Please don't take this comment as a personal attack. That said, you, and by extension maybe the people on your team, have no idea what you're doing (I have quite a bit of experience in scratch building drones along with an engineering degree). Not only did you grossly mislabel the different components on the flight controller, but the entire offline navigation flowchart powered by vision and comparing those to satellite imagery is complete nonsense. Additionally, the mechanical design stuff... have you ever even seen how an RC plane is built or put together one yourself? Anyways. Here are some concrete tips: - If you want to build a quality drone, especially one capable of offline navigation, you should not use dime-a-dozen IMU sensors directly on the main flight controller PCB. Especially the magnetic compass is sensitive to magnetic fields, like those induced in the circuitry around it. Just like the GPS, you want the magnetometer to be far away from other electrical components and metal generally. Additionally, for the gyro/accelerometer I would seriously consider putting them on a separate, vibration-isolated platform/pcb too. Just with this simple change, you can significantly decrease drift and cumulative error greatly increasing the measurement quality of even cheap sensors. - You don't need to invent a flight controller from the ground up for this, and in general, you are inventing too many components from the ground up. This is the death of many ambitious projects led by inexperienced people. You want to focus on getting a functional prototype working as soon as possible, that demonstrates the core feature (jamming resistance). To do this, you DONT need to scratch build a FC, a pixhawk will do just fine, and you DONT need to think up an airframe from zero either, just use a regular RC plane. Instead, focus your efforts on where the challenge is (this is basically a sensor and coding project). All the other stuff, like airframe, bomb bay, etc, are trivial once you've cracked the hard problems. Ardupilot or whatever flight control software is already open source, just modify the code (it's literally built for this). - You can scrap basically the entire visual pipeline system for offline/jammed navigation. It won't work, at least nowhere near to how you are describing it. There are countless issues with your approach. For example, you are using satellite imagery, which has an extremely long focal length and general optical characteristics. Any camera you mount on a plane that has a wide enough FOV to be useful will have significant perspective effects in the image, which means that the approach you've outlined is basically useless. Two images could be extremely close to eachother, yet because of for example a tall tree which will appear to point in different directions, have no matches. This isn't an issue in your scraped google earth images, because they are flat. Edge detection is also a very poor computer vision technique, useful basically only in extremely high-contrast scenarios where you have entire control over the lighting like in a factory automation setting. In the real world, especially in a "nature-facing" perspective, it will be basically impossible to use. Why? It detects edges in IMAGES, not in reality. This means that even non-real portions of the image, think shadows cast by trees, are detected as edges. As soon as the time-of-day changes or a cloud moves that changes the lighting, all the shadows will be different, completely ruining any attempt at trying to match edge-detection images and that's not even considering all the perspective issues. If you want to make this work with a camera only, I would look into SLAM or photogrammetry. While this won't match to a known value, you can use multiple images to very accurately calculate from where the picture was taken in 3d space, and then determine how far you've gone and traveled relative to a fixed reference point. Also, more generally, I strongly encourage you to not continue working on this project, or at least not publicly (releasing source code and such). While you and your team no doubt have the noblest of intentions, this doesn't mean that bad actors or terrorist groups won't repurpose your designs for malevolent and evil acts. Besides these abstract concerns, consider that might be real life consequences, ranging from for example being placed on a no-fly-list (happens faster than you think, would ruin your vacation plans for life) to actually being killed. This has actually happened--one of the notable founding contributors in the open-source firearms community where they were 3d printing guns--died under suspicious circumstances in Germany when special forces raided his home despite the fact that he spent a lot of effort staying anonymous and keeping his identity concealed. You, having a patreon, has already compromised your identity in the eyes of law enforcement and counterterrorism.

@NaeviusUAV

Ай бұрын

Hey! It is what it is. It sounds like you could give some good advice. Do you want to join the team? Application form is in the description

@gunlover92

Ай бұрын

While I agree with most of your technical analysis, no self respect ting terrorist is going to use this kids video as a blueprint for terrorist attacks they have access to the dark web or are taught by knowledgeable individuals from friendly nations

To remove noise from your 3D renders, you can add a "denoise" node in compositing with blender.

@NaeviusUAV

Ай бұрын

True but it often ends up looking so fake and flat and almost like clay. I prefer the noise cuz it looks as if it’s just a bad camera and not cgi

@bspringer

Ай бұрын

@@NaeviusUAV Then you can check out the dynamic seed. It will vary the seed and at least get rid of static noise (so now you've got dynamic noise, less noticeable and more realistic) Render Properties -> Advanced -> Seed -> the clock on the right

Since you're gonna be having multiple ways of determining position, velocity and acceleration might I suggest checking with your team regarding PID controllers and Kalman filter algorithms. Also if your drone has enough wing area, it might be worth considering adding solar cells. Apart from being a tiny range extender, they could serve as a sensor that determines the position of the sun during the day. Paired with knowledge of the time of day it might be another sensor for determining position and/or orientation of the drone.

@NaeviusUAV

Ай бұрын

Do you think the solar panels could really be accurate in practice? How do you determine if it’s a cloud instead, and is it not easier to just put a camera up there?

@atrumluminarium

Ай бұрын

@@NaeviusUAV Yes it's not perfect sadly. At that point I would imagine that a basic classifier model would need to be trained. As a starting point, an overcast is a very diffuse kind of light whereas direct sunlight would put shadows from parts of the UAV on some of the panels. The panels following the curvature of the aerofoils could determine whether the sun is in front or behind, etc. A bit of an outlandish idea but might be worth tinkering with if there are ever plans to embed PV panels in the wings.

@grekiki

Ай бұрын

​@@atrumluminariumYou can get the sun position with a clock and magnetometer already

mcu stands for "microcontroller unit"

@NaeviusUAV

Ай бұрын

My bad

@chrisalexthomas

Ай бұрын

@@NaeviusUAV learning everyday man, it’s the only way 👍 great job so far. I’m extremely curious about your dead reckoning navigation using maps to spatially locate yourself.

The "offline positioning system" is a really cool concept. I have done some remote sensing work. wondering if you can pair edge detection with feature detection (if not done already). Also loading a vectorized map before hand first from satellite but then once you map or survey an area keep that for future analysis. Really cool project and this particular system hits my specific interests. (GIS, Software Development, Computer vision, Etc)

@gabrielfinke7051

Ай бұрын

Also the SR-71 did something similar but with stars for its IMU

you may want to look into inertial navigation system. I think it'll easier than comparing to maps data which may be outdated/ there could be snow laying changing the overall apperance and so on

the same module that controls the servos could also control the ESC's I assume you're using PWM anyway for the servos so I think a module for everything PWM output related, would be best since it's super individual, the module could be cheap and small af too, it could communicate with the controller based on smth like UART or I2C but you probably are gona have a lot of ports for both otherwise the expansion would be pointless, but maybe you have smth else in mind?

cheap LIDAR-lite can be bought for a couple of hundred dollars and it works up to about 100m. TBH I think you should forget about rebuilding the wheel, a Pixhawk Cube already has temperature controlled IMU with multiple redundancy in the sensors. Then you can work on building a positioning system that can feed in to the flight controller. Tackle the 'little' job, rather than debugging the whole system. Good luck with the project.

I like how your project's going. I am working on a similar project and want to comment on the use of 2 component polyurethane to build the airframe. It's a terrible material to work with. its weak and brittle, It sticks to everything, deforms the mold when expands, and is not UV resistant. I would recommend to use EPP instead, or carbon/glass fiber if you wanna mess with composites. IMO working with carbon fiber would be less messy than working with the polyurethane foam. Good luck with your project!

@NaeviusUAV

Ай бұрын

Hey! Thanks for the advice. Can EPP also be molded?

@Argosh

Ай бұрын

@@NaeviusUAV theoretically yes, but no. The machinery required to do any kind of advanced foam production is prohibitively expensive. Even just ordering something would easily run you tens of thousands of dollars for just the mold.

keep going with this series its awesome

ALSO a flying wing with pusher propellers on the wing tips would be the best in terms of efficiency, payload, size, and Vtol too without needing to add the weight complications and price of more motors that would practically be a paperweight when not in landing or taking off

@pepegac9999

Ай бұрын

also the motors could be on the connected to the same servo as the control surfaces basically so you could accurately control the aircraft while landing or taking off with just 2 swiveling motors, it would by useful in flight too, making it much more responsive with much more control, you could maintain control probably even in a stall

also making it almost fully 3D printable would be A LOT BETTER, you can get filaments so light cheap and strong that the weight is on par with foam planes

@pepegac9999

Ай бұрын

also when ur talking about aluminum there is a filament for basically everything, u can even get 75% tungsten composite today, good lw pla with some asa and tpu would would be basically all you'd need maybe some composites

Another way to determine position is to receive broadcasted messages by any satellites out there and triangulating based on their published locations. having an ADS-B receiver that receivers nearby aircraft information could also lead to a nice approximate area discovery, all these steps will narrow things down a little bit by little bit. If you mix this in with Meshtastic devices, you could have a very robust platform capable to determining its position offline.

YOO on paper had the same idea for SOOOO LOOONG but never got the money or time to develop it though LOL, wish you the best of luck, I wouldn't be able to participate much but I can tell some ideas that would make it better along the way, if ur interested

Computer vision flight is overrated. I think radio environment navigation is the most powerful. Think about all the radio stations blasting 24/7. You don't even need to know where they are coming from, even though in a more complicated system signal source direction could help. But even just flying around and getting the signal strength of radio stations at all these points could create a map for you to navigate from. You could even fly with computer vision, just to collect the radio map data. With enough data you could fly offline in complete darkness.

@NaeviusUAV

Ай бұрын

The point is that it should work when it’s being jammed

@holski77

Ай бұрын

@@NaeviusUAV You can't jam all frequencies people are passively listening to at once. If you are passively listenign to certain signals, no one will be able to tell to which you are honing in on. Even if someone could jam all frequencies at once, depending on the typing of jamming, you may be able to use the jamming signal as another source to navigate off of.

Since the thing already has an accelerometer and gyroscope onboard, maybe add inertial navigation?

@SolarWebsite

Ай бұрын

Yeah, I've been wondering for a while: is such a thing available for hobbyists yet?

Yes

sorry, but doing visual odometry and aerial image feature matching is nothing new, and has been under development for decades, it sorta works (specially odometry, thats easier to do with out of the box solutions, but feature matching on a small drone would be SOTA.) but no offense, I doubt you could pull it off. your best bet would be getting a decent imu and dead reckoning. heck even radio navigation or stellar positioning would be more achievable

oh this is going on my save this is awesome

i sent you the mail

How fast is this supposed to go? Because above 200kph you can forget about EPS. Above 100 you will need a way to stiffen your hull for anything but the smallest of vehicles, be that some kind of internal ribs and strakes or a shell like fiberglass. This also heavily influences the possibilities for your power system. For extreme endurance you want to run ICE, most likely using nitro. A combustion motor will run you at least 300€ without being able to actually hover. Motors capable of providing that kind of power are 4 figures and above. If you want to go electric, you're even worse off. I've prototyped something like this a year ago, a cheap banggood 1800mm skyhunter epo kit, using a Hacker Motors A50-10S with a 6S battery pack and a 14x14 prop. The drivetrain alone runs you around 400€ and it delivers roughly 10kg of static thrust. But that will only fly for 10 minutes tops. Granted, it will do so at around 200kph, and scaling it back some might help, but you have spent 7 weeks and haven't even nailed down the basics.

@atrumluminarium

Ай бұрын

Tbh I doubt it will be designed to fly fast given the wing design that was shown and the purpose he described for it. Plus for sensor fusion purposes I think it's still beneficial to keep the GPS sensor on and then just shut it off when you know you've exceeded it's limit.

Why is flying further after having lost your GPS an option? Why put so much effort into something that should never ever happen. That one time it does happen you can just bail the drone?

@williamfosterXD

Ай бұрын

Defence against active jamming like what is happening with Ukraine and Russia