EVO 2 Crash while Dynamic Tracking

[Mike Mas]

It's 100% your own fault - you took off to quick and there were far too many obstacles. It's a drone with avoidance not an AI robot. Try you stunt elsewhere!

Fly Safe -Mike

 

[AstroNoot]

It's 100% your own fault - you took off to quick and there were far too many obstacles. It's a drone with avoidance not an AI robot. Try you stunt elsewhere!

Fly Safe -Mike

Well, I'm pretty sure that Autel markets it as an AI robot.
Even on the box it says "fully autonomous".

Autel already determined it was their fault and their engineers will fix it

 

[Mike Mas]

I just don't get why someone would spend 3 grand on a precision drone then go do something as stupid as attempt it to follow thru the woods. What benefit is there so you can post the video on the forum like others? You totally eared this crash and its no ones fault but your own!

 

[AstroNoot]

I just don't get why someone would spend 3 grand on a precision drone then go do something as stupid as attempt it to follow thru the woods. What benefit is there so you can post the video on the forum like others? You totally eared this crash and its no ones fault but your own!

Whatever man, just your opinion. The manual clearly indicates this is a standard scenario, and it should have handled it well. You can be damn sure that I'd expect my "3 grand" drone to avoid obvious trees with it 12 OA sensors.

The fact that Autel offered me a replacement, if I want, confirms that this drone isn't supposed to behave this bad in a forest environment.

Cheers.

 

[Mike Mas]

Astro - You're missing the whole point of obstacle avoidance - it's "NOT" designed as a "Play Toy" for children so they can run through trees and brush hiding from the drone therefore confusing the flight logic. It's in place so in the event a person or object is in the flight path, it will hopefully avoid an accident.

Just never understood why anyone would want to risk their drone doing obvious goofy stunts like tree dodging or checking to see if their drone can fly 8 miles away. These are not toys, they are designed for professional aerial photography, use it for its intended design.

If you want to play - get yourself a DJI Mini2 from Walmart for $299

Fly Safe - Mike

 

[AstroNoot]

Astro - You're missing the whole point of obstacle avoidance - it's "NOT" designed as a "Play Toy" for children so they can run through trees and brush hiding from the drone therefore confusing the flight logic. It's in place so in the event a person or object is in the flight path, it will hopefully avoid an accident.

Just never understood why anyone would want to risk their drone doing obvious goofy stunts like tree dodging or checking to see if their drone can fly 8 miles away. These are not toys, they are designed for professional aerial photography, use it for its intended design.

If you want to play - get yourself a DJI Mini2 from Walmart for $299

Fly Safe - Mike

Are you even aware of what dynamic track is? It's a fully autonomous flight mode that lets the drone follow a subject.

"Its intended purpose"

^ this is from Autel's official channel, by the way

 

[Mike Mas]

The flight log player in the app is created for a beauty and not for accuracy. Download the tool from that video description and check flight records yourself.

22th second, sat.count = 14
View attachment 10699
25th second, sat.count = 0, gps lost
View attachment 10700

Loss of satellites is not what made it hit the tree. Satellites provide no data for obstacle avoidance, in fact during these tasks the GPS is piratically in a standby mode.

Avoidance logic makes its decisions mostly from obstacles location and distance, drone altitude, speed, position and more importantly "clear space" to control the drone. In this case, the satellites were only there to provide "position" which was over-ridden second by second by the tracking logic. The only time the satellites would come to play would be if the drone cancelled the tracking event, then it would use GPS or ground cameras to hover in place.

Another point is there is no way the satellites would go to zero in that short period of time. Could be a glitch in the logs or a problem with the GPS processor.

Regarding the Autel video keep in mind they could have gone through 20 drones to get the piece they were fishing for. Do you also believe those pretty blue rings shoot out from the sensors are real too? It's all Hollywood smoke and mirrors. "Read my lips - nothing you see or read is 100% accurate. The most important question you can ask yourself is why did you do such a foolish thing!

To clarify myself, I'm not saying the Autel won't go through the woods, the EVO has possibly one of the best avoidance systems. What I am saying is you should not be testing it since unlike Autel, it will cost you big time to play!

Mike

 

[Jiri Bakala]

Last week on a shoot I had a similar crash in Dynamic Tracking mode. Unlike AstroNoot, in my case we were on a clear paved path lined with trees on the right side only. The mode behaves spastically and unpredictably. All I wanted it to do is to follow the gentleman on the longboard in a straight line. I started the flight straight behind him and instead of simply following the drone would swing quite violently from side to side. I don't think it's a software error as someone suggested but instead a programming issue. In other words, these crashes are likely to happen in similar circumstances.

 

[Jiri Bakala]

Loss of satellites is not what made it hit the tree. Satellites provide no data for obstacle avoidance, in fact during these tasks the GPS is piratically in a standby mode.

Avoidance logic makes its decisions mostly from obstacles location and distance, drone altitude, speed, position and more importantly "clear space" to control the drone. In this case, the satellites were only there to provide "position" which was over-ridden second by second by the tracking logic. The only time the satellites would come to play would be if the drone cancelled the tracking event, then it would use GPS or ground cameras to hover in place.

Another point is there is no way the satellites would go to zero in that short period of time. Could be a glitch in the logs or a problem with the GPS processor.

Regarding the Autel video keep in mind they could have gone through 20 drones to get the piece they were fishing for. Do you also believe those pretty blue rings shoot out from the sensors are real too? It's all Hollywood smoke and mirrors. "Read my lips - nothing you see or read is 100% accurate. The most important question you can ask yourself is why did you do such a foolish thing!

To clarify myself, I'm not saying the Autel won't go through the woods, the EVO has possibly one of the best avoidance systems. What I am saying is you should not be testing it since unlike Autel, it will cost you big time to play!

Mike

We shouldn't call anyone foolish for doing *exactly* what Autel shows as the use and purpose of a mode. I would like to remain constructive here and have Autel see, respond, and hopefully correct whatever is wrong with the tracking mode. Or, alternatively, issue a disclaimer on what NOT to do in that mode.

 

[Jiri Bakala]

Further to the behaviour of the tracking; I wish that it would attempt to stay close to the distance and angle of where its set up. What I mean is if you start the tracking behind the object, I wish it stayed behind and not try to get to more of a 45° angle, which it seems to try to do.

 

[AstroNoot]

Autel hooked me up with a new set of props, and assured me to get back with an email regarding what the issue was and how will they fix it. I told Gian, from support, that it would be better for it to be a hardware issue, rather than a software issue. If it's software, this means any Evo can randomly kiss a tree, in more or less complicated tracking scenarios.

Thanks @Jiri Bakala for your input

 

[AstroNoot]

Loss of satellites is not what made it hit the tree. Satellites provide no data for obstacle avoidance, in fact during these tasks the GPS is piratically in a standby mode.

Avoidance logic makes its decisions mostly from obstacles location and distance, drone altitude, speed, position and more importantly "clear space" to control the drone. In this case, the satellites were only there to provide "position" which was over-ridden second by second by the tracking logic. The only time the satellites would come to play would be if the drone cancelled the tracking event, then it would use GPS or ground cameras to hover in place.

Another point is there is no way the satellites would go to zero in that short period of time. Could be a glitch in the logs or a problem with the GPS processor.

Regarding the Autel video keep in mind they could have gone through 20 drones to get the piece they were fishing for. Do you also believe those pretty blue rings shoot out from the sensors are real too? It's all Hollywood smoke and mirrors. "Read my lips - nothing you see or read is 100% accurate. The most important question you can ask yourself is why did you do such a foolish thing!

To clarify myself, I'm not saying the Autel won't go through the woods, the EVO has possibly one of the best avoidance systems. What I am saying is you should not be testing it since unlike Autel, it will cost you big time to play!

Mike

You found that out going on a hunch...or ?
I'm pretty sure you don't actually know the decision tree of the tracking algorithm, since its 100% proprietary software.
The GPS system is never in "standby" mode. How else would it actually display the current distance from the takeoff point, how would it log its position every 100ms?

Also, it would be really stupid for a company with "Robotics" in its name to not use redundancy when doing an autonomous task. I'm clearly going on a hunch here, but i'm sure that the flight controller definitely uses GPS too, to determine its own speed. No one would rely only on binocular optic sensors with a max range of 30 meters when chasing cars, for example.

Please refrain from treating me like a kid. "Read my lips:" What kind of company would even push a feature for a product that has such a high failure rate?
Yeah, maybe they went through 20 prototypes to make it work, but definitely they didn't go through 20 drones to get that video

It's quite stupid to defend Autel like that. I like the company too, but you can't sweep this under the rug. The OA system on this drone is not a safety-only feature. It's part of its core functionality. The starpoint system, the dynamic tracking...

I'm not testing the dynamic track. I'm using it. By the book. Testing was done by Autel, i suppose.

To go by your words, I'd just go to Walmart and get myself an *asian brand* Air 2. Same camera sensor, crappy NFZ system, but guess what. It'll stop in front of a tree .

Thanks for your input nonetheless. I'll post the response from the engineer team, i'm curious about what the cause was, too.

----------------------------


@Jiri Bakala Yeah, i think Autel really wants to go for the "dynamic" in dynamic track.

Here's another video, shot minutes before the crash shown in the first post. There's clearly room for it to track the subject on the side with no trees, but it chose to go through the tiny gap in the branches

Even the Skydio is way more "calm".


I can't wait to see how Autel will improve in time.

 

[kenautelevo2pro]

so astro when you get your drone "repaired" will you go back out and do the same thing again to see if it "works" now?

look, there's a thing called "marketing." obviously autel wants their drone to compete and they have made a solid attempt to have their drone perform. no different than car companies who market their suvs but those car crash and get stuck all the time. it's not going to be flawless and autel is not going to tell you the feature is sometimes flawed and they're not going to put out a video with crashes. instead, they will try to take care of you best they can in the event you have an accident.

honestly, it's up to you do decide for yourself whether you believe what you are being told and shown. don't claim autel is "pushing" this on you....put your drone up in the air, test it out, and then push back. it really look week when someone continues to believe in what they are "told" instead of figuring it out for themselves; learn to realize when you're not being told....

in the end, you might conclude this feature, altho present, is simply not up to par and not very capable and then you move on with your life since there are a million other challenges to deal with. again, I have a skydio2. and after I fly the skydio, I know "capability" when I see it...and the autel doesn't have what it takes. no amount of user manuals, yt videos, or "sensors" can convince me otherwise, c'mon you're smarter than that. so what do I do? I use the autel for what it's great at: 6k stunning videos, solid long range and connectivity, long battery life.

you're never going to catch me trying to force my autel to perform at the level of my skydio, that's just ridiculous.

 

[Jiri Bakala]

You found that out going on a hunch...or ?
I'm pretty sure you don't actually know the decision tree of the tracking algorithm, since its 100% proprietary software.
The GPS system is never in "standby" mode. How else would it actually display the current distance from the takeoff point, how would it log its position every 100ms?

Also, it would be really stupid for a company with "Robotics" in its name to not use redundancy when doing an autonomous task. I'm clearly going on a hunch here, but i'm sure that the flight controller definitely uses GPS too, to determine its own speed. No one would rely only on binocular optic sensors with a max range of 30 meters when chasing cars, for example.

Please refrain from treating me like a kid. "Read my lips:" What kind of company would even push a feature for a product that has such a high failure rate?
Yeah, maybe they went through 20 prototypes to make it work, but definitely they didn't go through 20 drones to get that video

It's quite stupid to defend Autel like that. I like the company too, but you can't sweep this under the rug. The OA system on this drone is not a safety-only feature. It's part of its core functionality. The starpoint system, the dynamic tracking...

I'm not testing the dynamic track. I'm using it. By the book. Testing was done by Autel, i suppose.

To go by your words, I'd just go to Walmart and get myself an *asian brand* Air 2. Same camera sensor, crappy NFZ system, but guess what. It'll stop in front of a tree .

Thanks for your input nonetheless. I'll post the response from the engineer team, i'm curious about what the cause was, too.

----------------------------


@Jiri Bakala Yeah, i think Autel really wants to go for the "dynamic" in dynamic track.

Here's another video, shot minutes before the crash shown in the first post. There's clearly room for it to track the subject on the side with no trees, but it chose to go through the tiny gap in the branches

Even the Skydio is way more "calm".


I can't wait to see how Autel will improve in time.

I agree that the Dynamic Track is too spastic. I would much prefer if it simply followed the subject in a simple trajectory matching as much the subject's movement as possible. Of course the function IS INDEED part of its core functionality and we use it not to be "daring" or to do stupid stunts but to do the exact opposite; be careful, delegate some of the flying to the system when manual piloting AND camera work at the same time could be too challenging. (I am expecting for someone here to call me a bad pilot but I don't care.) Try to fly in those circumstances and get decent video result at the same time. I guess these kinds of shots call for a two person operation where a pilot can only focus on flying the airframe and someone else is doing the camera work (as with a DJI Inspire, i.e.). But we don't have that with the EVO II. So, my take for AR is to either be more conservative and tone the software down in making it fly in a more simple and safer fashion, or make it very clear that the OA's abilities are significanlty lowered in this mode.

 

[rvl]

I've did a small Dynamic Track test in a plain environment with blocked GPS.

 

[AstroNoot]

I've did a small Dynamic Track test in a plain environment with blocked GPS.

Thanks!
That confirms that the Evo 2 uses just the OA sensors in Dynamic Track , like Mike said.

Also, how did you manage to "tame" your Evo? It doesnt swing right and left like mine and Jiri's does. What firmware version are you on?

 

[rvl]

Also, how did you manage to "tame" your Evo? It doesnt swing right and left like mine and Jiri's does. What firmware version are you on?

Nothing special was done (except "foiled" GPS), but there is no complex/tall obstacles around.

Spoiler: The firmware is current stable (2.6.22)

Bootloader	ComponetName	Hardware	SerialNumber	Software
V0.0.0.0	DEV_DSP_RC	V0.0.0.0		V1.1.1.45
V1.0.3.0	DEV_RC	V0.0.0.4	R7A919481618	V2.0.4.7
V1.0.0.0	DEV_RC_PLAYER	V1.0.0.0		V3.0.13.0
V0.0.0.0	DEV_DSP	V0.0.0.0		V1.1.1.45
V0.0.0.0	DEV_CAMERA	V0.0.0.1		V0.2.31.37
V0.0.0.0	DEV_MOVIDIUS_1	V0.0.0.0		V0.2.31.37
V0.0.2.0	DEV_UAV	V0.0.0.5	HV5919526042	V0.0.4.41
V0.0.6.0	DEV_BATTERY	V4.0.0.0	B54820131980	V0.0.14.0
V1.3.0.10	DEV_GIMBAL	V0.0.1.1	G8K919522015	V0.1.49.0
V1.2.0.2	DEV_SONAR	V3.1.0.0		V1.2.1.25
V0.0.2.3	DEV_ESC1	V0.0.0.4		V1.0.3.6
V0.0.2.3	DEV_ESC2	V0.0.0.4		V1.0.3.6
V0.0.2.3	DEV_ESC3	V0.0.0.4		V1.0.3.6
V0.0.2.3	DEV_ESC4	V0.0.0.4		V1.0.3.6

 

[AstroNoot]

Nothing special was done (except "foiled" GPS), but there is no complex/tall obstacles around.

Spoiler: The firmware is current stable (2.6.22)

Bootloader	ComponetName	Hardware	SerialNumber	Software
V0.0.0.0	DEV_DSP_RC	V0.0.0.0		V1.1.1.45
V1.0.3.0	DEV_RC	V0.0.0.4	R7A919481618	V2.0.4.7
V1.0.0.0	DEV_RC_PLAYER	V1.0.0.0		V3.0.13.0
V0.0.0.0	DEV_DSP	V0.0.0.0		V1.1.1.45
V0.0.0.0	DEV_CAMERA	V0.0.0.1		V0.2.31.37
V0.0.0.0	DEV_MOVIDIUS_1	V0.0.0.0		V0.2.31.37
V0.0.2.0	DEV_UAV	V0.0.0.5	HV5919526042	V0.0.4.41
V0.0.6.0	DEV_BATTERY	V4.0.0.0	B54820131980	V0.0.14.0
V1.3.0.10	DEV_GIMBAL	V0.0.1.1	G8K919522015	V0.1.49.0
V1.2.0.2	DEV_SONAR	V3.1.0.0		V1.2.1.25
V0.0.2.3	DEV_ESC1	V0.0.0.4		V1.0.3.6
V0.0.2.3	DEV_ESC2	V0.0.0.4		V1.0.3.6
V0.0.2.3	DEV_ESC3	V0.0.0.4		V1.0.3.6
V0.0.2.3	DEV_ESC4	V0.0.0.4		V1.0.3.6

Then why do you have an in-app prompt to update?

 

[rvl]

Then why do you have an in-app prompt to update?

I think that was the app glitch.

 

[quinn karter]

I sent them the details of the crash (date and everything) and sent them the log too. So they can use any file they want.

Their audacity to tell me to buy my own props motivates me more to just send them the drone and make them repair it. In the end, i can't be 100% sure it's not a hardware fault. Imagine if there was a person instead of that tree

I wouldn't be using dynamic track around people anyway. If you do, when a person is close, walk in the opposite direction. I can't imaging the damage an Autel 2s props would cause when hitting a person. I don't use dynamic track unless i am in an open field, road or beach. If I use it in a wooded area that has a path cleared but still has tree limbs in the way, I will constantly watch the drone and move it away from the branches manually. Half the time it stops anyway. I certainly would ride a bike and have it follow me through a wooded path. The speed is too high. Most of the time if Obstacle avoidance is on, autels sensitivity levels are so high, that the drone will often stop in it's tracks. Then I have to turn the OA off and go around the object, then put it back on. I HAVE TO SAY, ALMOST ALL MY CRASHES OCCURED WHEN I DEPENDED ON THE DRONES AUTOMATED SYSTEMS. I rarely if ever use them now.