Monday, March 3, 2014

It's Raining UAVs

The Navy has publicly declared their intent to incorporate significant numbers of UAVs into carrier airwings.  Depending on who you listen to, carriers of the near future may include a squadron or two of UAVs on up to entire airwings of UAVs.  In addition, UAVs are being looked at for maritime surveillance, electronic warfare, tanking, air superiority, ASW, and any other role that manned aircraft now fill.  The general consensus seems to be that UAVs will become significant assets within the next few years and will dominate operations within a decade.

However, there are two related problems with this: cost and reliability.

As the UAVs take on more roles and become more complex their costs will rise sharply.  As an example, if we want a UAV that has the range, speed, stealth, payload, networking, etc. of an F-35, the UAV will cost what an F-35 does.  The incremental cost of manning an aircraft (the cockpit and life support) is miniscule and is largely offset by the additional datalinks and communications equipment needed for remote control as well as the more complex programming.  This is fairly obvious and straightforward.

What’s less apparent to most of us is the current state of reliability of UAVs in general.  Data on UAV operations is very difficult to come by but data from the U.S. Air Force Safety Center reveals that a total of 79 General Atomics MQ-1/9 Predator/Reaper fixed-wing UAVs have been destroyed in accidents, with another 21 seriously damaged. The Air Force acquired 248 Predators and 156 Reapers through the end of Fiscal Year 2011 (1).  That’s 100 damaged or destroyed UAVs out of 404.  I was stunned when I saw that.

What the data doesn’t indicate is the cause of the various incidents.  Anecdotally, communications loss seems to be the leading factor.

Back to the earlier point, how are cost and reliability related?  As the UAVs become more capable and complex and costs rise, the loss rate is going to quickly become unacceptable.  Consider the low side of the cost curve for our F-35-ish UAV example.  Losing 100 F-35-ish UAVs at $100M per aircraft for a total of $10B is simply unaffordable.

Consider the losses described above – those occurred in benign electromagnetic environments.  What kind of losses would be incurred in a hostile environment with challenged communications?

ComNavOps is on record as being doubtful about the controllability of UAVs in an electromagnetically hostile environment.  However, I had been under the impression that operations in benign environments were reasonably reliable.  This is an eye-opener.

Now, to be fair, I would assume that communications and control reliability are constantly improving.  Nevertheless, this suggests that we are still very far away from reliable control in combat.  Much, much further than I had thought.  The website lists 73 USAF Predator/Reaper/Global Hawk crashes since 2007 (2).

I’ve been unable to find corresponding Navy UAV data but I was able to document the following incidents in the last few years.

2 August 2010, an MQ-8 Fire Scout became unresponsive to commands during testing and entered restricted airspace around Washington, D.C -Wiki

21 June 2011 Fire Scout lost communications and crashed over Libya.

30 March 2012 Fire Scout crashed off the coast of Africa after it was unable to land on the frigate Simpson at the end of a surveillance mission. 

6 April 2012, another Fire Scout crashed in Afghanistan.

11 Jun 2012 BAMS-D (RQ-4 Global Hawk) crashed in Chesapeake Bay, MD

24 Oct 2013 MQ-8B crashed at Patuxent River

15 Dec 2013 MQ-8B crashed in the Atlantic

I’ve also found documentation of two BQM-74 drone crashes due to communications loss.  The BQM-74 is the drone that just recently crashed into the cruiser Chancellorsville during a training exercise.

I’m not advocating stopping the development of UAVs, just pointing out that the communications and control issues seem not to be as far along as the military would have us believe.  We also need to be cognizant of the cost-reliability relationship.  Finally, I’ll repeat:  I’m highly dubious about the controllability of UAVs in combat.


  1. My standard gag line about large full-spectrum UCAVs which can handle all facets of aerial combat goes like this .... if we initiate a crash program to develop such large UCAVs, bunches of them will crash before we finally get it all figured out.

    1. Scott, a crash program ... outstanding!

      On a serious note, though, why the enormous loss rate compared to a manned developmental program? Is it just the unmanned nature makes people willing to accept technical risks that we wouldn't otherwise? Or, is it the communications (as I suspect) where, unlike a pilot in the cockpit, if you lose comm you lose the aircraft? Something else?

    2. CNO, at a first order approximation you can think of the existing active drones as basically really advanced RC controlled planes (using really complex and round about RC methods). That combined with somewhat rushed development.

      The drones that the Navy is interested in have rather independent autonomous control systems, they really do fly themselves. A Predator/Reaper/Globalhawk is in contrast just remotely piloted. For X-45/X-47, there is no remote pilot, just a mission planner. The X-45/X-47 are designed/built to fly themselves.

      So, if the crash rate of the existing drones is comm related, the next gen drones will pretty much be immune to that issue.

    3. “On a serious note, though, why the enormous loss rate compared to a manned developmental program?”


      Big logical issue with this statement.

      The “problem” isn't that UAVs have an "enormous loss rate compared to a manned developmental program". It only appears that way because we have been flying UASs a lot more than manned aircraft.

      A "rate" is by definition some occurrence over a unit of time. So you need to look at both the numerator (# of mishaps) as well as denominator (# of accumlated flight hours).

      For aviation safety, the numerator is typically the number of Class A mishaps. This is a mishap in which the damage exceeds $1 million, total destruction of the aircraft, or there is a loss of life.

      The denominator is the number of accumulate flight hours. If two systems have the exact same number of Class A mishaps, but you fly System 1 twice as much as System 2 during some period – then System 1 will have a lower mishap rate than System 2.

      If one looks at the link below, MQ-4 Global Hawk and MQ-9 Reaper actually appear to have very similar Class A mishap rates to the U-2. And both are currently quite a bit lower than the F-16.


    4. Good reply, Matt. Always nice to see some raw data.

    5. B.Smitty, if you look at the raw source data, the 10 and 5 year average mishap rates are

      F-16 10=1.92, 5=1.96
      MQ-1 10=6.28, 5=4.79
      RQ-4 10=2.47, 5=3.34
      MQ-9 10=4.27, 5=3.17

      So, the raw mishap rates are all greater than the F-16 and are around 1.5X - 3.5X worse than the F-16.

      Source data is from the website

    6. CNO, there’s fundamental problems with that approach too!

      1. Clumping MQ-1, MQ-9, MQ-4 and F-16 into the same bins is way too simplistic. They don’t perform the same missions or have the same flight profiles. There’s probably a reasonable comparison to be made between MQ-9 and F-16 in strike mission, and U-2 and MQ-4 in ISR mission. MQ-1 is probably best compared to a low-altitude A-10 or a helo gunship

      2. Looking at “raw” rates over a five or even ten year period compares an aircraft that has been around 40 years (F-16) to UAVs which have been around a little more than a decade.

      Analysts control for this by looking at mishap rates as a function of total accumulated flight hours for that aircraft type (e.g. after 100K accumulated flight hours, F-16 mishap rate was X). This allows an ‘apples to apples' comparison.

      So... using the methodology and bins I just described and the Air Force Safety Center derived data:

      Mishap Rate at 10,000 accumulated flight hours.
      - ISR: U-2 and MQ-4 had similar mishap rates.
      - Strike: MQ-9 was lower than F-16.

      Mishap Rate at 50,000 accumulated fight hours
      - ISR: MQ-4 had slightly lower mishap rate than U-2.
      - Strike: MQ-9 was lower than F-16.

      Mishap Rate at 100,000 accumulated flight hours
      - ISR: No data on MQ-4, but if trends continue will be less than U-2.
      - Strike: MQ-9 was much lower than F-16.

      Bottom line: MQ-4 and MQ-9 don’t appear to be ’failing from the sky’ provided you look at the data properly. The trends all seem to point to the fact that UAS mishap rates over the long term are slightly lower than equivalent manned aircraft.


      PS - Here’s a link to the chart I cited earlier.

    7. On caveat to that, Matt. By its nature, the F-16 is flown much harder than any of the UAVs mentioned. F-16s have to perform air combat exercises. So the UAVs should have a lower mishap rate, all else being equal.

      But yes, I agree with you, you have to look at each system at the same point in its maturity cycle.

    8. B. Smitty - that is a good point. One final sidebar:

      - In the military's mishap statistics, any mishap that results in a loss of life is automatically elevated to Class A.

      - I'd imagine Class A mishaps involving high performance jets quite frequently involve pilot injury and fatality.

      - I've never seen stats which address the true "costs" of a lost pilot. I'd imagine if you tallied up the education, training, readiness, etc. it's easily in the single digit millions.

      - No UAS mishap have ever led to the loss of the operator's life. If the UAS crashes, he/she lives to fly another day.


  2. Com loss is probably a big factor, Navy Warships have all sorts of communications equipment and even they can’t keep all the channels open at all times. And you can put a lot more redundant systems on a ship then a aircraft, and have people standing by to fix, reconfigure, or shift to another mode as needed.

    Another factor is the lack of pilot awareness. There are probably lots of occasions where the pilot hears, or see’s or feels something is not right and either changes something or returns to base. Things that the drone operator might not pick up until its too late

  3. I am on the same page as the CNO here. The record of super sophisticated UAVs so far has not been the best. Remarkably simple ones with limited roles (gunfire spotting, recon, laser designation), on the other hand, are very reliable. This is a very easy place for us to over reach to quickly, such as another chance to repeat LCS. They will be very useful in lots of roles one day, but I bet combat aircraft and the situational awareness associated with being in the actual cockpit will prove superior for a long time. The lack of SA associated with sitting behind a screen at the worst, or in a simulator at the best, from a thousand miles away will lose us more combat aircraft-like and combat aircraft-like costing airframes far, far faster than we can afford.

  4. It should probably be pointed out that we are also talking about two completely different categories of UAVs here. The Predators/GlobalHawk et al are basically remote piloted drones with all that that entails. X-45/X-47 are autonomous remote commanded drones. The "pilots" of the X-45/X-47 don't really fly the planes, they more or less enter flight and operation plans for the planes.

    A lot of the work Boeing did on X-45 (which was subsequently used on X-47 A/B) was making the drones as autonomous as reasonable. Those X-47B carrier landings weren't flown by hand, the drones were told to land in pattern and did so, even to the point of one of them calling its own abort when it detected an issue and re-routing to an airfield. As part of their command set, they can even be told to follow another plane and will do it, regardless of if it is a drone or human piloted plane. They really are an entirely different bread than the drones currently in use. Boeing at one point even built a physics correct simulator with complex models of the drone hardware that did complete missions including comm drops. The X-45/X-47 basically can do things that no other drones out there can do, their software is more akin to a pilot and a terminus for remote control.

    Basically what I'm saying, is we are designing the next generation of drones to basically work on their own even in the presence of comm blackouts. They'll be given a mission plan, they'll execute their mission plan, if comms are available they can receive updated mission plans just like pilots. They'll come back and land without remote pilot intervention just like a manned plan. Etc.

    So if the majority of the existing UAV failures are related to RC plane like remote control issues, then the next generation of UAVs won't have those issues. Granted, they'll probably have entirely new sets of issues, but that's par for the course in both UAVs and manned planes.

    1. ats, in concept and on paper, you're right. The next gen UAVs will have no comm issues because they'll be autonomous. On the other hand, on paper, the LCS was an amazing ship, the LPD had no manufacturing problems, the JSF entered squadron service a decade ago and is the greatest plane ever developed, the Ford's EMALS and AAG worked flawlessly, etc. You get the point. The next gen UAVs will have their own set of problems. To be fair, I think you recognize that.

      Frankly, I don't think we can program autonomy to the degree we want for the next gen UAV. If all we want is for the UAV to fly through some waypoints and attack a fixed target, they'll undoubtedly do that although, come to think of it, we already have that - it's called a Tomahawk. If, on the other hand, we want a UAV that can decide its own path based on threats it senses, select its own targets after it arrives on site, co-ordinate targeting with other UAVs, engage in self-defense enroute, egress aggressively based on threats, and fight its way home, I'm pretty sure we won't be able to do that for decades to come, if ever.

      Look at the programming problems we're having now that are far less challenging. For example, the F-35 can't present data to the magic helmet or even operate the automated maintenance system despite a decade or so of development.

      I'm absolutely not saying we stop development just because there are challenges. I'm just saying that we need to recognize future reality and distinguish it from Powerpoint slides. Before we commit to all-UAV airwings, let's make sure they work - unlike the total committment to the LCS before we were sure it would work. To be fair, the UAV program is, generally taking a responsible, steady path of development and I have praised it in a post for exactly that. I am troubled, though, by the recent jump in postulated capabilities which smacks of LCS-ism.

    2. CNO,

      We've been doing automated route planning with threat avoidance for decades. My wife worked on a system for a Navy aircraft back in the '90s. Doing so real-time doesn't seem insurmountable.

      Heck, if Siri can get me from my house to my brother's house in California, with unplanned detours to see the world's largest ball of string, I think we can build something that can pick the best route through new threat emitters. It may take a while to make it robust, but at least we don't have to train every pilot who flies a manned combat aircraft how to do this. Once the software is written, it can be replicated as many times as needed at no cost.

      Automatic target recognition is a harder problem, especially when the enemy uses camouflage and deception. Of course, as we've seen in recent wars, manual target recognition is difficult in these situations as well.

      Coordinating and deconflicting multiple UCAVs and manned aircraft in a crowded CAS stack is another potentially difficult problem.

      Having multiple UCAVs autonomously coordinate with each other is another difficult problem.

      One way forward with these issues is to incorporate a manned "coordinator" to review choices made by UCAVs and adjust them as needed. Ideally this coordinator would be on a manned aircraft using hard to jam, line-of-sight comms. It could stand off, but may still have to penetrate air defenses to some degree. This may mandate a survivable, manned aircraft with similar range characteristics.

      LRS-B is a potential candidate. A two-seat F-35C might have been but, oops, no two-seater.

    3. B.Smitty, you're looking at the most simplistic case of threat avoidance - fixed, known emitters. Yes, we probably can do that. Now, what if want the UAV to factor in airborne threats (fighters and AEW), ground threats that cycle their radars, and so on? I don't believe we'll be capable of that for quite some time, if ever.

      The rest of your comments are good observations and demonstrate that you recognize the magnitude of the programming task.

      The manned coordinator is an interesting concept and worth some consideration although I can see some problems. The main one is that we want UAVs precisely so that we don't risk men. A coordinator aircraft will, presumably be akin to a Hawkeye in that it would require multiple people to man consoles. I'm dubious that a single man could handle the coordination. Such an aircraft would be non-survivable and/or require significant protection and then we're right back to what amounts to a manned strike. Still, worth some thought.

    4. CNO,

      Even adding airborne threats doesn't change my mind. I still think dynamic route planning is one of the easier tasks. We can build autonomous robotic vehicles that drive a 60 mile, complex, urban course without any human intervention.

      That's hard. Popup threats doesn't seem that bad. I have a feeling we could design a system that would do as well as a human in the cockpit at avoiding threats. We may still lose aircraft, but that's the nature of war.

      The local, manned coordinator is only need in a SATCOM-denied environment. The aircraft could also just serve as a comms relay station, outside the range of enemy jamming. That way, the bulk of the brain power could remain elsewhere in the world. It may not even need to be manned. Another UCAV could carry a comms relay payload.

    5. CNO,

      Its worth pointing out that in many cases the routes actual manned craft are using in response to dynamic threats are already computer generated. The systems to do military automated route planning have been around for over a decade now. The algorithms aren't incredibly complex and its actually something that a computer generally does both better and faster than a human. So I see no reason an autonomous UAV would be any worse at threat avoidance than a human pilot.

      The coordinator role isn't as complex as you are making it out to be. Its basically a negatory override response. A single back seat should be able to do it just fine. One thing that makes it a lot easier is that the autonomous UAVs use a variant of a flocking algorithm for group flying, so the coordinator would be handing generally a single command update for the group rather than one per UAV.

  5. You're spot on. At the center of all this is the attempt to run our U(X)V programs like F-35 or LCS. However, we do NOT need drones that do everything. Making an unmanned F-35 doesn't make it easier or better... it just makes it unmanned.

    We need to think how we can build platforms built from unmanned, not from manned, concepts. We should be thinking, for example, how to build a missile-bearing missile that can return to base, not a F/A-XX without a pilot. We can develop countermeasure drones, low altitude ASCM tubs, etc... Sure, one day we'll have our metal Mavericks... but not with our current tech. Development at lower levels will get us there.

    Our real salvation here is going to be breaking the autonomy barrier. Yes, if you're using an data-link that can drop or be hijacked, you're going to lose A/C same as if you were to have a pilot black out.

    1. As I see it, developing full autonomy in an unmanned UCAV which can handle a full-spectrum combat threat environment is an order-of-magnitude more difficult and expensive problem to solve than simply getting an unmanned airplane to automatically return to its launching carrier and land safely.

      Getting from here to there will involve years of slog-through-the-weeds software code development work, combined with build-a-little, test-a-little, learn-a-lot operational testing of each automation approach as it will be applied to an unmanned combat airframe.

      This must be a very deliberate, one step at a time process which ensures that the tortoise eventually crosses the finish line to the accolades of the race fans, while avoiding the problem of the flashy but overly-confident hare scampering in all directions through the DOD acquisition wilderness seeking the next magical technology solution.

    2. Scott, I agree completely. Unfortunately, the Navy seems to have abruptly changed from its steady, responsible development to a desire to jump a generation straight into ultra-sophisticated deep strike and/or air combat UAVs. It smacks of LCS thinking. In the space of a few months I've gone from thinking the UAV program was one of the best to thinking it's going to crash and burn like an airborne LCS.