MA flew itself into Active Track subject

[sar104]

Reason and logic? The fact that the object was in the air is irrelevant. It wasn't moving so "a relatively slow moving person or car" is still moving infinitely faster than the aircraft I was tracking. Even if it had been moving I have the ability to control the speed at which the aircraft moves. It's possible for a VTOL aircraft, which quadcopters are, to move slower than cars and people typically do. Nothing in the documentation indicates that tracking an airborne object should be different than a land or waterborne object.

I agree - it has nothing to do with the speed of the object, at least in this case. As mentioned previously, the problem is almost certainly that the tracking algorithm expects the motion of the tracked object against the background to represent the actual motion of the object. That's essentially true for an object on the ground. However, for an object in the air, the apparent motion against the background is a combination of the actual motion of the object and the parallax change as the viewpoint changes. It almost certainly has no way to separate those, and that's going to lead to strange effects when it tries to follow.

 

[ravenflight]

Wrong. A flying object can move much more quickly out of frame and there by lose tracking much easier. Think about it! One slight move of the stick on the drone being recorded and it's out of frame.....and there is the distance factor also which can change dramatically and quickly too.

A car can move much faster than the drone can. Some cars even accelerate faster. But that doesn't matter. In my experience with it, running too fast will cause it to lose track. That doesn't mean I shouldn't be tracking running objects.

 

[joejaxvic123]

"Do not try this at home"...?

 

[ravenflight]

I agree - it has nothing to do with the speed of the object, at least in this case. As mentioned previously, the problem is almost certainly that the tracking algorithm expects the motion of the tracked object against the background to represent the actual motion of the object. That's essentially true for an object on the ground. However, for an object in the air, the apparent motion against the background is a combination of the actual motion of the object and the parallax change as the viewpoint changes. It almost certainly has no way to separate those, and that's going to lead to strange effects when it tries to follow.

Yeah, that sounds reasonable to me. If that is the case, I think this problem could arise in situations on the ground too. Say if it was tracking someone on top of a hill with distant trees in the background and it couldn't see much of the ground.

 

[FEMit]

I can understand that it may lose tracking ability if the target is moving too fast. I've had this happen several times, and the drone will just stop. For it to charge forward and crash into the target seems a bit ridiculous to me, whether it's airborne or not. They need to implement a better failsafe where the drone will just stop and wait for user input if it is at all confused. Guess they havent tested enough use cases, yet. hopefully they add this one, now.

 

[sar104]

Yeah, that sounds reasonable to me. If that is the case, I think this problem could arise in situations on the ground too. Say if it was tracking someone on top of a hill with distant trees in the background and it couldn't see much of the ground.

That is plausible.

 

[ravenflight]

That is plausible.

I think the drone's profile, particularly against the trees, probably played a role as well. But I wouldn't expect it to do anything more than just lose or never acquire the track.

 

[sar104]

I think the drone's profile, particularly against the trees, probably played a role as well. But I wouldn't expect it to do anything more than just lose or never acquire the track.

Not sure about that. The Inspire was very clearly visible. I think the Mavic just had no way to tell how far away it was.

 

[Cheech Wizard]

I suspect that the problem is that active track was intended to track an object on a stationary background. The Inspire was far from the background and the parallax was completely different. Interesting test but I don't think the software had any way to cope with it. Expensive test as a result.

Interesting theory.

 

[tcope]

Wow... just read the comments about how the failure of Active Track was some how due to not reading the manual. I'm sure the software knew that the person had not done this and this is why it failed.

Active Track is not perfect. As a matter of fact, DJI really has not done a very good job with it. Other companies have done a much better job with this type of software tracking.

In my testing of AT I've seen it fail first hand. The software is designed to fly the Mavic directly at the object it's tracking. Sometimes it looses track of that object but the Mavic still flies right toward it.

In the case below the Mavic lost me but still was flying right at me. I noticed this right at the last second and stopped walking. The Mavic missed me by about 2 feet. Had I kept walking it may have flown right into me.

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

 

[Dabrain]

It will be interesting to see what DJI says about tracking another AC. I hope they respond to the situation.

 

[v8juice]

When I first started experimenting with active track I took my bright red car out onto the street with the t-tops out and locked in on it from about 15 feet or so away and 4 feet above the roof line of the car behind me and to the left a few feet. I proceeded to drive down my street for at least a couple of hundred yards at no more than 10 mph or so. Everything was going great until I stopped for a stop sign. The Mavic proceeded to perform what I can only guess was a kamikaze attack and fly directly into the car with me landing in my lap. The impact with the left side of the seat apparently killed the motors and it did a somersault landing in my lap upright. Neither the Mavic, me or the car had a scratch on it. On top of that I realized I had forgotten to press the record button. The footage would have been great to add road runner/coyote sound effects to it.

 

[AnzacJack]

Are you saying I did something wrong? I don't see what.

this place is starting to get way more serious. I am seeing post after post where people have no sense of humour and very quick to throw up defences and go on the attack. It was always interesting, educational and fun on these forums, now its a constant battle to reread our posts to ensure no one will be offended, before we submit. Its bad enough living in an nanny state and having to be diplomatic in our work lives.

anyway, sorry for your loss.

 

[djorgen]

I guess we can learn from this that active track isnt fail proof and that it should be used with extreme caution; be careful and be sure to have enough of a "buffer zone" between the drone and the target in case anything happens and you need to cancel the tracking.

Also active track might work less optimal when tracking airborne targets (opposed to ground targets) as they might be harder to track/follow somehow?

 

[joejaxvic123]

@v8juice: Heck of a story, I can't believe you forgot to hit record, that would have been a great video. Have you tried it again?

Lately I've been wondering about these automated features, too much room for error. Precise take off and landing problems, rth going awrie, active track freak outs, etc. I haven't used any of them and I'm not sure if the trust level is high enough to do so.

 

[dwallersv]

Are you saying I did something wrong? I don't see what.

I wouldn't say that.

What I would say is this serves as another good real-world example to review and train oneself to be ready on the Sport Mode switch any time the bird is flying itself. As soon as it started to misbehave, if you'd thrown the Sport switch, you could have avoided the collision. There was plenty of time.

Again, not criticizing you at all. Most people don't train themselves on this emergency option, so never think of it when the pressure's on. I haven't (going to now). It's worth spending some regular training time doing emergency stops with the Sport switch while the bird is flying autonomously to get the muscle memory, and reflexive response.

 

[ac0j]

I wouldn't say that.

What I would say is this serves as another good real-world example to review and train oneself to be ready on the Sport Mode switch any time the bird is flying itself. As soon as it started to misbehave, if you'd thrown the Sport switch, you could have avoided the collision. There was plenty of time.

Again, not criticizing you at all. Most people don't train themselves on this emergency option, so never think of it when the pressure's on. I haven't (going to now). It's worth spending some regular training time doing emergency stops with the Sport switch while the bird is flying autonomously to get the muscle memory, and reflexive response.

This ^^^^
Even in these "automatic" flight modes, the pilot is ALWAYS in control. Any related crashes, unfortunately, are pretty much pilot error.

 

[Just_nick]

I wouldn't say that.

What I would say is this serves as another good real-world example to review and train oneself to be ready on the Sport Mode switch any time the bird is flying itself. As soon as it started to misbehave, if you'd thrown the Sport switch, you could have avoided the collision. There was plenty of time.

Again, not criticizing you at all. Most people don't train themselves on this emergency option, so never think of it when the pressure's on. I haven't (going to now). It's worth spending some regular training time doing emergency stops with the Sport switch while the bird is flying autonomously to get the muscle memory, and reflexive response.

The pause button is there for a reason

 

[FEMit]

The pause button is there for a reason

Yeah no need to switch sport mode, can just use the pause button. However, I think it would best if dji made it so that if the pilot touched any of the RC controls during active track/gesture controls, the MA would just stop. It can be hard to find the pause button in time, if you're not prepared.

 

[ravenflight]

this place is starting to get way more serious. I am seeing post after post where people have no sense of humour and very quick to throw up defences and go on the attack. It was always interesting, educational and fun on these forums, now its a constant battle to reread our posts to ensure no one will be offended, before we submit. Its bad enough living in an nanny state and having to be diplomatic in our work lives.

anyway, sorry for your loss.

It's not just this place. It's the whole Internet. I expect to be attacked anytime I post anywhere and I'm quick to be defensive. Too quick, sometimes. I mostly avoid forums for this reason.

I wouldn't say that.

What I would say is this serves as another good real-world example to review and train oneself to be ready on the Sport Mode switch any time the bird is flying itself. As soon as it started to misbehave, if you'd thrown the Sport switch, you could have avoided the collision. There was plenty of time.

Again, not criticizing you at all. Most people don't train themselves on this emergency option, so never think of it when the pressure's on. I haven't (going to now). It's worth spending some regular training time doing emergency stops with the Sport switch while the bird is flying autonomously to get the muscle memory, and reflexive response.

This ^^^^
Even in these "automatic" flight modes, the pilot is ALWAYS in control. Any related crashes, unfortunately, are pretty much pilot error.

The pause button is there for a reason

The pause button or the sport mode switch would both have aborted the operation had I hit one of them in time. Unfortunately I thought the Active Track was working at first and it wasn't until the last second that I realized it wasn't going to stop. I'm used to the Inspire controller and in that split second I drew a blank on what the abort action on the MA controller was (and I didn't have my finger on the switch like I typically do with the Inspire). All I did was pull back on the right stick, which did nothing. Even if I had been able to abort, the brake distance is large enough that the collision was probably unavoidable by then. I'm sure there's no way to convince ac0j that this does not mean I'm an incompetent idiot but I feel like equipment malfunctions don't fall under pilot error.