Most combat aircraft will be autonomous by 2025

Read the forum code of contact

Member for

13 years 5 months

Posts: 593

It is my estimate that most combat aircraft will be unmanned by 2025. This means current and future UAV's like the Predator, but also older legacy aircraft that were previously unmanned, such as the F-16, the F-18, the F-22, the F-35, the Eurofighter, the Rafale, the Su-30, the Mig-21...

The reason for this is that technology is about to see major breakthroughs in two fields that will make this possible:

- The first field is robotics. I've seen video's of robotic hands and controllers being installed onto hobby planes, so the computer could "manually" fly them like a human would. This allows a computer to control any aircraft (or vehicle for that matter) that a human can (and more), by manipulating the controls the way a human would. This is important because it allows a computer to control aircraft without needing to introduce a digital control system, or to upgrade the existing digital control system to make it accessible to the controlling computer.

- The second fields is software and computing power, what most would call AI. Computers are getting powerful and smart enough to take over an increasing number of extremely complex tasks, most notably driving a car in busy traffic. USAF research has shown that software running on a low end computer will already beat human pilots in dogfights in most engagements. The UK has hinted that its latest UAVs are capable of extreme levels of autonomous mission execution, although the details remain secret.

These evolutions show that technology is quickly catching up to human pilots, and it is only logical that Air Forces around the world will quickly adapt to this technology, for a number of reasons:

- Cost: a human pilot is extremely expensive, needing to be paid a salary, health insurance, a pension, but most notable a large number of training flights to retain his skill. An AI would cost a fraction of that.

- Skill: a human pilot's skill tends to fluctuate, and flatten out, as well as be lost when he gives up his job. USAF research has shown that software can already systematically beat human pilots in dogfights, even with an inferior aircraft. These superior skills will never be lost once gained, and will likely improve exponentially as the software is improved and the cost of computing power continues to decline. In addition the skill level can easily be copied to other computers, allowing for an almost limitless number of Ace pilots.

- Endurance: AI doesn't get sleepy, tired, distracted, angry, scared, confused... It is 24/7 unblinking, can take G forces all day long and will not make stupid mistakes.

- Superhuman abilities: AI will be able to "see" in 360 degrees, with multiple sensors, as well as process data linked information. It will know the exact location and trajectory of every friendly and enemy aircraft, ground units, and launched missiles, and be able to compute all this information into a perfect battle plan. Which it will execute in perfect cooperation with its team mates as well as ground units thanks to their hive mind capabilities that will allow them to employ swarm tactics. And all this at almost instant speed, which will allow them to formulate and decide upon the most complex battle plans in moments.

- Expendability: at the end of the day fighter pilots are in an extremely dangerous position, and taking the human pilot out of the cockpit is one less body bag to be sent home.

All this combined suggests that in the next few years, companies will develop a sort of black box that will be able to fly pretty much any aircraft and execute any mission, cheaper and often better than a human could. And for obvious reasons Air Forces will adapt these systems, and those who fail to do so will pay a hefty price.

Original post

Member for

24 years 2 months

Posts: 5,396

IMO 2025 is too soon because the infrastructure isn't available. 2050 would be a better time scale for widespread autonomy within major Air Forces. Wide scale autonomy requires wide scale networks, and those do not yet exist.

Member for

12 years 7 months

Posts: 4,731

- Cost: a human pilot is extremely expensive, needing to be paid a salary, health insurance, a pension, but most notable a large number of training flights to retain his skill. An AI would cost a fraction of that.

-

when you producing less pilots and more drone operators than time will come there will be less pilot instructors and it can even impact production of pilots for Cargo planes/Civil airlines and the whole system of airforce and flight operations will be impacted. the Civil airlines have now sky high salaries.

Member for

13 years 5 months

Posts: 593

IMO 2025 is too soon because the infrastructure isn't available. 2050 would be a better time scale for widespread autonomy within major Air Forces. Wide scale autonomy requires wide scale networks, and those do not yet exist.

infrastructure? that's my point, you don't need any extra infrastructure, all you have to do is take the pilot out of the cockpit, put a computer in his seat so to speak and you're done

if anything you'll need less infrastructure because you won't need training flights anymore, or just a fraction of them
the USAF is about to buy up to a 1000 new trainer aircraft, if their jets were autonomous they wouldn't need any of those

by wide scale networks I'm guessing you mean data links so the operators on the ground can see what the aircraft see. but that's the point, there will be no need, the aircraft will only communicate with the ground when there is something to report

Member for

8 years

Posts: 17

The first field is robotics. I've seen video's of robotic hands and controllers being installed onto hobby planes, so the computer could "manually" fly them like a human would. This allows a computer to control any aircraft (or vehicle for that matter) that a human can (and more), by manipulating the controls the way a human would. This is important because it allows a computer to control aircraft without needing to introduce a digital control system, or to upgrade the existing digital control system to make it accessible to the controlling computer.

Most modern aircraft are fly-by-wire and flight control computers. These would be easier to integrate with than some mechanical system which are an obvious point of failure.
We already have remote control F-16's (used as targets), so this part is already solved.

Member for

14 years 3 months

Posts: 3,259

no way... transforming the current aircraft into drones would cost a fortune that nobody will be able to pay for. what's more, letting the aircraft do their job autonomously brings ethics problem, about target ID, killing of people by a machine with no human in the loop.

it may become reality some day, but definitely not with current platforms, which means not before 2050 at the very earliest

Member for

13 years 5 months

Posts: 593

no way... transforming the current aircraft into drones would cost a fortune that nobody will be able to pay for.

well no that's the point
the cost you'll save for not having to pay and train a pilot (and conduct CSAR missions to save him if he crashes) will save a fortune
in these times of constricted military budgets the economics of the thing will push them to switch to autonomous

what's more, letting the aircraft do their job autonomously brings ethics problem, about target ID, killing of people by a machine with no human in the loop.

there's no reason they need to be out of the loop
modern manned aircraft check with home base all the time to confirm their targets
there's no reason autonomous aircraft couldn't do the same thing

unless they're flying first day missions perhaps
but then aircraft will also engage targets beyond visual range. if anything it's safer because autonomous aircraft will not engage targets unless all their target parameters are confirmed
and if you need to confirm the target visually, you don't want to send in a manned aircraft into a dogfight with a Sukhoi, that's dangerous even for an F-22, never mind for any other type of aircraft. better to send in a drone

Most modern aircraft are fly-by-wire and flight control computers. These would be easier to integrate with than some mechanical system which are an obvious point of failure.
We already have remote control F-16's (used as targets), so this part is already solved.

yes of course, but initially it might actually be easier and cheaper for a computer to control an aircraft the same way a human would
not a programmer, not sure how hard it would be to integrate a separate computer into a modern FBW aircraft control system
but I think it's especially interesting for older aircraft
say a Mig-21, you can buy those for half a million or less. make them autonomous and put some advanced missiles on them, and you've got a cheap and effective weapons system
sure they won't be as good as an autonomous F-16, but they're fast and cheap as dirt
and because you'll only use them when needed maintenance costs won't make much difference either

Member for

9 years 6 months

Posts: 66

Sanum, while I understand where you're going, I think you are seriously under-estimating the time required to get there. The development cycle for highly sophisticated military aircraft these days is in the 15 to 20+ year range, from program initiation to IOC, and that's if everything goes fairly well. Examples such as the F-35 (1990s to mid-teens) and Global Hawk (early/mid 1990s to somewhere near the 2010 period, depending on what you consider IOC) are fairly typical, and certainly not saving the DoD much money so far (mission effectiveness is a different issue, but can also be questionable depending the mission circumstances). Budgets are already planned for 2017 and 2018 in most countries, so a new start in 2019 is probably a pretty quick start. Also, in most countries (certainly in the U.S.) saving money 20 or 30 years in the future doesn't produce much investment money in the current budget. That puts your 2025 point for "most combat aircraft" mostly in dream land (at six or seven years). Also, the money you expect to save by eliminating the pilot is eaten up by the requirement to train and man the ground-based technicians that plan and fly/maintain the aircraft (save one U-2 pilot, add 5-7 people in the Ground Control Station). You can play with the numbers a little depending on which story you're trying to sell, but there's a definite price to pay. We used to emphasize that UAVs are "unoccupied", but certainly not "unmanned", the people were just located elsewhere. Add to that the cost of communications and data movement, infrastructure, and satisfying location-unique political and air-traffic issues, and the bill adds up pretty quickly. There's also a frequently overlooked long range issue: a few decades down the road, where are the experienced pilots who inform the builders how the airframe needs to perform to accomplish a desired mission? Based on my nearly 50 years in the business, I'm not sure I want engineers to decide how to do the mission...their expertise does not lay in combat operations. I see this all the time in my current line of work supporting the acquisition of new systems; building new systems from an engineer's perspective is frequently not the same as building one that meets a user's need in the field.

So, I concur that automation and expansion of unmanned ("unoccupied" is still a better word, I think) systems will continue at a rapid pace in the future, but I think operationalizing it on a widening scale will be considerably farther down the road. Besides, what happens when the adversary hacks your system and tells your unmanned systems to shut down?? That's not an insignificant worry in these times!

Member for

14 years 3 months

Posts: 3,259

well no that's the point
the cost you'll save for not having to pay and train a pilot (and conduct CSAR missions to save him if he crashes) will save a fortune
in these times of constricted military budgets the economics of the thing will push them to switch to autonomous

well, today, the major delays for the F-35 are software related, besides the huge amounts of cash spent, it's still a good way from what's been promised, and it's mostly about giving clues to the pilot, managing this or that part of sensors and have the interact properly, and so on... Setting up an artificial intelligence that will be able to manage the aircraft and apply decisions in real time in combat is not years but decades away.

there's no reason they need to be out of the loop
modern manned aircraft check with home base all the time to confirm their targets
there's no reason autonomous aircraft couldn't do the same thing

problem being, most of the time, the visual ID is still required.. how do you ID visually when your only eyes are hundreds or thousands of miles away?

unless they're flying first day missions perhaps
but then aircraft will also engage targets beyond visual range. if anything it's safer because autonomous aircraft will not engage targets unless all their target parameters are confirmed
and if you need to confirm the target visually, you don't want to send in a manned aircraft into a dogfight with a Sukhoi, that's dangerous even for an F-22, never mind for any other type of aircraft. better to send in a drone

well, fact is, if your enemy has sukhois, you do send your fighters against them.. it's not the "writing" on its side (Su-xx) that will kill you

yes of course, but initially it might actually be easier and cheaper for a computer to control an aircraft the same way a human would

no, not in several decades

not sure how hard it would be to integrate a separate computer into a modern FBW aircraft control system
but I think it's especially interesting for older aircraft
say a Mig-21, you can buy those for half a million or less. make them autonomous and put some advanced missiles on them, and you've got a cheap and effective weapons system
sure they won't be as good as an autonomous F-16, but they're fast and cheap as dirt
and because you'll only use them when needed maintenance costs won't make much difference either

extremely hard today.. it's is not about having a remote controlled fighter, it is about having an aircraft flown and maneuvered by a computer in a rapidly evolving environment, which is simply a huge task to do. The amount of information to manage in real time is enormous, even for the latest computers available, providing you were able to program them properly which is even worse

Member for

11 years 2 months

Posts: 1,059

It is my estimate that most combat aircraft will be unmanned by 2025. This means current and future UAV's like the Predator, but also older legacy aircraft that were previously unmanned, such as the F-16, the F-18, the F-22, the F-35, the Eurofighter, the Rafale, the Su-30, the Mig-21...

The reason for this is that technology is about to see major breakthroughs in two fields that will make this possible:...

-... The first field is robotics.All this combined suggests that in the next few years, companies will develop a sort of black box that will be able to fly pretty much any aircraft and execute any mission, cheaper and often better than a human could. And for obvious reasons Air Forces will adapt these systems, and those who fail to do so will pay a hefty price.


You are watching way too many sci-fi movies.

You sound like a online version of the Popular Mechanics visions of the future, that started about a decade or so after the Wright Brothers first flight, which are 99 percent day dreaming by people who zero concept of how difficulties of reality vs. paper or computer generated wishful thinking.

Although that is a good reason for lessor countries to develop nukes. The EMP would make all these robot wunder craft useless.

Member for

9 years 8 months

Posts: 1,123

The near future will focus on manned unmanned teaming apparently. I am pretty sure the USAF has said they wanted to use the unmanned F-16s as munition trucks for the the F-35. For a2a I'm not sure the link-16 would be fast enough, but it may well be considering that the AIM-120D can get its targetting data from link-16.

Member for

13 years 5 months

Posts: 593

The development cycle for highly sophisticated military aircraft these days is in the 15 to 20+ year range, from program initiation to IOC, and that's if everything goes fairly well.

yes but that's for building an aircraft from scratch. I'm talking about upgrading the autopilot software

Boeing demonstrated in 2005 that a computer could autonomously detect an unplanned threat, decide which aircraft would be best positioned to attack, ask for confirmation from a human operator and then execute the attack
https://en.wikipedia.org/wiki/Boeing_X-45#X-45A

last year a graduate student developed an AI that could be an expert human pilot every single time in dogfighting simulations
http://www.popsci.com/ai-pilot-beats-air-combat-expert-in-dogfight

we don't need to develop this, it's already here. all we need to do is put it in aircraft and that's it

Also, in most countries (certainly in the U.S.) saving money 20 or 30 years in the future doesn't produce much investment money in the current budget.

they won't have to, companies will develop it and offer it
Air Force generals won't like it, but when politicians get wind of it that will change things
and if the West won't do it, poorer countries will, if only because it will be so much cheaper

Also, the money you expect to save by eliminating the pilot is eaten up by the requirement to train and man the ground-based technicians that plan and fly/maintain the aircraft (save one U-2 pilot, add 5-7 people in the Ground Control Station).

1) if you automate existing planes, you'll need as many ground crew to do maintenance

2) but you'll lose the pilot, and you'll need hardly any training flights anymore, which will cut the number of flights down by like 90% outside of combat tours. according to a Time article an F-16 costs about $22,514 per hour to fly and flies about 300 hours a year, in peace time. if you automate that F-16 and scrap 90% of the flights, then the cost goes from $6.6 million to $500k per year. on top of that a computer is less likely to make mistakes, like fly into the ground or collide with other aircraft, plus you don't lose a multi-million Dollar investment if your pilot dies, nor do you need to send out an extremely expensive and risky rescue mission if he gets shot down over enemy territory

3) the USAF is already looking to have on human operator control multiple UAVs. and landing will be fully automated, so there will be no longer a need for crews on the ground to land the aircraft (btw a large percentage of USAF UAV losses are due to pilot error on landing. the US Army suffers much fewer such losses because it has already automated the landings)

Besides, what happens when the adversary hacks your system and tells your unmanned systems to shut down?? That's not an insignificant worry in these times!

it is. which is a problem considering all manned aircraft today are digitally controlled, as are all satellites, weapons...
anyone who can hack an unpiloted aircraft will probably be able to hack all those others as well, they all have datalinks

Setting up an artificial intelligence that will be able to manage the aircraft and apply decisions in real time in combat is not years but decades away.

as I posted above, such software has been around for over 10 years

problem being, most of the time, the visual ID is still required.. how do you ID visually when your only eyes are hundreds or thousands of miles away?

Facebook knows when I post a picture if I'm in it, as well as any of my friends
I'm guessing by comparison recognizing a tank is a lot easier
http://www.aaai.org/ocs/index.php/AAAI/AAAI13/paper/viewFile/6435/6839

but that doesn't keep an automated aircraft from doing the same that human pilots do today, which is to contact mission control and confirm the target
unless you're flying deep into enemy territory and want to maintain radio silence, but then you're likely hitting GPS coordinates or targets likes missile launchers and radar, and yes computers have become pretty damn good at recognizing those, the F-35 does that automatically

it is about having an aircraft flown and maneuvered by a computer in a rapidly evolving environment, which is simply a huge task to do. The amount of information to manage in real time is enormous, even for the latest computers available, providing you were able to program them properly which is even worse

take a look at how the Tesla's drive in traffic, where there are waaaaaay more factors to take into account
by comparison flying or even dogfighting is super easy because of the limited number of actors involved (aircraft + missiles)

You are watching way too many sci-fi movies.

I read an article published by Boeing on what their software could do in 2005
if you run a multi-billion Dollar company that develops some of the most advanced autonomous aircraft on the planet, then do please tell us why your program failed where Boeing succeeded

The EMP would make all these robot wunder craft useless.

and likely fry every computer on every missile and manned aircraft as well, since they'd all have the same level of EMP resistance
if that's your worry, you better get those Sabres out of the graveyard

QUOTE=Hotshot;2374625]For a2a I'm not sure the link-16 would be fast enough, but it may well be considering that the AIM-120D can get its targetting data from link-16.[/QUOTE]

I'm guessing by "datalink not fast enough" you mean that the F-35 pilot will manually steer the F-16s in a dogfight?
because at BVR combat that's not relevant
and WVR, as I mentioned software running on a Rasberry Pie will let a computer outfight any human pilot
http://www.zdnet.com/article/raspberry-pi-ai-vs-usaf-colonel-guess-who-wins-in-sim-dogfight/

especially when datalinked with an F-35 with the latest sensors they'll kick ass
and you don't need much data for that, all you have to know is the location of all aircraft in the area (simple GPS location + altitude), their bearing and speed (again a simple string of numbers), and the computers will compute from there how best to engage
there's no need for large sized data streams with detailed images of the enemy, an AMRAAM doesn't need that either to do its job

Member for

9 years 8 months

Posts: 1,123

I'm guessing by "datalink not fast enough" you mean that the F-35 pilot will manually steer the F-16s in a dogfight?
because at BVR combat that's not relevant
and WVR, as I mentioned software running on a Rasberry Pie will let a computer outfight any human pilot
http://www.zdnet.com/article/raspber...-sim-dogfight/

especially when datalinked with an F-35 with the latest sensors they'll kick ass
and you don't need much data for that, all you have to know is the location of all aircraft in the area (simple GPS location + altitude), their bearing and speed (again a simple string of numbers), and the computers will compute from there how best to engage
there's no need for large sized data streams with detailed images of the enemy, an AMRAAM doesn't need that either to do its job

You may be right, maybe they'll do it. It might even be possible for 1 F-35 to control several F-16s with enough automation. For a2g it would be easier of course.

Member for

13 years 6 months

Posts: 1,138

no way... transforming the current aircraft into drones would cost a fortune that nobody will be able to pay for. what's more, letting the aircraft do their job autonomously brings ethics problem, about target ID, killing of people by a machine with no human in the loop.

it may become reality some day, but definitely not with current platforms, which means not before 2050 at the very earliest

One could argue that traps and mines are nothing but crude machines that already kill a man (or many men in case of naval mines) without a man in the loop.

The complexity of the machine doesn't change the principle. Of course this is a somewhat simplistic approach but nonetheless remains true.

Member for

13 years 6 months

Posts: 1,138

=Hotshot;2374625]For a2a I'm not sure the link-16 would be fast enough, but it may well be considering that the AIM-120D can get its targetting data from link-16.

I'm guessing by "datalink not fast enough" you mean that the F-35 pilot will manually steer the F-16s in a dogfight?
because at BVR combat that's not relevant
and WVR, as I mentioned software running on a Rasberry Pie will let a computer outfight any human pilot
http://www.zdnet.com/article/raspberry-pi-ai-vs-usaf-colonel-guess-who-wins-in-sim-dogfight/

Errrr No, a raspberry pi is not fast enough to out-fly a human. The amount of parallel information that a human brain processes every second is still too much for a Pi. Just the number of channels for data you'd have to link to it is beyond the capacity of the board.

Let's not exaggerate here.

Member for

14 years 3 months

Posts: 3,259


as I posted above, such software has been around for over 10 years

that one is sufficient.. so basically, LM is dragging its feet since late '90s for bits of software that are peanuts comapred to what you ask for, and nobody uses anything close to what you pretend will be "most used" in less than 10 yeas for... what? Fun of it?

I'm done with it... next you'll claim that flying cars existed back in 1985 because Marty McFly went to the future in one of them...

Member for

16 years 7 months

Posts: 3,765

Most combat aircraft will be autonomous by 2025

No it wont.
This entire topic is a severe "wishfull thinking" burst.

Member for

18 years 5 months

Posts: 204

No it wont.
This entire topic is a severe "wishfull thinking" burst.

:very_drunk:

Member for

13 years 5 months

Posts: 593

You may be right, maybe they'll do it. It might even be possible for 1 F-35 to control several F-16s with enough automation. For a2g it would be easier of course.

of course. all the human pilot has to do is confirm the target. at that point the computer will decide which aircraft in the swarm is the best positioned to engage the target, based on location, speed, altitude, weapons, stealth... a human can't compute that many detailed and changing variables at such speeds
and to calculate this you need only a limited amount of data, a lot of which you can gather passively by radar or optical sensors

One could argue that traps and mines are nothing but crude machines that already kill a man (or many men in case of naval mines) without a man in the loop.

and the most obvious ones, cruise missiles
the latest ones locate their targets autonomously from what I understand

Errrr No, a raspberry pi is not fast enough to out-fly a human. The amount of parallel information that a human brain processes every second is still too much for a Pi. Just the number of channels for data you'd have to link to it is beyond the capacity of the board.

lol, humans are no longer involved at the highest levels of the stock market trading. that's because computers there now work at microseconds, with a complexity far beyond the ability of a human to follow

I'm going to quote the relevant parts of the article, I'm getting a feeling you didn't bother to actually read the latest scientific research on the subject
http://www.zdnet.com/article/raspberry-pi-ai-vs-usaf-colonel-guess-who-wins-in-sim-dogfight/

Retired US Air Force colonel Gene Lee has been fighting and winning against AI opponents in simulators since the 1980s, but the seasoned tactical expert has admitted defeat to ALPHA, which only required the power of a Raspberry Pi to outmaneuver him.

According to Lee, ALPHA is "the most aggressive, responsive, dynamic and credible AI I've seen to date" -- so much so that after flying against ALPHA in realistic battle conditions, he goes home "feeling washed out" by his opponent.

"I'm tired, drained and mentally exhausted. This may be artificial intelligence, but it represents a real challenge," Lee said.

"Until now, an AI opponent simply could not keep up with anything like the real pressure and pace of combat-like scenarios."

During a recent simulated battle, Lee was unable to score a kill against ALPHA, which second-guessed Lee's every move to shoot down its human opponent during each engagement.

"I was surprised at how aware and reactive it was. It seemed to be aware of my intentions and reacting instantly to my changes in flight and my missile deployment. It knew how to defeat the shot I was taking. It moved instantly between defensive and offensive actions as needed," he said.

that one is sufficient.. so basically, LM is dragging its feet since late '90s for bits of software that are peanuts comapred to what you ask for, and nobody uses anything close to what you pretend will be "most used" in less than 10 yeas for... what? Fun of it?

you actually make a very good point
the F-35's software is probably the most advanced on the planet
if you take out the need for it to translate all its data to something the pilot can understand, it would probably be able to execute most missions autonomously

I mean according to LMT and the Pentagon it can detect, identify, track and engage targets by itself
its mostly just waiting for the pilot to approve the game plan
at that point there's no reason for the pilot to be in the same place, he can be in a bunker watching the data feed
or in a stealth mission you can set the F-35 to execute the mission autonomously, like a cruise missile

Member for

9 years 8 months

Posts: 1,123

of course. all the human pilot has to do is confirm the target. at that point the computer will decide which aircraft in the swarm is the best positioned to engage the target, based on location, speed, altitude, weapons, stealth... a human can't compute that many detailed and changing variables at such speeds
and to calculate this you need only a limited amount of data, a lot of which you can gather passively by radar or optical sensors

The F-35s share raw data between themselves to perform the fusion. They need the hi bandwith MADL for that. But it is surely possible to use the link-16 with a lower level of fusion and still be effective.

They will surely make computer simulations to test tactics using manned and unmanned fighters. Is it better to send the F-35s first to take out the enemy fighters, and then send the UCAV F-16s to finish of the other easiest targets ( with F-35 sensor help ) or is it better to send the UCAV F-16s first, knowing they would have a high chance of being shot down ( the F-35s staying a bit behind providing sensor coverage ), and then finish off with the F-35s?

Member for

14 years 3 months

Posts: 3,259

er, Sanem, you seem to believe that AI "flying" in a simulator is the same as flying a real machine... and if you're at that, it's hopeless