Thread Tools
Feb 21, 2020, 09:59 PM
Registered User
Thread OP
Discussion

Thinking about developing sub 250g cinematographic drone


Hello RCGroups! When I first got into the hobby back in 2015, it was a lot of fun! I left for some time but periodically updated my drone as parts broke and needed to be replaced/upgraded. I started out with an Arduino with MultiWii, mounted to DJI FlameWheel frame with a Basecam gimbal. At the time this was the state of the art for DIY drones, although now we these tiny highly integrated FC's running betaflight/cleanflight/baseflight/INAV for all the small racing drones! Impressed but also a bit frustrated with the rapid developments and obsolescence forcing me to constantly upgrade the FC, I switched to the more mature Ardupilot platform running on a PixHawk clone with a Storm32 Gimbal with a Yi 4K+ actioncam.

Sadly, it seems as regulations around the world have gotten quite restrictive and I am "out of the loop" with the state of the hobby, both in terms of legality, but also in terms of good entry-to-mid range gear and parts for DIY builds. Currently I see a bit of a gap in the hobby for small lightweight cinematographic drones. The DJI Spark and Mavic are very impressive but are very much proprietary platforms. I want to see a more open platform that allows more experimentation and tinkering. Of course to make a very small and reliable drone platform, one has to give up things like modular design, but that's fine since I have the ability to design custom PCBs and electronics. Small is no problem for me lol.

My thought is to design an ultralight that can fly with 1 or 2 Sanyo 3500mAH ncr18650ga cells, can pair with whatever the most popular TX is right now (Ideally running some version of OpenTX) that also supports telemetry over the same link, and utilizes a small camera module, although this will either require extensive hardcore hacking (reverse engineering a typical smartphone CSI camera sensor) and use of an FPGA or a big SoC to interface it, or signing an NDA to get the technical data package from a company, which is very unlikely. If I can get this figured out, then I can build a miniature gimbal around the module and really get the size and weight down!

Questions for the community:
  • Is there much if any demand for such a product? No worth pursuing if there is no interest!
  • It sounds like 250g is the upper limit for most parts of the world before stupid crap like registering are required.
  • What TX do you guys recommend? My FlySky FS-T6 is an unreliable piece of junk. The i6 looks like a good budget option but I want something that is practically designed for OpenTX, is less flaky, and work on a standard that's isn't expected to go obsolete. (I want to integrate a receiver, gimbal, telemetry, and FC all on one PCB)
  • I think Lithium-Ion cylindrical cells are a very good choice, Unless there are higher energy-density flatpack Li-Ion (commonly but erroneously referred to as "LiPos.")
  • Are there any good open-standard digital FPV systems? I know DJI has a decent locked down system
  • I might also add a thermal imaging camera, I got a SEEK Thermal 320*240 sensor which is cheap but uses USB and needs to be reverse-engineered as well for good shots.
Last edited by Power Max; Feb 22, 2020 at 02:57 PM.
Sign up now
to remove ads between posts
Feb 22, 2020, 03:43 AM
Fan of just about anything RC
SoloProFan's Avatar
Sounds like an interesting concept. Digital FPV is still in development, DJI and the Caddx Vista have set the bar but indeed are closed systems, you need a DJI goggle to use it, which makes it expensive, and not everyone likes the fit of these goggles.

While some countries are trying to put restrictions on any quad with a camera, staying below 250g can still be benficial, as currently in some countries that class is still free of extra rules.

The best all round transmitter these days seems to be the Jumper series, which support multiple protocols. So you are free in choice of receiver.

Single PCB designs might help reduce weight, and you don't have complex wiring. The drawback though is that if one component fails, the board often is useless.
Latest blog entry: For the love of the hobby!
Feb 22, 2020, 03:32 PM
Registered User
Thread OP
Quote:
Originally Posted by SoloProFan
Sounds like an interesting concept. Digital FPV is still in development, DJI and the Caddx Vista have set the bar but indeed are closed systems, you need a DJI goggle to use it, which makes it expensive, and not everyone likes the fit of these goggles.

While some countries are trying to put restrictions on any quad with a camera, staying below 250g can still be beneficial, as currently in some countries that class is still free of extra rules.
So far as I understand the US is relatively free of restrictions sub-250, with registration required for anything over, with or without a camera. Airmap can be used to request permission to fly within restricted airspace with certain limitations (areas near airports are split up into zones and you can get automatic approval to fly below a certain altitude depending on the zone.) This system is a very good compromise to share airspace, but I am not looking forward to the requirement of having to carry another burdensome transponder! and sadly FPV googles seem to be unallowed entirely...

Quote:
Originally Posted by SoloProFan
The best all round transmitter these days seems to be the Jumper series, which support multiple protocols. So you are free in choice of receiver.
So since I am running ArduPilot, would having a "wireless UART" connection to it allow full control and telemetry over a single link (MAVlink)? AFAIK many of the TX radios allow basic telemetry of some things but I'd imagine having full control over setting up waypoints and using something like qgroundcontrol or mission planner would require a laptop or smartphone with the app and a bluetooth or 802.11 wlan link, or a USB tether from the phone with an OTG cable, which I had limited success with.

A very important feature for me would be to have some way of controlling the camera gimbal in flight, mainly pitch. Especially if external foot petal or something could be hooked up so I don't need to take my thumbs of the sticks.

Quote:
Originally Posted by SoloProFan
Single PCB designs might help reduce weight, and you don't have complex wiring. The drawback though is that if one component fails, the board often is useless.
Unfortunately that is one of the downsides of highly integrated solutions. With my extreme cheapness, I have gotten very good at repairing PCBs Louis Rossmann style (with a hot air rework station, stereoscopic microscope, advanced microsoldering skills, etc.) and had to repair my GPS, Flight FC'es, and gimbal controller and the seperate IMU board a number of times, mostly due to reverse polarity or the board coming into contact with 12V from the battery one way or the other or by very bad crashes ripping components, connectors, and inductors clean off the PCB.

This can at least partially mitigated by using lightweight components to reduce the chance of parts getting sheered off in a crash (inductors!) and either encapsulating the PCB in a conformal coating or a small case like the pixhawk with specific connectors (Molex PicoBlade)
Feb 22, 2020, 04:02 PM
Fan of just about anything RC
SoloProFan's Avatar
I have no experience with Ardupilot, only CleanFlight, BetaFlight and a tiny bit of OpenPilot. So can't comment on that part.

An highly integrated PCB can work well, if either treated with care, or with some headroom built in, like ESC being rated high enough to handle more power than the installed motors are likely going to draw, etc. But you probably already were aware of that.
Latest blog entry: For the love of the hobby!
Feb 22, 2020, 04:32 PM
Registered User
Thread OP
Quote:
Originally Posted by SoloProFan
I have no experience with Ardupilot, only CleanFlight, BetaFlight and a tiny bit of OpenPilot. So can't comment on that part.

An highly integrated PCB can work well, if either treated with care, or with some headroom built in, like ESC being rated high enough to handle more power than the installed motors are likely going to draw, etc. But you probably already were aware of that.
You raise a good point. Integrating the ESCs will be difficult. Manufactures have a tenancy to overrate them, as the current ratings (both for batteries and for ESCs) tends to be peak maximum specs and not continuous. Still, even 10A is a lot.

When it comes to PCB production, one of the steps is to specify the thickness of the copper traces. 1oz copper (meaning 1 ounce of copper per square inch) is typical, with 0.5oz for inner traces (boards that are 4, 6, etc. layers) however I've seen some higher end ESCs use 2 or 3 oz copper enabling better current handling and lower ohmic losses. Layout is also very important to reduce parasitic inductance and increase robustness, but these are design details. It's also not ideal to have sensors or sensitive digital busses (SPI, I2C) near any high power stuff, as voltage and current transients can cause radiated and conducted interference. (differential signals like CANbus are much more immune to this.)

2 oz or higher boards also limit the minimum thickness of traces and the minimum clearance between traces and a ground/power plane. So a single board solution with both very high current parts requiring thick copper, and small signal stuff with very fine pitch BGA or QFP parts may be a challenge. 5 mil clearance is the limit with most PCB manufactures for 1oz copper, and the QFN packages used for IMUs already push that limit. I suppose you could have a double-sided layout with 2oz on the bottom layer and 1oz on the top with 0.5oz 3v3 power planes in the center, but this is going to be an expensive solution requiring more advanced 4 layer PCB and extra manufacturing steps for a double-sided board.

The TL;DR is that these will probably need to remain separate with a board-to-board interconnect or a connector and short jumper. Thankfully it seems that single board ESCs already exist commonly, albeit a bit expensive still compared to 4 or so discrete units.
Last edited by Power Max; Feb 22, 2020 at 04:44 PM.
Feb 23, 2020, 12:36 AM
Registered User
Its difficult to outdo DJI, since they spend $$$.
Average hobby racing drone is 10% tape, 10% ziptie, 10% extra wiring, and 10% super cheap electronics.

Without spending $$$ your solution will be kinda rough.

Camera:
You could design/integrate an IMX, like how smarphone do.

Zoom/Gimbal:
Weigh a lot. Can pull out of existing drone.

Drone Size:
Carbon Fiber frame can be 15g for 4", 30g for 5".
7" possible for a sub250g. Tubular even lighter.

Electronics:
Sinusoidal & field oriented control commutation.
For image processing and AI need at least 4core Cortex like SD820 (cheap).

Basically you could build drone around smartphone pcb which already has camera.
Feb 23, 2020, 02:24 AM
Fan of just about anything RC
SoloProFan's Avatar
Quote:
Originally Posted by romangpro

Basically you could build drone around smartphone pcb which already has camera.

That's an interesting idea, smartphones usually also have an accelerometer, to read the phone's orientation and rotate the screen if the phone is tilted. Maybe that can be used for the flight controller? Since we're talking lots of DIY here, maybe it's feasible.
Latest blog entry: For the love of the hobby!
Feb 23, 2020, 12:59 PM
Registered User
Thread OP
Quote:
Originally Posted by SoloProFan
That's an interesting idea, smartphones usually also have an accelerometer, to read the phone's orientation and rotate the screen if the phone is tilted. Maybe that can be used for the flight controller? Since we're talking lots of DIY here, maybe it's feasible.
It is a brilliant idea! I've actually considered it seriously in the past, but after a bit of research these are the reasons no one has done it:

The SoC in a mobile phone is more like a computer than a MCU. The speculative, out-of-order nature of it means that while code executed is deterministic, the time it takes to execute isn't. Modern SoCs make decisions on-the-fly about what variables to cache, which to keep in RAM or even storage, etc. Worse yet, if you are running a "normal" OS like Android or Linux, then the CPU scheduler may prioritize other tasks and cause massive unexpected delays. For a PID control system, is a fatal issue that will result in an unreliable performance. You have to sample the gyro and acc at a consistent sampling rate and continue to use those values process that data to compute the new throttle for the motors at periodic time intervals. Failing to do so will result in an unreliable control loop. MCU's, by contrast, are simple and deterministic in how long they take to execute code. Developers can literally look at the disassembly and tell you exactly how many nanoseconds a subroutine will take to execute! They generally will either write "bare metal" applications (like the case with MultiWii/BaseFlight/CleanFlight/BetaFlight, etc which make heavy use of very low level hardware peripherals directly) or write an application/threads in an RTOS which gives freedom in making a custom CPU scheduler to ensure all your important time-sensitive threads and processes execute by a certain deadline while still giving some time for code that is less sensitive to timing, like a user interface. An RTOS also abstracts the hardware a bit making your application more portable.

The other issue is simply difficulty. As per above, we would need to port something like ArduPilot to the SoC by creating a custom RTOS for it and probably have to reverse engineer the SoC to figure out how to use the peripherals, interrupts, DMA, etc. Then figure out how to interface the MIPI/CSI camera module, as that is almost certainly going to be controlled by some sort of co-processor, which would be undocumented! Lots of "fun" for the god level hacker , I certainly don't have the know-how to actually do all this!

TL;DR is that it could by made to work, Android will not work beacuse it chooses when to execute this hypothedical PhoneCopter app lol. Even with god-level programming skills, Jitter in the control loop may result in unstable and poor performance. You can generally see this in practice by looking at the attempts people have made for Raspberry-pi Quadcopters (which always end up using a seperate MCU for the PID) and other real-time tasks like CNC machines, 3D printers, etc. Most will use MCUs for simple, fast control loops and a computer for higher-level control authority.



The last issue is a lot simpler. There just isn't any available I/O on a modern phone! I've taken too many of them apart to look at the insides. It *might* be possible to use something like a single board computer (SBC), which is in-effect the same thing as a phone motherboard, except many SBC's break out all the GPIO's for other peripherals. We would need GPIO's that have timers associated with them to creating the PWM / OneShot signals that drive the ESCs, and some for inputs from an Receiver, (although I suppose you could probably use the other wireless radios to establish a UDP network socket.... Again, WAY outside of my capability lol.
Last edited by Power Max; Feb 23, 2020 at 01:15 PM.
Feb 23, 2020, 02:25 PM
Registered User
Thread OP
Quote:
Originally Posted by romangpro
Camera:
You could design/integrate an IMX, like how smartphone do.
I plan to use an FPGA (Arty Z7 Zynq7000) for my prototypes to interface a MIPI CSI camera module like the ones found in smartphones. This is above my level of skill at the moment so I will need to learn as much as I can about those modules and see if anyone else has attempted this. This is by far the hardest part of this project!

Quote:
Originally Posted by romangpro
Zoom/Gimbal:
Weigh a lot. Can pull out of existing drone.
Hopefully by making a small quad, I can take advantage of the square-cube law. This fundamental law is what allows microquads to survive falling from great heights and surviving! Giant drones get destroyed even if they were built exactly the same because the mass increases by the cube while the surfaces supporting the structure only grow by the square.

Quote:
Originally Posted by romangpro
Drone Size:
Carbon Fiber frame can be 15g for 4", 30g for 5".
7" possible for a sub250g. Tubular even lighter.
Composite materials are amazing! Tubes work well because with any material, most of the structural rigidity comes from the outer surface. With monolithic materials like aluminum, you have a lot of material internal that does not actually do much in the way of strength. I like the Tech Ingredients video on the topic, and my drone body may be either carbon fiber tubes, or what is demonstrated here.

Quote:
Originally Posted by romangpro
Electronics:
Sinusoidal & field oriented control commutation.
For image processing and AI need at least 4core Cortex like SD820 (cheap).
Oh boy, this is a tough topic and I have been researching it for about 2 years now.

FOC reduces electrical noise, improves low speed performance, etc. The problem with FOC is that it's very complex, requires measuring the current into 2 of the 3 phases and requires knowing the position of the rotor. I don't quite know how the cheap ESCs manage to do all this. I think they just measure voltage and integrate it to get a rough estimate of current. I do know they use back EMF from the unpowered phase to determine the approximate position of the rotor and extrapolate assuming constant speed. This heuristic works remarkably well for high speed / low torque, especially since you have a large flywheel (the propeller!) helping buffer out torque/acceleration rippleand making dead-reckoning estimations of the position of the rotor more accurate.

For a Gimbal, I need to have high bandwidth / low speed operation at and around 0Hz. The "ESC" approach will not work, and I demonstrate the issues with low speed operation here. The algorithm used by those ESCs featured is particularly cumbersome 6-step commutation type.

This means that you can't rely on back EMF as a form of feedback for gimbal motors. I need some way to measure the position of the BLDC rotor. The best way is with a hall-effect sensor or with a pot like the Yuneec 4K gimbal. Both these solutions increase the complexity and cost. The Storm32 and other similar gimbals instead opt for a 3 axis IMU mounted to the camera, and you can compute the position of the 3 motors with euler angles. The math for this 3D stuff gets quite complex but it again works remarkably well! The biggest problem is drift due to there being no way to determine absolute yaw position when stable. An additional Mag can help alleviate this problem but I have not seen this approach used.

Besides not wanting to measure 6 AC currents, another issue with FOC on a gimbal is you have to deal with nasty nonlinearity's like magnetic cogging. I think most gimbal controllers work by simply forcing a fixed 3 phase output with some amplitude, overpowering the effects of cogging torque 10 fold which allows the rotor to snap to the position that reduces the angle between the electrically-produced and permanently-excited magnetic field of the rotor. Increasing the amplitude of the 3 phase current sets the "stiffness" of the rotor. Any small offset in either direction will cause a stronger tendency to correct back to 0 degrees unless you go far enough past 90 degrees and the rotor snaps to the next closest position that achieves 0 degree skew. So the idea is to treat the BLDCs like a stepper motor and use microstepping to have continuous control over the angle. While this is very inefficient and causes the motors to dissipate a lot of waste heat, it's very simple and the most common approach even in professional units.

Quote:
Originally Posted by romangpro
Basically you could build drone around smartphone pcb which already has camera.
I wish lol. There are a LOT of issues with trying to use a phone, highlighted in my previous reply.
Last edited by Power Max; Feb 23, 2020 at 08:49 PM.
Feb 25, 2020, 09:04 PM
Registered User
LemonC's Avatar
TLDR... but I caught the gist...


GOoD news, a bunch of people have beat you all to the chase and there’s plenty of documentation for you to catch up.

Look into the Dison mini Cinewhoop group on Faceboot I mean book... why ban that wbsite title..? The political section should be banned if the mere mention of fb is imo...

But yeah find the dison mini cinewhoop group. Basically decased gopros on 85mm micros and variations of that variety.
Feb 25, 2020, 10:33 PM
Registered User
Thread OP
Seriously? the word Facebооk is banned? Fuсkіng hell. That's some seriously retarded shіt. It isn't hard to bypass restrictions in place like that anyway, and I have just demonstrated. (don't tell the mods lol!)

As much as I hate Facebооk and their total disregard and disrespect for user privacy, blocking the name of the site is even more stupid then drone registration and remote ID to be banning the word... what a world we live in today full of butthurt and sensitive snowflakes. ��‍♂️��‍♂️��‍♂️

I'll check that FАСЕBООK site out. Thanks!
Last edited by Power Max; Feb 25, 2020 at 11:55 PM.


Quick Reply
Message:

Thread Tools

Similar Threads
Category Thread Thread Starter Forum Replies Last Post
Discussion Thinking about developing sub 250g cinematographic drone Power Max Commercial UAV/FPV 8 Mar 10, 2020 04:01 PM
Discussion biggest sub 250g drone challenge wrightbrothers Multirotor Drone Talk 4 Feb 25, 2020 05:54 AM
Help! Uk law, what classes as a 'drone' and sub 250g laws tomtom001 Model Aircraft & Drone Advocacy 51 May 02, 2019 10:21 AM