John Hawley Talks About UAV Controls (Video)

Video John Hawley Talks About UAV Controls (Video) 20

Posted by Roblimo on Friday April 17, 2015 @05:13PM from the monocopter-bicopter-tricopter-quadcopter dept.

Video no longer available.

John 'Warthog9' Hawley was the boss sysadmin on kernel.org before he jumped to Intel in April, 2014, as an open hardware technical evangelist. He last showed up on Slashdot in June, 2014, with his Dr. Who-inspired Robot K-9. Now he's talking about flight computers for quadcopters, specifically ones based on MinnowBoards. Last month (April 2015) he was speaking at the Embedded Linux Conference + Android Builders Summit. That's where he and Timothy Lord had this conversation about flight controllers for UAVs, which makes it a fitting sequel to yesterday's video, which was also about controlling drones with real-time Linux.

Tim: John, first, in talking about flight in-flight computing, you praise 3D Robotics control hardware.

John: Yes.

Tim: And one thing you say is that, let me get this right, you said that serial UART – Serial UART is the greatest thing in the world. Why is that?

John: So, it’s a – serial UARTs are absolutely fantastic. They’ve been around forever, everybody understands them and they’re a really easy way to get bidirectional communication going. You know, even on embedded platform servers, these are things that a serial UART can give you all the insight in – potentially has the ability to give you all the insight into a system. They’re in some cases even more useful than a monitor hooked up to the computer. And with a serial UART, it gives you the ability to have a very simple, very straightforward way of dealing with communication. And since the MinnowBoard Max has a couple of serial UARTs available on it and the Pixhawk has a serial UART that has a well-defined API to communicate with it, this gives me a really easy, simple, straightforward way of dealing with the communication between the two devices unlike something like CAN that can have this very binary protocol that you may or may not have to decode or guess that. A serial UART is just text and text is one of the easiest things to figure out.

Tim: Is battery compatible and

John: Yeah.

Tim: as much as it can be?

John: Yeah, I mean, serial UART’s have been around for practically since the dawn of modern computing, so

Tim: And with embedded stuff especially you don’t have a lot of other ways to what is going on?

John: Yeah, you know, a lot of embedded systems, they just don’t have a lot of other communications options. I mean, you know, USB is great, but sometimes you don’t have USB or you don’t have the power for USB.

Tim: Drones, they are really good example of the constraints that are faced by any embedded system, you got weight, power and in the case of drones, you are also adding aerodynamics, and at the same time, what you are at least right now, you are not advocating is that you move all of your flight control to sort of a new system.

John: Yeah.

Tim: Explain why is it worthwhile to actually strap on a Linux computer for some other functions even if you are not using that for your flight control?

John: So, yeah, one of the biggest reasons why you don’t want to reinvent the flight computer wheel is, the flight computer wheel is really hard. There is a lot going on, even with this particular drone, there is a lot going on with the flight control systems. You’ve got eight routers here, so you’ve got to manage eight different PWMs to keep the drone in flight, you’ve got to deal to with the accelerometers and gyroscopes. And at some point, depending on what you are trying to do, you may just overload your base system and keeping things aloft. So if you effectively treat a Linux computer as a side coprocessor, that means that your flight computer, in this case, it’s a Pixhawk, can do exactly what it does best, keep the thing aloft, go from Point A to Point B and deal with the things to keep the drone in flight. That means that I personally can spend more time working on the computer vision aspects and the things that I actually care about and with a really well defined protocol to communicate between the MinnowBoard and the Pixhawk, in this case, serial UART. It means that I’ve got a very simple straightforward way of dealing with the – dealing with the complexity that the Pixhawk is dealing with. I just say, go to this, add this new flight point and the Pixhawk handles that problem.

Tim: You can be as real-time as you want

John: Yeah, yeah.

Tim: ...but you are less concerned about crashing because of that.

John: Yeah. If something goes wrong on my computer vision processing or heaven forbid, Linux kernel panicked because I’ve written a bad driver for something, if that happens, the UAV crashes. If that is at least farmed off to a secondary system, in this case, the Pixhawk, it means that I have abdicated my responsibility to solve that problem.

Tim: You say one thing that people do when they get a drone like this is they say put a camera on it and they put up, take pictures, that’s sort of the universal human urge. But one product that you described today is actually – it goes quite beyond that, and is out of the University of Washington Vancouver in the [inaudible] system.

John: Yeah.

Tim: Can you talk about having mapping applications and talk about what savings you get in time of processing when you can actually offload the task that it was doing?

John: Yeah. So this particular project they purchased us almost a year ago now. And what they are doing is they take a quadcopter and then traditionally they hook a camera up to it, go out to the survey trees or these groves that they have got and they lock the quadcopter, fly around, take a lot of pictures and then go back and post-process these photos into 3D models, very accurate 3D models. And the reason they are doing this is to gather scientific data over time on how the tree is growing, which directions it’s going, does the tree shift based on light patterns, what’s going on, they are just collecting basic science. And the problem with the way they have – or they have been doing this is that they will go out, they will gather all this data, they may drive for six hours to get to these trees that they have marked and are keeping track of. Go out, take all this data, takes a little while to fly a quad around a tree, come back and then post-process this data for 20 hours to see if they have got a really accurate 3D model.

Tim: Can you explain why it’s quite so much post-processing?

John: So the post-processing is, is that they have got thousands and thousands of images and they have to figure out the correlation between each image independent of every other image. So it’s just a lot of data that needs to be chunked.

Tim: They are creating the image entirely by actually framing.

John: Yes. So I mean it’s basically like if you are using Hugin to create a panorama, well they are doing the same basic idea although they are creating a 3D model. So what they are finding is – or what they were interested in when they contacted us is they wanted to loft a MinnowBoard MAX on a similar platform to this, keep the camera going, but they wanted to see if they could get down to doing some of the basic processing in real time. So take the exact same data that’s coming into the camera and process it onboard. And the reason they wanted to process it onboard is that streaming large amounts of data from a UAV back down to a computer on the ground is actually sometimes very hard because your RF may be noisy, you may lose contact. I mean there is a lot of ifs when you start dealing with radio communications, particularly at some distances. But if you are processing it onboard, you can actually make some more intelligent decisions. So if you’re – if you know that you’ve got a bad frame, you can mark that position in your mapping and come back to it an hour later or something. So if you are getting glare off of the camera lens in a specific spot, you can come back to it. So what they found was by using the GPU and the CPUs on the MinnowBoard hooking it up to a preexisting fly computer, I think they are using the Pixhawk that they can do this, that they can actually do real-time processing of a basic 3D model while the quadcopter is in place. They are taking advantage of lot of extra information that they weren’t getting before out of the camera like the GPS data. And so, they’ve got some help on the relative location, but they are also still doing the same basic image processing. And by the time the quadcopter lands, they actually know whether they got a viable 3D model or not. So, just knowing, driving six hours and flying the quad around and then landing it knowing that they’ve got good data right then, is a huge win for them.

Tim: Sure. One thing I want to ask you about that came up in your talk as well is the power differential it takes between flying a drone like this, it’s a multi-blade [indiscernible] (8:31) you got versus a fixed-wing aircraft, power that’s controlled, you don’t have the glide effect when you’re using your blades as your wing. [It’s not to repower on a suspicion]. So, talk about the power difference that creates.

John: So, I mean, the biggest difference you’ve got is that whether a fixed-wing aircraft, you’ve got a glide factor, which means with a single motor, you’re basically just pulling yourself through the air and what’s providing your lift is actually the wings themselves. So, as long as you can keep moving forward, you still have lift. With a quadcopter or a helicopter, what’s providing your lift is actually the rotors themselves. So, as soon as the rotors stop for any reason, your lift stops, which means you’re coming down. And this means that you’re relying significantly more on the rotors and the power that’s going to the rotors for staying aloft, which means fundamentally something like this, this is a 3D robotics X8, weighs with the battery about five pounds to launch. It can stay in the air for about 15 minutes. Using the same batteries and a similar – a motor that has similar output to these motors, you can probably keep a fixed-wing aircraft in the air for an hour or two.

Tim: What’s the actual power drawn on these motors?

John: So, I don’t remember exactly off the top of my head, but I believe the batteries are capable of discharging at 55 amps at about 18 volts. So, it’s a pretty substantial output particularly when you’re – most computer scientists or people who are dealing with computers are used to like 5 and 12 volts and couple of amps and here, we’re talking something that’s 55 amps. I mean, case in point, you can kind of see the width of these wires for the battery, they’re fairly non-trivial, so.

Tim: It’s like it could almost go to a car battery.

John: They could almost go to a car battery. Now, I mean, the nice thing is that you’re not going to discharge that, that full-load all at once, but it can spike, so.

Tim: Now, this also brings me a little bit to the previous question about the power drawn versus the aerodynamics, you’re adding a board on, but it’s not really significant part of the power drawn from the batteries, compared to what these motors do.

John: Yeah. So, the board itself, the MinnowBoard uses about 2, 2.5-amps of power at 5 volts, and we’re talking about 11 or 12 volts of power. As I was just saying, the motors – the batteries for this can discharge at 55 amps at like 18 volts. I mean, the MinnowBoard, when you start talking about those kinds of numbers, is almost noise. So, I mean, admittedly

Tim: There is not much gain in under-clocking it.

John: Yeah, there is no gain in under-clocking it. Even the weight of an extra battery, just to power the MinnowBoard is negligible when you’re talking about, the UAV can only stay aloft for 15 minutes.

Tim: Right. If you’re on space, it’ll be a different story.

John: Yeah, yeah, yeah, if you’re in space, power is a completely different problem, but – or even a fixed-wing aircraft, I mean, if you’re not talking about a UAV that can stay aloft for an hour, the battery consumption and battery resources for the MinnowBoard or any platform become a much bigger deal than you need to power it for 15 minutes.

Tim: Yes, one more thing.

John: Yep.

Tim: What ties into it isthe problems you can encounter. Now I think you can probably get MinnowBoards at a good price.

John: Yep.

Tim: At the same time, you’ve got a $100 plus board that’s strapped to a multi-$1000 drone here, so what are some of the disasters you hit with the drone hardware or so?

John: So the drone hardware is something, it’s not for the weak of heart, particularly like you said that even this platform, you buy a ready to fly, it’s multi-thousand dollars, even in kit form, you’re still talking parts of $500 to $900 and then quite a bit of time to build it. So these are things that – in a lot of respects, they are a labor of love. So if you build one of these things, you have to be prepared that this either – expense of both either in time or money, you’re going to crash it, and when you crash it, things can go horribly, horribly wrong.

Tim: How do you lost complete air frames?

John: I have not lost complete air frames yet, thank goodness. But the nice thing about the entire UAV community is that there is a lot of like make or do-it-yourself just simple – let’s make generic parts and let’s glue them together in different ways. I mean there is a lot of parts even on this drone that are all off the shelf. I mean the motors are off the shelf, there is nothing weird about them, you can go and put these on any air frame, the propellers, again, straight up, they’re standard, they are just plastic rotators of the right dimension and the right pitch. And even – I mean most quadcopters are going to be like this. So even if you do get into a crash, it is – you can recover from it, it may cost you a bit, but it is doable, so

Tim: When you go from there then to the disasters you’ve had with the electronics you’ve added, what are some ways that you actually hurt the board that you put on there, what about the power draw and that computer people generally used to 5, 12, how does the power affect the longevity of your hardware?

John: So one of the easiest ways to fly a board is the power you are dealing with off of the battery, again, you are talking 18 volts at these many – tens of amps of power, and if you don’t have a good power converter because the MinnowBoard takes five volts, it’s possible for things to spike and flow through and you will smoke boards. So that’s always a problem. And you play with these things, you go – this works, this works, oh this smoked; this works, this worked, oh, this smoked. And these are just kinds of things that over time, you build up like this way, it works, this way it doesn’t. And like I said, if you are going to crash, you are going to lose parts this, that. And for the most part, I end up attaching the MinnowBoard relatively far out on the exterior which means that if something goes wrong, it is one of the first things in the line of potentially being damaged, so

Tim: How about collisions, there are no moving parts here on the MinnowBoard, but do collisions end up that intense?

John: I haven’t seen bent pins or too much along those lines. I mean the way I end up attaching the MinnowBoard is basically with industrial strength Velcro to the frame so I can call it off if I need. But – and I’ve not that slightly lose a few times, but for the most part, I haven’t seen too much damage from collisions based on that.

Tim: Do you have a garage full of parts?

John: Do you have a garage full of parts? So in fact the case that I have running around with me, I have about half a drone’s worth of parts in there. So [push comes], if I’m out and running around and I have not had a horrible crash, there is a chance that I could rebuild the drone on the fly, so – and that’s not uncommon for folks who are lofting these larger – and this would generally be considered medium-size quad. I mean some of the bigger ones are over a meter in diameter, but generally people carry around spare parts – buckets of spare parts so that

Tim: Would you call those people the drone community?

John: I would call them the drone community.

This discussion has been archived. No new comments can be posted.