Skip to content

Air Superiority Drones

March 15, 2010
tags:

I thought the following comment at Strategypage, in a post about turning old F-16s into unmanned targets very interesting:

The UAV version of an aircraft is superior, in some ways, to one with a pilot in it. This is mainly because pilots black out when the aircraft makes turns too sharply, at high speed. The air force discovered how effective this capability was during the 1970s, when they rigged some jet fighters to fly without a pilot, and had them go up against manned aircraft.

So,we’ve had this capability since the 1970s and have yet to deploy an air superiority UAV? I suspect someone is going beat us to it eventually. Then there is Gorgon Stare.

About these ads
13 Comments leave one →
  1. Scout permalink
    November 3, 2010 2:34 pm

    I’m pretty sure they already exist in great numbers and varieties.

    Why did Obama cut funding on the F-22 calling it obsolete?

  2. Jim permalink
    April 3, 2010 5:21 am

    Matt,

    I am taking the whole system into account. Below I address your concerns about jamming and AI by discussing the state of the art in each area. Keep in mind that the stuff I am discussing is simply what can be done using the latest ideas in research, not what is cheap enough and tested enough to actually deploy yet (although it certainly will be in the not-so-distant future).

    >If UAV loses comms to its operators back in CONUS

    They will be local, as I said before. I, like the rest of the people here, am very nervous about anti-sat missiles, but the solution to that is simple: get the pilots within RF range. That eliminates everything except the last RF hop.

    As for the last hop, there are numerous ways to defeat jamming (increase power, error correcting codes that are as redundant as necessary, frequency and channel hopping in seemingly random patterns, spread spectrum techniques, filtering out of any non-random jamming signals, probabilistic coding such that can be distinguished even out of random noise, network coding, cognitive radio, etc.). Our enemies thus far have just had low-tech radios. Even with the most advanced jamming techniques (short of a gamma ray burst from outer space), completely blocking all forms of communication on the entire spectrum is not impractical.

    Of course, even if the enemy manages to jam the system (which won’t mean complete black out, but more along the lines of reduction of bandwidth below what is necessary for effective control), we can fall back on AI to control the plane, just like the F-22 falls back on its pilot.

    >And I think we are a long way from having an AI that can out-think an even mediocre fighter pilot.

    As I said below, I disagree on that one.

    First, pilots are already heavily assisted by AI in a modern fly-by-wire system. In fact, an F-22 wouldn’t be possible for a human to fly without these systems.

    Second, the drone has an enormous advantage in its ability to maneuver to because of the lack of a meat payload inside to black out. Envelopes matter just as much, if not more than pilot ability. A person cannot take much more than 4-6 Gs for a sustained period without passing out. Even the best flight suits can only add a couple Gs at most to that number. Just for comparison, the electronics in an artillery shell are engineered to take 15,500 g (http://www.iechome.com/news/032003.htm). So, the computer can take pretty much anything and the only limit is the air frame. The US has built platforms before that can take 100 gs (the Sprint Missile). And the F-22’s airframe is easily capable of knocking out the pilots (http://abcnews.go.com/US/wireStory?id=8249793). I’d guess that a drone has a 5 G+ advantage here, which gives it a lot of wiggle room to be less intelligent than a human pilot.

    Third, the modern AI toolbox is much more capable than most people realize. Sure, we don’t have computers that can talk like a human, but we do have computers that can do almost everything else. Modern machine learning techniques are pretty incredible at complex pattern recognition. Compared to many of the problems we deal with in AI, an air-to-air combat scenario is actually fairly simple. Computers currently design medicines and machines much better than humans can using genetic algorithms, trade stocks using support vector machines, forecast weather with artificial neural networks, route all of the packets on the Internet using distributed shortest-path heuristics, and do many other applications that would be literally impossible for people to do. Also, a computer can process input from all of its sensors faster, in parallel, and in much more detail than a pilot can. The only advantage that I can think of that pilots currently have is the image processing aspect, but that gap is closing (and might already be good enough for collecting enough state to do air-to-air combat, I’m not up on the latest in that area).

    Fourth, even if all of the above still isn’t enough to beat a human, we can simply overwhelm the human with raw numbers. Good pilots are extremely expensive to train and maintain. The planes that support good pilots are also extremely expensive to train and maintain. The one thing that drones have absolutely kicked ass at so far, is being cheap per unit. If we can go build 10,000 drone fighters and don’t care if some of them get shot down, then we can use a zerg-rush style strategy and always win.

    If you don’t want to believe me on the AI stuff, fine. I have studied this stuff in detail and am giving my personal opinion. You are entitled to yours, but just know that educated people do disagree with you. Also realize that AI will never get worse, it only gets better over time.

    To conclude, I personally believe that most, if not all, of the raw ingredients are available now. The only real constraint is engineering costs for getting such a system right. The thing about drone fighters is that they not only have to be as good as humans, but they also have to be cheaper, so that will keep humans flying fighter planes for the time being.

  3. Matt permalink
    April 2, 2010 5:38 am

    Jim,

    Jim.

    Again — you have to consider the whole system.

    If the F-22 “system” loses comms to the AWACS, there is a fail-safe — the pilot in the seat. If UAV loses comms to its operators back in CONUS there is no fail-safe — the system breaks. And I think we are a long way from having an AI that can out-think an even mediocre fighter pilot.

    The US does broad-band jamming all the time, and we go out of our way to make sure the unit that’s doing it is protected and survivable. I would think a smart bad guy would do the same — so your assumption that we’ll maintain complete control of the RF spectrum in wartime is a bit suspect.

  4. Jim permalink
    April 2, 2010 3:35 am

    Matt,

    We already have to have comms that very reliable for our current fighters. An F-22 does not operate by itself to be at its most effective. It needs the help of AWACS, sat data, the rest of the squad, and tons more to really be a reliable winner. So, I’m inclined to think that a UAV is not much worse off than current fighters.

    Second, like I said above, some fairly simple AI control system would probably suffice given that it would have a huge advantage in maneuverability. If the AI turns out to be good enough, then the comms problem becomes equivalent to what the F-22 needs.

    Finally, broad jamming just makes one the highest priority target in today’s radio-dependent battlefield, so I don’t think that is a serious problem.

    Either way, I think the DoD is building them anyways, so hopefully we will have some actual evidence in a few years. Building them is the only way to improve the problems we’ve discussed.

  5. Matt permalink
    March 30, 2010 9:59 am

    Jim,

    As I’ve told Mike before, one has consider the whole unmanned air system, and not just the air vehicle. There is one glaring gap in a UAV fighter concept — reliable comms back to the operator.

    The RF spectrum is already in pretty short supply and they aren’t making any more of it. In addition, the bad guys can do an awful lot of interesting stuff to deny our use of it – like jamming and satellite destruction/denial.

    Until we lick that problem, my personal opinion is that a UAS fighter is a non-starter.

  6. Jim permalink
    March 30, 2010 7:52 am

    I’m a PhD student in computer engineering with lots of background in computer architecture, networking, and AI. I am also a lifelong aerospace nerd (almost picked that as my focus instead of computers) and military nerd. Here is my take…

    In my opinion, the AI for flying a fighter does not need to be extremely advanced for it to be able to defeat a human pilot. If the practical envelope of the plane is opened up into ranges that black humans out, even a less intelligent system than a human can probably outmaneuver and win in a dogfight. Also, drones are cheaper so we can put up 10 drones against each human if necessary. Another thing to consider is that video games and various other applications have pushed AI much farther than most lay people realize. Machine learning techniques are pervasive in nearly every major computer system that is newly deployed. Overall, I am confident that the flight control system is not an issue. Also, these planes will almost certainly be fully human-controlled for the time being or, if necessary, at least human-assisted with auto-control for fast reaction times during certain types of engagement.

    Secondly, any issues with network latency between the human pilot on the ground and the plane can be overcome. For a fighter, the pilot will almost certainly have to be in theater to reduce latency (for a recon plane it doesn’t really matter, but bouncing off sats or crossing ocean cables adds crucial milliseconds that are easily eliminated by proximity). As for in-theater latency issues, I do not think these will be a problem. It is standard radio-latency and we are very good at dealing with that. Note that it isn’t just the speed of light that slows down communications across the whole world, but also router queue delay, latency to go to LEO and back, and much more. All of these go away with having the pilot close by. However, the pilots can be moved back to Kansas or Utah or where ever if it is determined that the AI is good enough for engagements (remember, the F-22 doesn’t really need to dogfight, it wins by beating the enemy before it is even detected).

    Finally, if computers cannot handle the job now, someone will figure it out in the next few years, especially given the huge advantages an air-to-air drone could give (mainly cost, which the brass and politicians will love). From my perspective, security is a much bigger concern than the AI or network performance issues. In theory, the security of a well-designed network protocol is sound, but implementation always differs slightly from the specification. Therefore, a key cost for this (and all drones) is to spend lots of effort on verification of the security software implementation. We cannot afford to have someone with a laptop and a cheap transmitter to take over a fighter plane.

    Either way, I guarantee the guys at Lockheed and Northrop are working very hard on solving these issues as we speak and they rarely disappoint. I fully expect the DoD to have a public announcement of a fully functioning air-to-air combat drone around 2020.

    I hope my perspective was useful to you all.

  7. Chuck Hill permalink
    March 16, 2010 3:00 pm

    When we talk air to air combat, a lot of times the advantage goes to the pilot with the best reaction times. When you start talking about drones, just the signal time between sensor, operator, and platform becomes significant unless the operator is close to the sensor/platform. Remember those long delays when the news anchor on the East Coast is talking to the correspondent in ‘stan. We can’t expect operators sitting comfortably in Kansas to fight air to air half a world away. The speed of light is too slow. Even a few hundred miles might be a problem.

  8. Jed permalink
    March 16, 2010 10:10 am

    Ooops – I should qualify my last statement: I think a “missile launch vehicle” is a possibility – carrying some AIM120C aloft, with missile guidance being provided by AEW or other fighter aircraft – but I don’t consider that to be unmanned fighter. Also same issues where it comes to degredation of C2 and missile uplink datalinks.

  9. Jed permalink
    March 16, 2010 10:08 am

    Science fiction Mike – until the ‘artificial intelligence’ is good enough to create a Cylon Raider, forget about it.

    What makes an U.S. airforce F15 or F16 better than a Mig 29 or Su30 series fighter ? The highly trained professional in the cockpit !

    Until there is a computer that is as capable at multi-tasking, sensor fusion and communications management as the pilot, air-to-air fighter UAV is a non-starter.

    As for flying them remotely via data-link, please believe an ex-Navy Signaller who also did time in the Royal Corps of Signals as a reservist and is now an IT professional, your statement: “I don’t see the bandwidth issue as a game changer. Important, yes, but manageable.” is way off the mark.

    Listen to the latest installment of the MidRats podcast, where the reliance of USN surface units on IP comms via satellite is slated. This issue was also discussed recently in comments on Solomon’s SNAFU blog – cyberwarfare and simple jamming are going to bite the US in the arse at some point if you think you can use all these UAV’s against a better equipped and prepared enemy – and as we seem to be talking about air-to-air use I presume we are not talking about Afghanistan.

  10. Matt permalink
    March 16, 2010 7:57 am

    Mike,

    A UAV is not the same as a drone. Any unmanned air vehicle can be a target one time — being able to sense an adversary and shoot back is what makes a UAS valuable.

    A UAS is much more than simply the UAV. You have to consider the whole chain — which includes the off-board operators and the communications link.

    If you don’t see communications bandwidth as a potential show-stopper for a fighter-type UAS, I would urge you to look closer at this issue.
    The Services are struggling as it is to find sufficient bandwidth for their long-dwell ISR UASs.

    I would imagine a fighter-type UAS would require much larger “pipe” per platfrom, and the data transferred would have to be very high fidelity, multi-spectral (radar, EO/IR), and near instantaneous.

  11. Mike Burleson permalink*
    March 15, 2010 8:37 pm

    I don’t see the bandwidth issue as a game changer. Important, yes, but manageable.

  12. papa legba permalink
    March 15, 2010 7:22 pm

    Also, I’m curious about the bandwidth requirements for the Gorgon Stare sensor. A wide-angle viewport of any resolution seems like such a signal hog that it would beg to be jammed.

  13. papa legba permalink
    March 15, 2010 7:20 pm

    A target drone is very different from a craft that has to operate in hostile airspace with live weapons. Converting an old plane to a target drone is far short of using them as UAV’s. Target drones like this have been around for decades– F-4’s were fitted as such. I’ve heard of proposals to load down old airframes with explosive and use them as cruise missiles, as well.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

Follow

Get every new post delivered to your Inbox.

Join 236 other followers

%d bloggers like this: