MA flew itself into Active Track subject

[joejaxvic123]

Hey guys, sorry if i came off a little harsh on the automation topic. I guess I'm a little apprehensive about using them, with all the stories I hear. I know, I mostly hear the negatives. I guess I should have said use them, but always be prepared to hit the abort button.

 

[dwallersv]

I didn't know kamikaze was a possibility.

New feature in the latest firmware release. Didn't you read the Release Notes?

 

[ravenflight]

I at least got a cool shot of the I1R with the MA in manual before the incident.

I1R

• ravenflight
• Apr 11, 2018

 

[FEMit]

I at least got a cool shot of the I1R with the MA in manual before the incident.

I1R

• ravenflight
• Apr 11, 2018

any updates from DJI?

 

[ravenflight]

any updates from DJI?

From DJI:

NON-WARRANTY PER DATA ANALYSIS. Misoperation of Intelligent Flight Mode (Active Tracking) FLY008 1. Unit was in P-GPS mode and responsive to RC input. 2. At t=111s, user triggered Active Track via APP. 3. At t=112.8, unit started pitch forward toward object selected via app and collided with Mavic Air (information provided by customer) and fell to ground. Conclusion: Misoperation of intelligent flight mode. User selected the wrong object to track and was en route to designated location and made impact with another craft.​

I've responded asking for clarification about selecting the wrong object and what role obstacle avoidance should have played in the situation. They seem to be saying the object collided with the MA rather than the other way around so there may be a miscommunication about what happened. They want $469 to repair the Inspire and $444 to repair the X5R. I'd certainly prefer not to pay anything since I feel like this was clearly a product malfunction but, all things considered, it's not that bad.

 

[ravenflight]

DJI's response to my follow-up questions:

Obstacle avoidance has limited sensing areas, the object it may collide with also needs to be large enough and have sufficient contrast with the background. There are many more factors which are outlined in the user manual. The sensing size is roughly a 3 x 3 foot square, which is much larger than an Inspire. Obstacle avoidance was added to help prevent accidental collisions with walls, tree trunks, etc. Flying into a relatively small object like an Inspire, is not a failing with the obstacle avoidance, which is why the data analysis result is non-warranty.

In addition, using ActiveTrack on a smaller object, is not intended, which is why the title of the analysis is "Misoperation of Intelligent Flight Mode." The manual states it is capable of tracking people, vehicles, and boats.
​

I'm considering the matter closed and have gone ahead and paid the ~$1k repair bill. I still feel like this is a product malfunction but DJI is well within their rights to refuse to help. If the sensing size is 3' square then this may explain why it has a hard time tracking people, particularly children, since few people are 3' wide. Caveat emptor, I suppose. It's still a great little drone but calling the autonomous flight modes "intelligent" may be overstating the case.

 

[swooshdave]

I think people overestimate the capabilities of some of these automated functions. Packing the technology to actively track, avoid and still fly is not trivial. It has limitations. Tracking a flying drone is one of them.

 

[FEMit]

Yeah DJI clearly used all their outs to get out of helping a customer. I guess you need to spend more than a couple grand on their products for them to go the extra mile for you. Their "smart" features leave much to be desired. Hopefully they buy up Skydio's tracking system and implement it into future products. Their stuff is impressive.

 

[dwallersv]

I still feel like this is a product malfunction but DJI is well within their rights to refuse to help. If the sensing size is 3' square then this may explain why it has a hard time tracking people, particularly children, since few people are 3' wide.

I'm much more inclined to judge this a problem of poor/incomplete documentation, rather than a malfunction.

To be clear, AT is a very complex system of sensors, local processing (software) on the bird, and then more processing on the app. Only the engineers truly know if your situation was operating properly within the design limitations, and therefore not malfunctioning -- but rather was challenged with something it wasn't equipped to do. DJI seems to be claiming the latter, as an engineer myself the tech details and reasoning are solid, so I'm inclined to believe them.

Rather, as is often the case with sophisticated, complex tech like this, attempting to balance clear, simple operating instructions with detailed and complete enough information to avoid pilot error due purely to information ignorance failed in the latter.

This doesn't let DJI off the hook. They didn't tell you some of those pertinent limitations in detail that you're getting now as support detail. So you simply could not have avoided the situation. It's their fault, they should pay.

However, it IS important to identify the correct point of failure, so that the problem can be improved upon. For example, documenting the 3x3 ft detection window (and at what distance from the Drone? An appendix in the manual with exhaustive details about this stuff, reviewed by the engineering team, would be a big step forward), and some other things need to be added to the manual.

 

[dwallersv]

I think people overestimate the capabilities of some of these automated functions. Packing the technology to actively track, avoid and still fly is not trivial. It has limitations. Tracking a flying drone is one of them.

Agree, and this is largely borne of marketing hype setting expectations, then utterly inadequate documentation to know just how capable -- or not -- these features are in different circumstances.

I can (and have) reproduced a few of the more stunning AT examples seen in some of DJI's marketing vids. However, it's anything but "autonomous". The entire shot has to be planned and done over several takes to wring out the eccentricities of the feature.

Being less spectacular, like just following you and your dogs on the beach, and it comes much much closer to realizing the fully autonomous "camera operator" idea. I've run an entire battery on the beach for about 14m of video and several times completely forgot the thing was following me and the beasts around at about 30ft.

If I'm actually actively watching and controlling the aircraft while using AT on a subject, it can be exceedingly useful to relieve some of the workload so you can concentrate on the shot, composition, framing, etc. Just having the system control the gimbal and yaw to keep the subject in frame while flying is immensely useful when you learn how to make the feature work FOR you, rather that you working AGAINST the feature.

Like anything, it takes lots an lots of practice and experience. I'm still pretty much a piker, but I've been playing with it a lot since I got my Air (rear OA got me more interested in trying to use it seriously), and I actually pull AT out of my bag of tricks intentionally now to shoot certain things. It's become a useful tool, rather than a gimmick.

An owner will never "get" that, though, looking at it the way DJI sells it to them, and then trying to achieve the unachievable with little to no experience with the feature.

If you shoot mostly distance, vistas, forests, beaches, etc., then there's little point in spending the time investment to really "get to know" AT and how you can use it. However, if you shoot people and other subjects that move quite a bit relative to their background, it's worth diving deep on AT (that is, frustrating the hell out of yourself for a few dozen hours experimenting) because there will be situations where it will be a great help.

 

[dwallersv]

...it's worth diving deep on AT (that is, frustrating the **** out of yourself for a few dozen hours experimenting) because there will be situations where it will be a great help.

Really, REALLY? That word is too vile for this site? (h e double hockeysticks, perfectly suitable for broadcast TV round the clock and church on Sunday). Gawd, why did I start posting here again? And fellow members, Mavic brothers, now watch this post get deleted...

 

[Prismatic]

Really, REALLY? That word [ref: "frustrating the **** out of yourself"] is too vile for this site? (h e double hockeysticks, perfectly suitable for broadcast TV round the clock and church on Sunday). ...

Um ... you appear to have a very limited vocabulary regarding the four-letter intensifiers available in vernacular English.

 

[Deleted member 114292]

You’re all being punked. Totally fake video.

 

[The Editor]

You’re all being punked. Totally fake video.

Which video are you referring to?
There are two posted in this thread.

If you are referring to the OP’s video then it is not fake.

 

[Deleted member 114292]

Which video are you referring to?
There are two posted in this thread.

If you are referring to the OP’s video then it is not fake.

And you know this how

 

[Prismatic]

And you know this how

You're the one claiming it's fake. What's your evidence? Please share your expert digital video analysis with us.

 

[Deleted member 114292]

You're the one claiming it's fake. What's your evidence? Please share your expert digital video analysis with us.

“extraordinary claims require extraordinary evidence”
First off the "story" is hard to believe.
No one thing below is not possible, its when you add them up.
1. Discontinuity in the footage between the "crash" and the fake spin.
2. The fake spin - there is not a conceivable force that would cause the drone to spin up that fast, or at all for that matter.
3. The spin looks pieced together from stills
4. No stills of the damaged AC.
5. And the biggest NO FLIGHT LOGS. Show me the kmls of the two AC's converging.
Why would he not post this in a crash forum?
"It perceived the Inspire as an enemy." really? Y'all gonna swallow that one?
Personally, I find it quite amusing first that we would do this and second that you guys took the bait.
So really it's you believers that have little evidence.

 

[sar104]

“extraordinary claims require extraordinary evidence”
First off the "story" is hard to believe.
No one thing below is not possible, its when you add them up.
1. Discontinuity in the footage between the "crash" and the fake spin.
2. The fake spin - there is not a conceivable force that would cause the drone to spin up that fast, or at all for that matter.
3. The spin looks pieced together from stills
4. No stills of the damaged AC.
5. And the biggest NO FLIGHT LOGS. Show me the kmls of the two AC's converging.
Why would he not post this in a crash forum?
"It perceived the Inspire as an enemy." really? Y'all gonna swallow that one?
Personally, I find it quite amusing first that we would do this and second that you guys took the bait.
So really it's you believers that have little evidence.

You really should have read the thread. The flight characteristics and frame rate are consistent, and the behavior was caused by a lack of sophistication in the tracking algorithms when the object being tracked is nowhere near the background - in other words it couldn't cope with the parallax. An interesting and, in hindsight, not surprising event.

 

[Deleted member 114292]

You really should have read the thread. The flight characteristics and frame rate are consistent, and the behavior was caused by a lack of sophistication in the tracking algorithms when the object being tracked is nowhere near the background - in other words it couldn't cope with the parallax. An interesting and, in hindsight, not surprising event.

Great lets see the flight records.

 

[FEMit]

“extraordinary claims require extraordinary evidence”
First off the "story" is hard to believe.
No one thing below is not possible, its when you add them up.
1. Discontinuity in the footage between the "crash" and the fake spin.
2. The fake spin - there is not a conceivable force that would cause the drone to spin up that fast, or at all for that matter.
3. The spin looks pieced together from stills
4. No stills of the damaged AC.
5. And the biggest NO FLIGHT LOGS. Show me the kmls of the two AC's converging.
Why would he not post this in a crash forum?
"It perceived the Inspire as an enemy." really? Y'all gonna swallow that one?
Personally, I find it quite amusing first that we would do this and second that you guys took the bait.
So really it's you believers that have little evidence.

Great lets see the flight records.

LOL so glad you solved this deep fake video from two years ago ?.Thanks for showing all of us the light! /s