Algorithms for quadcopters to use back-EMF to detect obstacles close?

[AmericaPacific42]

TL;DR Summary

Please post URLs to articles about how to make flight software to process the back-EMF data (from each separate electric motor) to detect if obstacles are close (as the air pressure would go up if the motor is closer to a wall compared to empty air.)
Algorithms to detect obstacles with the forward optical camera would also help; so far they all require expensive LIDAR!

Would it help to study Verilog (VHDL) or Field-Programmable-Gate-Arrays (FPGA) if self-crafted x64/aarch64-assembly would take too much power/be too slow?

 

[anorlunda]

I presume that you mean drone quadcopters, not passenger ones. It reminds me of an old Farside cartoon.

 

[Borek]

TL;DR Summary: Please post URLs to articles

You do know that the forum rules are "do your research first, come with questions later"?

 

[AmericaPacific42]

I presume that you mean drone quadcopters, not passenger ones. It reminds me of an old Farside cartoon.

View attachment 318738

I asked for help to create autopilot software for aircraft, whether they carry people or are just cameras. The latter is all I am able to afford myself, but I would be happy to help write the software for both.

 

[AmericaPacific42]

You do know that the forum rules are "do your research first, come with questions later"?

Yes, did my research. All I got from searches were results about how to detect if a crash already occured from back-EMF, or how to detect obstacles with extremely limited laser sensors that contribute nothing to image capture.
I take it back for other reasons though; the whole quadcopter design is way too noisy. Birds fly just as fast, completely quiet, with better agility to dodge obstacles possible compared to the quadcopter design. Plus they fly for miles.
I doubt that I would ever be allowed to give the quadcopters good software, but if I was allowed, the physical design would still be fundamentally flawed beyond hope. But all the people paid to program them are so retarded that they can't even make good software, which just makes me sick.

 

[AmericaPacific42]

Sadly it seems nobody has made a camera with bird-like flight.
Should I just buy some servo motors for muscles, metal ribs for structure, tarp to cover the ribs for lift, a camera, radio setup, plus microprocessors, to make a cyber bird? The biggest work seems to be to write the firmware/software. Will somebody help me create this? I am very suprised that it was not created yet! It seems like it be fairly easy for a roboticist from MIT or Harvard to have created a fairly good copy of a bird that flies.

 

[berkeman]

Sadly it seems nobody has made a camera with bird-like flight.

I'm pretty sure that RC birds have been built. Where were you looking?

 

[AmericaPacific42]

The closest result is this ( https://www.festo.com/us/en/e/about...ghts-from-2015-to-2017/bionicswift-id_326830/ ) which lacks the ability to avoid obstacles because it requires a pre-programmed flight path that assumes that all objects are always where they were at the time the flight path was uploaded. Plus, the hardware itself looks fake, with no parts list available, and no way to buy or pre-order.

 

[berkeman]

A quick Google Images search on RC Birds seems to turn up lots of hits...

 

[AmericaPacific42]

I swear I searched before I wrote that, I'm clueless at how I missed this, but I appreciate it!

 

[Borek]

Birds use just sight. DJI drones have no LIDAR yet as far as I know they do have some obstacle detection capabilities based on the image processing (never risked tested it though, even if I am flying one).

EMF from motors doesn't sounds viable to me. You have a motor emitting EMF noise, an EMF detector near the motor (so bathed in the noise) and you want the detector to detect signal reflected from an obstacle that is quite likely just something small and organic (say branch/leaves). Reflected signal is pretty low, noise is pretty high - I can be wrong, but for me these things don't add up reliably.

 

[AmericaPacific42]

Birds use just sight.

This is what I wished to add support for to the firmware (so they would be able to use the forward optical camera to detect/avoid obstacles, plus store the area of the obstacles+their velocity, so it could guess where they are after it flies past them.)

DJI drones have no LIDAR yet as far as I know they do have some obstacle detection capabilities based on the image processing (never risked tested it though, even if I am flying one).

The best available is DJI's obstacle avoiders that use bulky lasers (not LIDAR, my bad,) to detect how close it is in that exact direction to an obstacle. They have the scan width of pencils and there is no memory, so once objects leave the extremely narrow field of view of those laser sensors, the aircraft is unable to avoid the obstacles. Also, only the very large and expensive drones have sensors that point up/down/left/right/forward/backwards. They also have no memory and no way to estimate where objects will be from velocity, because they just say what is the range to the closest object in that direction, with the laser range-finders.

EMF from motors doesn't sounds viable to me. You have a motor emitting EMF noise, an EMF detector near the motor (so bathed in the noise) and you want the detector to detect signal reflected from an obstacle that is quite likely just something small and organic (say branch/leaves). Reflected signal is pretty low, noise is pretty high - I can be wrong, but for me these things don't add up reliably.

I hoped to write assembly (x86-64 or aarch64,) FPGA or VHDL code to process the electrical load of the motors, because it will vary if the air pressure goes up. The air pressure will go up if that particular motor gets close to a wall. This will make it take more power from the battery, which is the data that I hoped could be processed by a fast algorithm to avoid obstacles.
Especially if the data from all of the motors is used together, it should be possible to make up for gaps the data would otherwise have (eg gyroscopes or other tools that tell the craft which way is up relative to itself, could help it had problems due to breezes/gusts that got mistook for walls. It could also use a compass to help tell which cases the load varied because of obstacles versus which cases varied due to the atmosphere.)

Where local authorities allow aircraft with mics, you could also add a mic to help it detect the waveforms that are reflected from nearby obstacles, with a circuit that is able to filter out the waveforms of the loud motors themselves, to copy how mammals (eg bats or whales) echolocate. It should be easy to make the firmware automatically take whatever data sources are available to give the best possible estimate of perfect behaviour (eg the software, if we make it, should work for all aircraft, whether they have all of these data sources or just some of them.) I will help make it for free if somebody is okay with remote work.

 

[AmericaPacific42]

Just simulate how fish[a] or birds[ b] are able to use the air/water pressure to avoid obstacles or fly/swim close together without a crash. It should be possible to detect the pressure from how the motor load varies, because higher pressure air takes more power to displace, right?
This probably works best with more motors, such as a hexacopter, or if other data sources are used together with it, because obviously birds or fish with just 4 cells spread over their body to detect pressure would probably crash a lot.
Could RC birds have small pressure detectors attached to the base of each feather to collect data about the atmosphere (to process to detect obstacles)? They seem to work with a motor (or 2 max,) which seems very difficult to get all of the data required from, so I proposed small, solely passive, motors attached to each feather just to detect pressure.
Please tell me which algorithms already exist that most closely fit this problem, or where I should look, or who I could help make this if somebody already hopes to create such.
a) https://www.sciencedirect.com/science/article/abs/pii/S0022519316302004
https://www.scientificamerican.com/...s-lateral-lines-sense-pressure-and-predators/
b) https://www.google.com/search?q=how+birds+avoid+obstacles+from+the+side+or+below
https://www.futurity.org/wingspan-birds-flight-drones-772212/

 

[AmericaPacific42]

What about just some help with algorithms to make use of the forward optical cameras that all aircraft (plus most smartcars) already have? There would still be some blind spots, but it would be a lot better compared to how it all works today.
Could we at least get them to detect the velocity vectors so that they are able to predict where obstacles (people, other cars, dogs, etc.) will move to?

 

[pbuk]

I would be happy to help write the software for both.

Then you need to look at what is required to be recruited by one of the many compaies each with teams of hundreds of people doing just that.

https://en.wikipedia.org/wiki/Avionics_software

What stage of education are you at, what courses are you taking and what is your nationality?

 

[AmericaPacific42]

Then you need to look at what is required to be recruited by one of the many compaies each with teams of hundreds of people doing just that.

https://en.wikipedia.org/wiki/Avionics_software

What stage of education are you at, what courses are you taking and what is your nationality?

Certification is my only blocking point. I was always self-taught (due to asperger's disease/autism,) so I lack a degree. All of the jobs always required a degree, or lots of referrals from previous jobs. I have years of practice as a writer of (X)HTML, CSS, JS, TCL, PHP, MySQL, BASH, C, C++, Java, Swift, plus I created (from scratch) implementations of every algorithm I read about.

 

[Vanadium 50]

Before going down the tough path of doing this with a drone, have you considered the easier problem of doing this with a Roomba?