the same timed. Just like a multi-core, multi thread cpu.

It’s called “muscle memory,” but when that “really good drummer” started out, they might have had problems patting their head and rubbing their belly. Even “Commander Data” had problems with that one…

Pat Head And Rub Tummy

But, Iron Butterfly’s “In A Gadda Da Vida” drum solo is Magic…

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

In a Good World, we both can be right… so let’s agree to disagree…

 

[blackbirdsr72]

And here my uneducated guess is that the pure white obstacles are too hard for the sensors to detect separately, even with the subject in contrast.

 

[mobilehomer]

It’s called “muscle memory,” but when that “really good drummer” started out, they might have had problems patting their head and rubbing their belly. Even “Commander Data” had problems with that one…

Pat Head And Rub Tummy

But, Iron Butterfly’s “In A Gadda Da Vida” drum solo is Magic…

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

In a Good World, we both can be right… so let’s agree to disagree…

Obviously, you do not know any drummers. It is NOT muscle memory. Every song is different. Jazz is different than country or rock. I know guys that excell at ALL genre. You are flogging a dead horse.

 

[Aether]

What I find strange is why the drone didn’t halt once the target subject disappeared behind the wall. From past experience with the M2P when an object disappears or becomes significantly smaller in the defined tracking area the drone pauses.

I snapped a few screenshots of the video and bumped up the brightness and contrast to 100% and noticed there was still variation between pure white and the color of the environment (even while making the image monochrome). Perhaps what you see on the camera and the notable differences aren’t what the sensors see — just speculation. It is odd to me how the drone still tried to follow an object that was no longer present though. It would be nice if DJI offered settings where one could fine tune the tracking. I’ve never had much confidence in it and see waypoints as more reliable.

 

[EpicFlight]

I am not an expert on obstacle avoidance, nor am I privy to DJI's obstacle avoidance firmware's algorithms, but I am a computer programmer and as much as we say that a computer can multi-task, it can't. It can do a lot of things at one time, but in reality, it really just does one thing at a time. It works on one task for a bit, and then it works on the next task, and so forth…

Boy I hate doing this on these forums, but wrong is wrong.

The above is so wrong I don't know where to start. I'm an engineer, now retired, that was designing pipeline multicore processors in the 80s, with multiport RAM. I guarantee you there were multiple threads of execution taking place in parallel back then, doing "more than one thing" at exactly the same moment in time. Today...

Even single-core pipelined architectures can add two values while at the same time performing a cache access, during the same cycle. And even though modern compilers are very good at creating compact machine code with out-of-order operations, programmers STILL need to be aware of how parallel pipelined and multicore processors work and write parallel-aware high-level code if speed and efficiency is a major concern.

Surely you know this, as a programmer?

Strict serial processing hasn't been a limitation since the 1970s.

 

[EpicFlight]

What I find strange is why the drone didn’t halt once the target subject disappeared behind the wall. From past experience with the M2P when an object disappears or becomes significantly smaller in the defined tracking area the drone pauses.

ActiveTrack 5.0.

One of the marketed features is prediction of where a target is going to be if lost, so the drone attempts to continue predictive tracking for a few seconds, hoping to pick up the target again when clear.

 

[Eloy Nogueira]

Good Morning.
I don't know how it works in other countries, but in Brazil there is legislation for drones and a specific case arises that is not mentioned in the law, so it has to follow the general law.
You cannot make an inversion of values.
Example:
1) There is no specific law for the house: follow the general law until one arises.
2) If they make a law for drones indoors. The person flies in the house car garage. Is the law worth it?

Realize, the law is genetic and applies to ALL situations. It just doesn't apply if a SPECIFIC law has been made including an exception (example, flying indoors).

Is there any law that a person has to wear a seat belt in a car inside the 3rd basement of a mall? No need, the law is generic. It will only stop following the general, broader law, if they make a law that "you can ride without a seat belt in your car if you are in the 3rd basement of a mall"

Conclusion: the law applies everywhere and for all pilots. Only if, and only if, you have a specific law for the case you are in then you should follow the specific rule.

 

[Skywatcher2001]

So I've read thru this and no one mentioned the metal structure overhead which (most probably) caused interference.

On the older M2P . . . when inside a structure like this (several videos here with missions inside grocery stores); the satellites would drop considerably to a point where the bird goes into ATTI Mode which causes it to drift (without GPS hold). Maybe this is irrelevant in the M3 but the M2P suffered from this constantly. Just a thought . . .

 

[GadgetGuy]

So I've read thru this and no one mentioned the metal structure overhead which (most probably) caused interference.

On the older M2P . . . when inside a structure like this (several videos here with missions inside grocery stores); the satellites would drop considerably to a point where the bird goes into ATTI Mode which causes it to drift (without GPS hold). Maybe this is irrelevant in the M3 but the M2P suffered from this constantly. Just a thought . . .

In those other videos, it's likely the exceeding of the ground sensors' height limits for maintaining position, rather than interference from above, that is the problem.

 

[BigAl07]

Good Morning.
I don't know how it works in other countries, but in Brazil there is legislation for drones and a specific case arises that is not mentioned in the law, so it has to follow the general law.
You cannot make an inversion of values.
Example:
1) There is no specific law for the house: follow the general law until one arises.
2) If they make a law for drones indoors. The person flies in the house car garage. Is the law worth it?

Realize, the law is genetic and applies to ALL situations. It just doesn't apply if a SPECIFIC law has been made including an exception (example, flying indoors).

Is there any law that a person has to wear a seat belt in a car inside the 3rd basement of a mall? No need, the law is generic. It will only stop following the general, broader law, if they make a law that "you can ride without a seat belt in your car if you are in the 3rd basement of a mall"

Conclusion: the law applies everywhere and for all pilots. Only if, and only if, you have a specific law for the case you are in then you should follow the specific rule.

In the US of A the FAA has no authority over any true INDOOR flights. Now if the "arena" is not completely isolated from the National Airspace (no way for the aircraft to Autonomously/Accidently get into the NAS) it would fall under the FAA's regulations since the aircraft "could" end up in the NAS.

 

[Phantomrain.org]

Yes I learned I cannot play hide and go seek with my drone. Thank you

Phantomrain.org
Gear to fly in the M3 in the Rain.

 

[Paulark]

ActiveTrack 5.0.

One of the marketed features is prediction of where a target is going to be if lost, so the drone attempts to continue predictive tracking for a few seconds, hoping to pick up the target again when clear.

In my opinion, obstacle avoidance should overrule any other command the drone tries to execute. So yes, it may have tried the predictive tracking, but as soon as the sensors picked up the wall (and the sensors should have picked up the wall…they picked up all the other walls it flew by, that’s why I don’t buy into the monochrome theory) it should have stopped every other process and hovered.

For example, you’re following this hot gal over a bridge. She disappears behind a car then a section of the bridge between you and her vaporizes. Just because you are 99% sure she’s just behind a car and about to re-emerge, you don’t keep walking because….well, there’s no bridge and you fall to your death.

I don’t know, maybe depends on how hot she is….

P.S., if we could stay on topic and not diverge into the ethics or legality of the flight, that’d be great. With the prior knowledge of the flight the workers received, and being indoors, the flight was both ethical and legal, end of story.

 

[canadian drone videos]

Why did this happen?

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

Link to phantom help log

No contrast for the sensors to read, this is covered in the Mavic 3 manual.