Joined
Sep 6, 2018
Messages
3
Reaction score
5
Age
56
Hello all, I’m new to the forum and a relatively new pilot, as I’ve had my P4 advanced only for a few months.
Unfortunately, I am writing after a total loss crash in which my drone ended up on the steep side of an unreachable coastal cliff. I would appreciate if someone more expert than me could give me some honest feedback on whether the accident was my fault (as DJI claims) or due to a malfunctioning of the RTH routine, as it is my strong belief.
I uploaded the flight log with this post and made it available also under Crash log - Google Drive

Summarizing briefly, I was driving the P4 advanced along a tall coastal cliff. It was a beautiful day, with great visibility, and the GPS signal was always present. After arriving at the visible edge of the cliff, curious to see what was on the other side, I sent the drone behind the edge. As expected, shortly thereafter the remote lost connection and the failsafe RTH routine was automatically launched. I was not too overly concerned as there was plenty of open space to fly back. What happened afterwards, however, left me speechless. In a situation in which the drone had a tall cliff on one side, but was completely open on all the others, the RTH, after controlling it for 53s, achieved the brilliant result of crashing it against the cliff. That made me speechless for a couple of reasons: with all obstacle avoidance tools the P4 advanced has, the RTH routine could not detect that the path was free in all directions but one, where there was a giant wall. Secondly, in the presence of good GPS signal, it could have simply adopted the dumb strategy of flying the outbound path backwards.

Anyway, the reply I got from DJI was in my opinion pure BS: the accident is my fault because
a) I should only fly in “open zones” and not in “improper environments” (not sure what’s more open than the coastline along the sea)
and
b) “the vision system cannot work perfectly” in situations with “dark light” and “complex topography”. It was around noon, full daylight, and I described above how complex the topography was, which will be visible clearly in the flight log.

Any feedback would be highly appreciated. Thank you in advance,
Marco
 

Attachments

  • DJIFlightRecord_2018-08-21_[13-24-24].txt
    369.9 KB · Views: 324
What was the RTH height? Remember that obstacle avoidance isn't a guarantee as it may not detect certain patterns. Without looking at the log, it sounds like a case of the RTH being too low for the surroundings.
 
Secondly, in the presence of good GPS signal, it could have simply adopted the dumb strategy of flying the outbound path backwards.
Unfortunatey, RTH does not work that way. The aircraft will climb to your specified altitude then fly the shortest route to the home point. Still looking through your data, but there is approximately 40 seconds of data missing due to downlink loss. I will say however, that immediately upon re-aquisition you were giving the aircraft full throttle up during the RTH procedure. That is neither here nor there at present. Just something to point out for the moment.
 
Judging by the elevation profile, your takeoff spot was much lower than the cliff sides. Your RTH setting was far too low. You would have needed at least 150m to clear the cliff.
 
  • Like
Reactions: Oso
Judging by the elevation profile, your takeoff spot was much lower than the cliff sides. Your RTH setting was far too low. You would have needed at least 150m to clear the cliff.
Just a quick look before taking my son to school. I'd have to agree with you @Fly Dawg . The RTH was far too low for the significant increase in elevation from HP.

Also, he flew out in strong wind which is always bad. See how the speed dropped from about 26mph to -0- almost instantly when RTH kicked in. It could not overcome the wind which was also mentioned in a warning during flight. That could have been overcome with the correct RTH altitude I think. Just a lesson to others reading this thread to account for high winds going out.

I'd have to also think that GPS may have been compromised by flying so dangerously close to the cliff face. I'm not an expert on that, but my guess is that accurate triangulation would have been difficult. Experts?

At first glance, I would have to agree with DJI on this one. I'll look again when I get home later today.

Ciao.

20180906 italnsd.JPG
 
Unfortunately, I am writing after a total loss crash in which my drone ended up on the steep side of an unreachable coastal cliff. I would appreciate if someone more expert than me could give me some honest feedback on whether the accident was my fault (as DJI claims) or due to a malfunctioning of the RTH routine, as it is my strong belief.
Your case is a difficult one to analyse and it has taken quite a lot to analyse and get an understanding of what happened.
The first thing I note is that you flew from a point about 30 metres above sea level and flew toward a cliff about 150 metres high, nearly 500 metres south of the home point..
You could have set the RTH height to clear any anticipated obstacles but left it at the default 30 metres.
You flew behind a part of the cliff about 350 metres away.
For much of the flight, you were close to the cliff and the VPS is indicating rocks 5-6 metres below the Phantom.
At 1:46, (380 metres from home) while flying at 10 metres/sec obstacle avoidance activated.

From this it would seem that you were flying out of visual range, fairly fast and within a few metres of the cliff edge.
I have only been flying for 4.5 years but that looks to be risky flying - too risky for me.
Tall sea cliffs can generate strong wind and updraft conditions. We don't know what the winds were like?

I sent the drone behind the edge. As expected, shortly thereafter the remote lost connection and the failsafe RTH routine was automatically launched. I was not too overly concerned as there was plenty of open space to fly back. What happened afterwards, however, left me speechless. In a situation in which the drone had a tall cliff on one side, but was completely open on all the others, the RTH, after controlling it for 53s, achieved the brilliant result of crashing it against the cliff. That made me speechless for a couple of reasons: with all obstacle avoidance tools the P4 advanced has, the RTH routine could not detect that the path was free in all directions but one, where there was a giant wall. Secondly, in the presence of good GPS signal, it could have simply adopted the dumb strategy of flying the outbound path backwards.
At 2:46, 576 metres from home and flying at full speed 12-13 metres/sec and lose signal for 12 seconds.
When signal is regained the Phantom is 10 metres further away and starting to RTH.
It is at the same altitude but is facing toward the home point and almost stopped with VPS indicating rocks 7 metres below.
One second later, you lose signal again for 23 seconds.
When you regain signal, the Phantom has ascended to 30 metres RTH height and is facing home with rocks 8-10 metres below and going slowly.
You use the left stick to climb another 10 metres to 40 metres and leave RTH to bring the Phantom home.
The speed climbs to 8 m/s at 3:32.4 but when the VPS height drops from 7 to 3 metres, the speed drops to almost zero.
My best guess here is that the Phantom is very close to the cliff and obstacle avoidance has slowed it?
Speed stays at <1 m/s until 3:46.8 when the speed increases to 7 m/s but the pitch and roll indicators suggest the Phantom has clipped the rocks and is not recovering well.
The VPS height shows that there are rocks just below the Phantom.
At 3:52.8 it spins to the left and tips up and the recorded data ends when signal is lost completely.
The whole time it was in RTH, the Phantom was pointing straight toward the home point with a cliff edge close to the east pf it.

I sent the drone behind the edge. As expected, shortly thereafter the remote lost connection and the failsafe RTH routine was automatically launched. I was not too overly concerned as there was plenty of open space to fly back. What happened afterwards, however, left me speechless. In a situation in which the drone had a tall cliff on one side, but was completely open on all the others, the RTH, after controlling it for 53s, achieved the brilliant result of crashing it against the cliff. That made me speechless for a couple of reasons: with all obstacle avoidance tools the P4 advanced has, the RTH routine could not detect that the path was free in all directions but one, where there was a giant wall. Secondly, in the presence of good GPS signal, it could have simply adopted the dumb strategy of flying the outbound path backwards.
The obstacle avoidance system is not magic and has a number of limitations which could have contributed to the crash, but the pilot could easily have moved the Phantom out from the cliff where there was nothing to hit and you would maintain signal.
The Phantom and it was very close to a huge obstacle and out of visual range.
I would have to ask why you stayed so close to the cliff where loss of signal was possible and the risk of collision always present.
Anyway, the reply I got from DJI was in my opinion pure BS: the accident is my fault because
a) I should only fly in “open zones” and not in “improper environments” (not sure what’s more open than the coastline along the sea)
and
b) “the vision system cannot work perfectly” in situations with “dark light” and “complex topography”. It was around noon, full daylight, and I described above how complex the topography was, which will be visible clearly in the flight log.
It's unfortunate that your Phantom was lost but I would have to side with DJI's assessment.
Your flying was very risky but there was no need to fly so close to the cliff, out of visual range and where you could lose signal.
You seem to have limited understanding of how RTH would work and too much faith in the obstacle avoidance system.
It is not magic and would have had trouble with the cliff edge being beside the Phantom as it was flown closer and closer.
The flight could have easily been carried out safely but it was the actions of the pilot that put the Phantom in a situation that caused the crash.
 
For what it is worth, just a different view, including the estimated terrain profile and flight altitudes. As you can easily see, 30m RTH is far too low. The cliff is over 100m. Although granted you were over 50m at the impact point, and RTH would have returned at 50m as opposed to descending to 30m.

CLIFFS.PNG
 
6BA8D13D-FBD2-42B8-8F11-C4E185664DF0.png
First of all, I want to thank everyone for the time spent on analyzing my log and giving me feedback. I learnt a lot from what all of you wrote.

Meta4 is correct, I had a poor understanding of how RTH works, as I didn’t know that it necessarily had to clear in height all the surroundings. However, I must say that the user manual is at best very misleading in describing failsafe RTH. I can’t really figure out how the sentence “The aircraft will plan its return route and retrace its original flight route home” could possibly describe the effective RTH procedure rather than my (wrong) interpretation of going backwards along the outbound path.

It is indeed fair to assume that i had too much trust into the obstacle avoidance capabilities. However, if it’s true that I flew very close to the cliff, upon taking control the RTH flew even closer to it. And I still have a hard time accepting the fact that even with the wrong value of RTH altitude set, the RTH would fly against such a, in Meta4’s words, “huge obstacle” while the path was clear on all other directions. This was not the classic easy-to-miss tall tree, it was a freakin giant cliff lol. I don’t know the specifics of the RTH algorithm, and it is my fault, but I must say I am underwhelmed. In an idealized extreme scenario in which I steered the Phantom around an infinitely tall wall, but of limited width, and I started RTH, the algorithm would always result in a crash by design.

Another assumption I made is that once the RTH takes over, the flight is all automated, and the pilot can re-assume control only by canceling it. From Meta4’s description of my flight “You use the left stick to climb another 10 metres to 40 metres and leave RTH to bring the Phantom home” this appears to be another wrong assumption of mine. This assumption in turn led to my belief that RTH would choose the safest possible path back home depending on the information available (GPS, vision system, etc.)

Finally, Fly Dawg, thank you for posting that great 3D flight path pic. Can I ask you how you generated it? I created a kml file from my log on airdata.com and imported on Google Earth, but I only managed to get the much crappier representation below.

Again, thank you all for sharing your expertise.
 
the user manual is at best very misleading in describing failsafe RTH. I can’t really figure out how the sentence “The aircraft will plan its return route and retrace its original flight route home” could possibly describe the effective RTH procedure rather than my (wrong) interpretation of going backwards along the outbound path.
The manual is a bit sketchy on this and it's taken users a bit of experimenting (in safe locations) to work this out.
In earlier models, the Phantom would just climb to RTH height and come straight home.
With later models and greater processing power, on losing signal the Phantom will retrace its path until it regains signal - but it can only do this for a short path, maybe 20-30 seconds.
This is unough to regain signal im most situations.
If signal is not regained, the Phantom climbs and goes to RTH normally.
It is indeed fair to assume that i had too much trust into the obstacle avoidance capabilities. However, if it’s true that I flew very close to the cliff, upon taking control the RTH flew even closer to it. And I still have a hard time accepting the fact that even with the wrong value of RTH altitude set, the RTH would fly against such a, in Meta4’s words, “huge obstacle” while the path was clear on all other directions. This was not the classic easy-to-miss tall tree, it was a freakin giant cliff lol. I don’t know the specifics of the RTH algorithm, and it is my fault, but I must say I am underwhelmed. In an idealized extreme scenario in which I steered the Phantom around an infinitely tall wall, but of limited width, and I started RTH, the algorithm would always result in a crash by design.
If the obstacle had been a tree or a cliff and it was right in front of the Phantom (and lighting wasn't a problem), it probably would have easily avoided it.
The issue was trickier because your RTH path was almost parallel to the cliff face but slowly converging toward it.
The obstacle wasn't in front of the obstacle avoidance sensors where they would have easily detected it..
This assumption in turn led to my belief that RTH would choose the safest possible path back home depending on the information available (GPS, vision system, etc.)
RTH was trying to come straight home.
If the sensors had detected an obstacle in front, they could have taken appropriate action but the cliff was in the blind spot and slowly getting closer.
 
  • Like
Reactions: italnsd
Thanks Meta4, now I can make much more sense of the dynamic of the accident. It’s just a pity that I did not come to this forum before I started flying. Would you happen to know what’s the max horizontal angle from the forward direction at which the P4 can detect obstacles? And vertically?

Thank you again!
 
Thanks Meta4, now I can make much more sense of the dynamic of the accident. It’s just a pity that I did not come to this forum before I started flying. Would you happen to know what’s the max horizontal angle from the forward direction at which the P4 can detect obstacles? And vertically?

Thank you again!

I have learned a lot on this site by reading the unfortunate accidents of others. This is a great web site for learning.

I am usually flying in a heavily wooded area here in Georgia. I always set the RTH altitude way higher than needed to be safe. I tested the RTH a few times. It works well. Just be sure to set the altitude correctly and have a GPS lock before launching. The RTH altitude is distance above the home point. Don't forget to note that if flying above rising terrain.

It is a shame your drone is lost. It is an expensive way to learn.
 
Thanks Meta4, now I can make much more sense of the dynamic of the accident. It’s just a pity that I did not come to this forum before I started flying. Would you happen to know what’s the max horizontal angle from the forward direction at which the P4 can detect obstacles? And vertically?
According to the specs:
Vision System - Field of View
Forward: 60°(Horizontal), ±27°(Vertical)
Backward: 60°(Horizontal), ±27°(Vertical)​
So it's seeing 30° either side of straight ahead.
 
  • Like
Reactions: JWH

Members online

Forum statistics

Threads
143,066
Messages
1,467,358
Members
104,936
Latest member
hirehackers