J-XX Fighter Aircraft

Re: J-xx

I don't know anything about the extent of technology that goes into a UCAV so maybe someone can explain it. Wouldn't the primary defense and Achilles' heel into defeating a such an aircraft would be to interrupt communications to the UCAV? Is it that hard to disrupt signals that give instructions to the UCAV? This is a perfect example of cutting off the head so the body will die. I don't see UCAVs taking a primary role unless they have technology like you see only in the movies. It's just like the UAVs today. Yeah they work great just as long as the enemy has limited anti-air capabilities. So what's the point when you can spend on something less and does the same job?

Re: J-xx

AssassinsMace said:

I don't know anything about the extent of technology that goes into a UCAV so maybe someone can explain it. Wouldn't the primary defense and Achilles' heel into defeating a such an aircraft would be to interrupt communications to the UCAV? Is it that hard to disrupt signals that give instructions to the UCAV? This is a perfect example of cutting off the head so the body will die. I don't see UCAVs taking a primary role unless they have technology like you see only in the movies. It's just like the UAVs today. Yeah they work great just as long as the enemy has limited anti-air capabilities. So what's the point when you can spend on something less and does the same job?

Click to expand...

To expose radar postion.
Clog up radar and air defence systems.
Waste the enemy air defence missles.
Confuse the enemy of which is main aircraft and which is not.

(Refering to the J-7 UCAV right now)

But for the UCAV you guys refering to. Maybe long range high level stealth bomber. Nucleur??

Re: J-xx

A problem with UAVs is that if laser point defense systems for fighters (such as that one planned for the JSF) are a workable proposition and significantly degrade the lethality of BVRAAMs, then dogfighting will come back into play, and I doubt that an AI guided fighter could fight on par with a human controlled fighter in dogfighting, given human ability to innovate and improvise.

Re: J-xx

Skywatcher said:

A problem with UAVs is that if laser point defense systems for fighters (such as that one planned for the JSF) are a workable proposition and significantly degrade the lethality of BVRAAMs, then dogfighting will come back into play, and I doubt that an AI guided fighter could fight on par with a human controlled fighter in dogfighting, given human ability to innovate and improvise.

Click to expand...

If a piece of software can defeat the world's best chess player, a piece of software can also defeat a human in a dogfight.

Re: J-xx

AmiGanguli said:

I don't know. The level of AI required would be impressive, and in general, AI development has been much slower than people predicted a couple of decades ago.

You would need the fighter to be able to act in some way independently in the event that it loses contact with the operator. It would be unacceptable to just crash, or cruise until fuel runs out and then crash. You could be over a city when this happens.

The only realistic open with current AI would be to head home if you lose radio contact, but then you would have to have a whole lot of confidence that the enemy can't jam your communications. I'm not sure if that's realistic.

Unfortunately the cooler option of fighting autonomously is still science fiction, and probably will be for quite some time. I'm not confident that autonomous fighters will be possible before the F-22s need to be replaced.

... Ami.

Click to expand...

Right now in PC flight simulator games, you can see some very imipressive manuevering done by dogfight AI. And this is just commercial game software, much less if you are actually devoting large resources to this issue. AIs have electronically fast reflexes, and do not suffer from G forces. The AI can lift off, do their missions, come home, and land in a strip. Heck they can land in a carrier too.

When you have a robot flying a fighter, this allows you to move past into the next realm of manueverbility---post 9G. You can reach the point where even missiles would easily be outmaneuvered. When you reach to this point, its back to Star Wars---projectiles and direct fire weapons. Bye bye missiles.

An AI can compute and determine exactly what the plane's energy state is, getting data from its speed, altitude, direction, engine thrust, its weight, enemy position, velocity, heading, and determine the counter move for it.

Its easier to test the AI itself through a virtual environment. Handling flgiht control is another layer that can be dealt seperately.

Re: J-xx

The bottleneck isn't the AI really. ACM isn't a general intelligence task that requires learning or logic. Imo it is sensors. For example, developing a truely autonomous visual sensor is a tough challenge.

Re: J-xx

crobato said:

Right now in PC flight simulator games, you can see some very imipressive manuevering done by dogfight AI. And this is just commercial game software, much less if you are actually devoting large resources to this issue. AIs have electronically fast reflexes, and do not suffer from G forces. The AI can lift off, do their missions, come home, and land in a strip. Heck they can land in a carrier too.

Click to expand...

That's not the limitation.

In a game the computer generates the environment and has a perfectly accurate map of its surroundings.

The AI in a real fighter would have to generate its own map of the environment using a combination of GPS, radar, and camera input. GPS is subject to jamming and can only tell you your own location. Radar is subject to jamming, has trouble dealing with stealth, and has limited ability to actually identify targets. Cameras have trouble with clouds and limited range.

Humans are exceptionally good at dealing with partial and inaccurate information. They can make reasonable judgments about how accurate a given sensor is in a given situation, and combine the information in and intelligent and context sensitive manner. They know when to trust their eyes and when to trust their radar. This has turned out to be a really tricky problem for AI.

The state of the art right now is represented by the contestants in the DARPA Grand Challenge:



They've pulled off some impressive stuff, but they don't have to deal with other contestants trying to shoot them down.

Landing on a carrier is actually pretty easy for a computer because you get highly accurate information from the carrier's instrument landing system - the carrier wants to be landed upon.

... Ami.

Re: J-xx

AmiGanguli said:

That's not the limitation.

In a game the computer generates the environment and has a perfectly accurate map of its surroundings.

The AI in a real fighter would have to generate its own map of the environment using a combination of GPS, radar, and camera input. GPS is subject to jamming and can only tell you your own location. Radar is subject to jamming, has trouble dealing with stealth, and has limited ability to actually identify targets. Cameras have trouble with clouds and limited range.

Click to expand...

Who told you that you need to generate a map? Cruise missiles don't. They have the maps pregenerated and built in. The AI does the same thing. It already has the map of the battlespace predetermined and added to its memory. All its sensors need to do is verify its location against this map.

Radar can do quite a good job of identifying targets. We have both cooperative (interogative) and noncooperative means of IFF. Identifying signals by matching them to a database of known signatures can determine the hostility of the target.

Cameras have trouble with clouds and limited range but does human eyes. Between cameras and human eyes, cameras can achieve much greater luminosity values that enable them to see better in the dark. In addition like IRSTs, they can scan with a much greater FOV and with much greater range.

Just like face recognition algorthyms used in many cameras, you can also have plane pattern or outline recognition algorthyms that scan an entire picture and home in on the visual image of a plane, or other possible targets.

Humans are exceptionally good at dealing with partial and inaccurate information. They can make reasonable judgments about how accurate a given sensor is in a given situation, and combine the information in and intelligent and context sensitive manner. They know when to trust their eyes and when to trust their radar. This has turned out to be a really tricky problem for AI.

Click to expand...

There is something called fuzzy logic that does that for AI. All you need is to take the incomplete information and match it against a set of conditions.

The AI can combine visual, radar and infrared data much faster than any human can, then integrate that with the plane's energy and flight condition.

The state of the art right now is represented by the contestants in the DARPA Grand Challenge:

Please, Log in or Register to view URLs content!

They've pulled off some impressive stuff, but they don't have to deal with other contestants trying to shoot them down.

... Ami.

Click to expand...

Doing things against a ground map is actually much more complex. A flight AI has its job much easier because it does not need to negotiate around obstacles. Basically you fly in a straightline between two points or to different way points.

Re: J-xx

Planes of same level tech as missiles will always have a hard time evading those missiles for the simple reason of agility. Unmanned planes may double the G limit, compared to today's planes but still, it is much easier to change direction of a 0.2 ton object than of a 10 ton object. I envision throttleable ramjet missiles which would use max speed to approach the target as fast as possible then slow down a bit to ensure maximum maneuverability.

As for AI, it may be all fine and dandy for attacking fixed targets, or even doing CAP missions, but going after mobile ground targets, in battlefield interdiction mode or even close air support - that will take another generation of AI to pull off.

Also, comparing chess AI to combat AI is silly. Though possible permutations in a chess game are numerous (somewhere i read it is 10 to the power of 120) that is still nothing to basically infinite number of variables in a game of war. In chess there is no randomness, there are set rules and where a figure can either be on a position or not be. In combat, just the gradients of a certain position are limitless, not to mention the number of positions, number of figures in the game and possibilities of their mutual interaction is impossible to calculate.

Contrary to various headlines in popularized science mags and shows - AI is still on the level where it depends on data inputted by its programmers. There are no real, independent learning artificial intelligences. And for decades its been impossible to tell just when that breakthrough might be made. We very well might not live to witness it.

On a positive side - learning AI may not really be necesarry for true UCAVs. With enough conditions inputted into its programme - the sheer brute power of computer calculation - combined with ruthless efficiency and reaction time in those few variables AI does know how to deal with - combined with performance increase due to removing the human limit enforced on the structure and possibly combined with greater numbers of planes (as there might come a point where even the most advanced software/computer is cheaper when mass applied than ever increasing wages of professional pilots) - all that might be enough to overcome the human opponents in most cases and thus justify those ucavs. Still, i don't see that happening in the next 30-50 years.

Re: J-xx

crobato said:

Who told you that you need to generate a map? Cruise missiles don't. They have the maps pregenerated and built in. The AI does the same thing. It already has the map of the battlespace predetermined and added to its memory. All its sensors need to do is verify its location against this map.

Click to expand...

By "environment" I don't mean the just the ground, but rather all the space around the fighter, including the ground, but also objects on the ground and in the air. Enemy fighters and missiles in particular. A video-game AI has 100% accurate information, a real-life robot has incomplete information. Incomplete information is very very hard for computers to deal with.

Cameras have trouble with clouds and limited range but does human eyes.

Click to expand...

But people can make good decisions even in the presence of uncertainty (at least in spatial navigation - humans are very bad at some other areas of uncertainty, but this we do extremely well).

Between cameras and human eyes, cameras can achieve much greater luminosity values that enable them to see better in the dark. In addition like IRSTs, they can scan with a much greater FOV and with much greater range.

Click to expand...

Certainly. And human operators can interpret the information from an IRST. The computer has serious problems understanding the data.

The AI can combine visual, radar and infrared data much faster than any human can, then integrate that with the plane's energy and flight condition.

Click to expand...

Yes and no. It can combine the information, but it can't interpret the information nearly as well. Especially because the reliability of the information will depend very much on the situation.

Where this stuff works today it's always in controlled environments where you can make a lot of assumptions about what sort of inputs the sensors will get. Even in the DARPA contest, the trees always look like trees and random objects don't move into the scene. Those robots would have serious problems if you started wandering around the desert in a tree costume. Actually, just wandering around in their field of view at all would probably confuse them. In battle the enemy will deliberately try to confuse you.

... Ami.