fpvracing.tv is an online community for the rapidly growing sport of drone racing. I'm looking for somebody to help maintain the site, which is a rails app hosted on Heroku. Email address is admin at the domain.
Helicopters, like all flying machines, have much stricter safety regulations while also being extremely low volume. This means that for every part that goes into a production aircraft R&D, tooling setup, and testing will far outweigh the cost of materials, machine time, and assembly. This is very different from an a car where the cost of goods sold (materials, manufacturing, and assembly) is many times more than the sunk costs.
For example, if I wanted to make 30 complex six layer PCBs it can cost on the order of $9,000 at $300 per board. Realistically, $8,000 of that is the cost of labor for setting up all of the machines for the production run but if I wanted to make 1,000 of them the price can drop to $30 or less per board because the static setup cost is now spread across many more units. In aviation, every part is essentially custom and not made very often so you have no choice but to pay that huge overhead every time someone orders a chopper unless you a) batch together manufacturing (large inventory cost and risk) or b) maintain the machinery so that it's always set up to make your parts (large capital equipment underutilization cost). Either way you've got overhead that costs much more than the parts and is unavoidable. If you're lucky you can buy the machining equipment and save lots of money by renting out time like Boeing does with their multimillion dollar five axis machining centers which make very precise turbine blades for a variety of applications like power plants and dams. This rarely makes sense for a business to do however, because then they've got two businesses to worry about.
People are so used to cheap mass manufactured goods that many don't realize just how much more it costs to make anything custom.
Seems like there is an opportunity here? Presumably milling machines and so on will start to reduce this overhead, so that the overhead becomes just the cost of creating the original design?
Come to think of it, there might be some opportunity in standardization. I'm not an aviation engineer so what I say may make no sense, but a good enough set of "bricks" that would let you assemble anything from small twin-seater to a passenger plane would benefit from economies of scale, even if initially such a design would be more expensive and less performant than a custom build.
I guess since it isn't already happening, I'm probably missing something obvious.
Weight and balance is critical for airplanes. You can design the bricks, but you can't stick them together in any combination because a brick that will hold the tail wings for a 400 passenger plane is too heavy for a 100 passenger plane: the plane will not balance right and always be in nose up stall.
Also, there are not enough airplanes made. Even if the bricks idea can work, we still wouldn't make enough to give it significant enough economics of scale.
Helicopters are extremely weight limited. In order to allow for any sort of useful payload, all of the airframe parts have to be designed for a specific model (or range of very similar models). The shapes and mechanical properties needed aren't well suited to highly automated mass production.
Avionics and engines are somewhat standardized "bricks", though.
What are you working on in Shenzhen? And how would I go about finding a company to produce a custom shape/size lithium polymer battery for a prototype?
> My point is that if you make a successful product, you're going to get cloned anyway.
If you're going to get cloned either way, how will you make selling your hardware a sustainable business?
I love hardware (robotics, incidentally) but I'm so hesitant to release a hardware product because of the inevitability and futility of competing against cheaper clones. I've seen it happen again and again in the nascent world of drone racing - somebody comes up with an innovative new design and there's a clone on Banggood three weeks later. Then most people just buy the clone. I don't see how it can be sustainable.
In the short to medium-term, it's a hybrid product+consulting business model with consulting gigs to add features, custom sizes, or integrated with specific testing frameworks. Long term, competition is based on brand, build quality, distribution, ISV partnerships, and the ability to keep innovating. Also, it's a B2B product, so success or failure is predominantly driven by the quality of the sales team and their process as much as anything else.
The last 10% always looks really tough for me. For example we kind'of have self-driving cars right now - but when will we have self-driving cars without a driver's cockpit where I can play poker in the back with complete faith that it will handle all situations intelligently, navigate any terrain and manage any situation - seems like 50+ years to me for just that.
If you want your cars to literally never make a mistake you might have to wait a little longer, but reaching human or superhuman driving ability shouldn't be that hard. Humans are pretty badly equipped for driving, with our shitty reflexes, two eyes and easily distracted brain.
No, it can't. They still need to be able to understand that a toddler has stepped into the street, that a pedestrian has fallen down drunk in the crossway, that dude X thinks he ought to be able to go first or simply doesn't give a shit what the car thinks. You can't solve this problem by blinking away the world.
All of the things you mentioned can (and most have) been solved even in the current bunch of self-driving cars.
A toddle stepping into the street? A modern self-driving car sensor can "see" it better than the average driver -- and it doesn't even have to involve actual vision.
I'm not sure what you're arguing against exactly. Actual current research problems with self-driving-car vision are totally different (e.g. disambiguating signals when raining or snowing etc). And nothing insurmountable about these either.
>that dude X thinks he ought to be able to go first or simply doesn't give a shit what the car thinks
Whatever he thinks, if it translates to an actual action in his car, other automatic cars can respond to it in the blink of an eye.
Yes, if you've programmed the scenario in. We might solve specific problems, sure. But how large is the problem space? How much of it have we mapped, and how well does the machine cover it? If there is a long tail of obscure problems that the machine cannot handle because it was not taught how to, machines will remain shitty drivers because they can't respond with creative judgment to a new situation.
Explicitly programming scenarios in is not how it really works.
>How much of it have we mapped, and how well does the machine cover it?
Enough for Google cars for example to have logged over 300K miles in actual conditions with no incidents.
Also enough to have such things (besides Google's) in pilot operation in several cities the world other, for things ranging from cargo transport to mass transportation (self-driving busses).
It's not like "programming the scenario in" is some huge switch/case statement that needs to cover all possible arrangements of things on the street -- it's machine learning algorithms with several rules and invariants to check and various corrective responses when those are off, and the smartness comes from the combinations of such rules.
>machines will remain shitty drivers because they can't respond with creative judgment to a new situation.
The think is, with the appropriate machine learning algorithms they can both add experience and respond with creative judgement to new situations -- they don't have to have hardcoded responses to them from the start.
Note what it says: "“Our test driver, who had been watching the bus in the mirror, also expected the bus to slow or stop".
So he made the same assumption as the car did. So what else could it do, magically guess?
Self-driving cars are not infallible -- they model the same assumptions we do. But a benefit is that they can also react much faster than we do -- which won't always save the day of course: if some idiot driving against the traffic, for example, suddenly appears in front of such a car, they will still be an accident in all probability.
This thing drove into a bus, and, also, didn't learn and needed to be reprogrammed so it would drive into buses again. You are being a credulous fanatic, and I am done.
Machine vision is the same (or at least was before all this convnet wizardry that's going on these days). You could get something pretty good (say 80% accurate) in an afternoon, or something good enough to demo (say 90% accurate) in a week. Then it'll take you 6 months to get to 99% accurate, and 5 years to get to 99.9% accurate. And if you're analyzing one image a second, that's still more than three failures an hour, which is nowhere near useful for any kind of production system.
> but when will we have self-driving cars without a driver's cockpit where I can play poker in the back with complete faith that it will handle all situations intelligently, navigate any terrain and manage any situation
We could have them now if we equipped all vehicles and pedestrians using roads with simple device to broadcast position, velocity, and planned route.
Do you know of any good resources for learning about using P2P and WebRTC to cut down on livestreaming bandwidth costs? This is highly relevant to something I'm working on at the moment.
We also have whitepapers easily accessible on our website, and a free dashboard where you can test our technology in a few clicks.
In terms of open-source projects, there are some projects like Bem.tv but they lack robustness when you try to scale.
I agree, and since our DARPA stuff did really well the content guys have been working on getting more tech and robotics stuff.
We're currently brainstorming ways to have more effective feedback from users about the content. It's definitely in need of improvement.
And yes I am! I still try to poke the content guys every once in awhile to try to get involved somehow. All of us on the dev team and tech team have been following FPV racing religiously.
