MM take a swim

[boblui]

Drone does not now what is underneath?
Just put your hand under it, u will see how it will move by itself !

I meant the drone only sees SOMETHING below. It does not know what it is in the sense that whether that something alone is good for positioning and height detection is uncertain. Therefore it combines the outputs of all the sensors including the GPS, barometer and IMU to come up with a more reliable estimation of it's position and height. That mechanism is called sensor fusion. If the output of one sensor deviates too much from those of others, it will determine that the output of that sensors is not trustworthy and ignore it.

The theory of the bottom sensors being "confused by the water" so the drone landed into the water simply doesn't make sense. One doesn't have to be a drone expert to see that. If the drone is misled by the water into believing that the surface beneath is too close, it will go up to avoid it. If the opposite is true and the drone thinks that the surface below going away, it will simply hold it's altitude. That's the behaviour we all have been seeing isn't it ?

How can any "confusion" to sensors causes the drone to land and announce it with a message prior to doing so ?

There are two sensors (and a camera) there, one of them is for positioning under 10m
Although, in manual does clearly state that drone is flying 10m+ ONLY on gps and above 10m ONLY on VPS, be sure that it uses both of them

The manual of MM only says VPS will be enabled when the height is less than 10 meters. I am sure it will continue to use other sensors such as GPS and IMU to judge it's position. You can do a simple experiment to verify this. Hover the drone above a big carpet and move the carpet to cheat the drone into believing that it's drifting. Will it "correct" it's position by following the carpet ?

...thats infrarend sensor, and i thing that is cousing problems in this situations..

An explanation on the mechanism behind will be more convincing.

 

[boblui]

.... the downward facing sensors ..... are prioritized over GPS

You can do the Mini-over-moving-carpet experiement I described in the post above and see the result for yourself. I have done it. As least for the Mini, VPS does not have priority over GPS, it doesn't even have priority over IMU. Do the experiement indoor and you will see that.

.....This approach only gives problems when the VPS gets confused signals such as with wildly varying reflectivity from water for the IR proximity sensor.....

This is the first time I hear that the VPS camera, which detects horizontal movements ( linear motion and yawing ) works together with the IR proximity sensor which detects distance from obstacles beneath. They detect totally different things, how can they work together and to what ends ?

 

[scro]

@boblui I think we're pretty much in agreement but just using different ways to describe the same thing.

You describe it better by calling it sensor fusion, whereby inputs from all the sensors are considered and a solution to the drone's position/movement is determined. This output from the fusion processor is usually better than any one input the fusion engine receives (GPS, VPS, IMU).

What I'd meant when I said VPS is prioritiesd at low altitude was that the VPS signals have a significant effect on the fusion processor output. GPS (if available) is still used as an input and can, with the IMU, outvote an erroneous VPS input (eg moving carpet experiment).

I have done the "moving carpet" experiment indoors, with no GPS. The drone followed the carpet, which seems to be contradictory to your results. ? Fast movement of the carpet was not followed so there is likely a threshold where the IMU input is favoured over the VPS. As IMUs are more prone to slow drift errors than VPS it makes sense that slow movement of the carpet is followed by the drone (ie IMU input is filtered out with a high-pass filter) and fast movement of the carpet is not followed (IMU signal favoured over VPS)

For me the term "VPS" includes both the camera and the IR sensors, so when I say the camera and IR sensor work together they each detect their respective things which when combined give a VPS solution for the drone's position/movement.

Getting back to the original thread topic, I can't think of any logical reason why an auto-land is initiated when over water. I can understand the drone not realising it is too close to the water as it can't see the surface properly, but what do the sensors detect over water that causes an auto-land? If the drone thinks it is too close to the ground the normal response is for it to gain altitude rather than land.

 

[boblui]

@boblui

I have done the "moving carpet" experiment indoors, with no GPS. The drone followed the carpet, which seems to be contradictory to your results. ?

The result I got with the Mini and M2P is that the drone would follow the carpet for a few inches and then stopped. My feeling is that it takes some time for the flight controller to realize that something fishy is going on. The carpet was moved slowly. It was also done indoor so no GPS. I haven't repeated it for enough number of times to come to a firm conclusion though.

 

[scro]

I have a Tello and a Mini. Both followed a slow moving carpet for a couple of meters, and neither followed a fast moving carpet indoors, with no GPS (I say that because I have had GPS indoors on some occasions when near a window!) I'd test it again to see if I can replicate it and get a better feel for when/if it will follow the carpet, but my Mini is currently parked in a tall tree?.

I've had the Tello follow leaves blowing along in the breeze, but then it doesn't have GPS as a reference to know any better.?

 

[kjmasson]

What are other instances where the mini is programmed to "auto-land" (for whatever reason - other than low battery). I know when I hit the land button and press the confirmation circle to confirm, it initiates landing, but I don't know of other scenarios (other than low battery) where the mini just decides on its own to land.

Can DJI not build a verification question into the app that always requires pilot response before auto-landing - much like when you purposely hit the "land" button to initiate a landing.

I would think if auto-landing is a result of confused sensors, a pilot verification action would easily resolve these water landings whereby it simply hovers in place until the pilot verifies.

 

[boblui]

What are other instances where the mini is programmed to "auto-land" (for whatever reason - other than low battery).

The drone has entered a no-fly zone. There has been such a report recently.

 

[Motorcycle Rider]

Left control stick full up should abort any landing attempt it may try to make, unless the battery is dead, or too far away to have signal between the drone and remote.
As a side note, I've flown my mini over water, dozens of times, down to as low as 5 or 6 feet, without so much as a slight hiccup.

 

[scro]

Left control stick full up should abort any landing attempt it may try to make, unless the battery is dead, or too far away to have signal between the drone and remote.
As a side note, I've flown my mini over water, dozens of times, down to as low as 5 or 6 feet, without so much as a slight hiccup.

I think it's lower than 5-6ft when the issues really start. I've flown to within 2-3 ft of water multiple times and not had any issues, though I kept a very close eye on the drone and was ready to intervene.

 

[Sallee]

The drone has entered a no-fly zone. There has been such a report recently.

nope! i live couple of meters from NO FLY zone (red zone) i unlocked blue (authorization) zone.. so when drone come for exemple 10m from red zone, it send u info that u are entering unauthorized zone,and resoult is that U CANNOT move drone in that direction,, like a wall, just wont move in that direction! u can go up down, back left right.. but forward (to the red zone) u can not! so no autoland,

The theory of the bottom sensors being "confused by the water" so the drone landed into the water simply doesn't make sense. One doesn't have to be a drone expert to see that. If the drone is misled by the water into believing that the surface beneath is too close it will go up to avoid it.


The manual of MM only says VPS will be enabled when the height is less than 10 meters. I am sure it will continue to use other sensors such as GPS and IMU to judge it's position. You can do a simple experiment to verify this.

Do u know how light is ''bent'' in water, object seems close in water? IR is light, and drone use it for positioning and also for proximity, if drone ''thinks'' that when he is above water 3m for example that he is 1m from ''some surface'' which IR falsely think because of light through water effect, when u push down stick it will go to land,

and dont think that VPS is not active above 30m? Drone does not know EXACTLY on what alt it is ( yes barometer can roughly estimate, but only roughly) VPS is ON all the time, but it is receiving info below 30m because sensors range is that.. like radar of 30m range which is always on and trying to scan surface..
ALSO one example, one guy in forum had problems with MM in fog in one flight, whenever he pushed stick down, drone activated landing procedure, he had big trouble to put drone to lower alt without activating landing procedure, then he just gave up and let it land.. DJI was asked about it, there answer was sensor confused by fog, so drone thinked it is just above ground so every stick down activate landing.

 

[Webcruise]

At least the Mavics should be smart enough to understand that it’s flying over water (at least ponds and sea) since it’s displayed on the map and not allow landing unless the pilot (i.e. on a boat) is there too.

 

[Sallee]

At least the Mavics should be smart enough to understand that it’s flying over water (at least ponds and sea) since it’s displayed on the map and not allow landing unless the pilot (i.e. on a boat) is there too.

so MM should read maps now? (For exemple like which drone? )
Fly up to 5km, have 30m flight time,2.7K video, under 250g .. and read maps, all that for 400$?
Should it also need to sing a song too?

 

[scro]

Nice idea to get it to read maps, but you need maps to be accurate to 10cm or so, and the drone to know its position on that map to the same accuracy for it to start to be useful. In reality the discrepancy between maps/satellite images and consumer GPS is several meters on a good day. Try switching on GPS on your phone and look on Google maps - see where that puts you compared to your actual location on the ground.

 

[scro]

@Sallee The light bending as it goes through water explanation only works if the IR sensor is detecting the riverbed/seafloor/bottom of pond and not the water surface. You'd need to be much lower than 50cm from the water surface for it to think it was within landing proximity of the seafloor. The IR beam would pick up some reflections from the water surface and this should cause the drone to rise up to about 50cm clear of the water surface again.

Also, when hovering 50cm above land autolanding is only initiated after holding the down stick fully down for around 1 second. Just a blip of down stick isn't enough to trigger auto landing. There must be something else going on to cause it to happen over water.

Does anyone know if the IR proximity sensor works on reflection intensity, or time-of-flight? ToF would seem like the better method to reliably measure distace over several meters with surfaces of varying reflectivity.

 

[scro]

so MM should read maps now? (For exemple like which drone? )
Fly up to 5km, have 30m flight time,2.7K video, under 250g .. and read maps, all that for 400$?
Should it also need to sing a song too?

Mine sings a wee song every time I start it up