USAF/US NAVY 6th Generation Fighter Programs - F/A-XX, F-X, NGAD, PCA, ASFS news

[Scott Kenny]

That's just a box with a computer in it. Hardly generation defining. Why do you even need to host the AI mission manager on a fighter vs on an E-7, or on the ground/ship? We're very much talking about outer loop control vs the likes of a sensor fusion algorithm that is inner loop and much more dependent on the architecture.

I'm not seeing why integration with older platforms is an issue. You've just got a different interface that you need to adapt to what is already there.

Ideally, you'd have the AI mission manager on the CCA, so that any plane with the right datalink can tell the CCAs what to do.

But I suspect that the mission manager is expensive enough that it needs to be on the 6th gen aircraft in order to make the CCAs sufficiently cheap and attritable.

 

[Hanz2k]

But I suspect that the mission manager is expensive enough that it needs to be on the 6th gen aircraft in order to make the CCAs sufficiently cheap and attritable.

Yes, and no . It is software and it will be airframe agnostic defined only by mission set. Airframe capabilities incl. available weapons are only one of the variables. If you can run GPT at home, then most probably you can put such model on the CCA airframe.

 

[GTX]

What are the de facto expected capabilities of 6th gen fighters?

• Increased size and thus space for installation of future-avionic upgrades
• Larger/more spacious weapons bay(s)
• VCE/ACE Engines
• Much more fuel for longer range/patrol time
• Increased computation power, cooling and power generation for DEW and much more powerful sensors
• Enhanced sensors
• Ability to command Loyal Wingmen/CCAs
• All-aspect broadband Low Observability

and so on...

Since we're talking about CAC's aircraft; way I see it, the J-36 design fulfills all of these requirements perfectly.

I am curious as to how you determine that this platform has all of those capabilities based upon the videos released to date...

 

[Josh_TN]

Since the LLMs are starting to show collapses due to them "eating"/analyzing LLM-created bullshit, I doubt that we will see LLMs on military aircraft.

Something as ubiquitous as LLM AI that has been eating garbage all its life is dangerous and unnecessary. Aircraft AI will be given narrowly defined data that has been tagged and cleaned by humans and used for specific tasks to give automation a facsimile of intuition: recognizing new EW wave forms and countering them with ECM, recognizing new target types or targets that are masking their EO/IR signature with camo or light emitters, and of course the software agent(s) that actually fly the CCAs. There is little reason not to install one on manned NGAD as a fallback; it could handle surprise situations while the aircrew focused on managing a wider network of unmanned aircraft.

 

[Josh_TN]

That's certainly where I want the CCAs to be, but I don't know that the AI will be able to get there in 5-10 years.

AI has been 10 years away for about the last 30 (at least), just like fusion power.

I think the progress in AI in the last 5 years rather staggering. The LLMs went from chat bots to passing the bar exam. I suspect by 2030 AI will be used with no oversight in high risk opponent airspace. Geo fence it up to the FEBA and then let it go weapons free. With large numbers of disparate air vehicles, it will probably be faster to organize the effort itself rather than having a human attempt to micromanage the swarm.

 

[Josh_TN]

Yes, and no . It is software and it will be airframe agnostic defined only by mission set. Airframe capabilities incl. available weapons are only one of the variables. If you can run GPT at home, then most probably you can put such model on the CCA airframe.

Computing power increases in power and/or drops in cost at a staggering rate. Perhaps the first few increments of CCAs will not have a “big picture” type software agent, but I think a few iterations in they will. If we are talking about off the shelf commercial processors, it simply will not be prohibitively expensive to give a multi million dollar CCA an air management AI. Smaller stuff - more disposable UAVs and expendable decoys - can be directly controlled by the CCAs.

 

[_Del_]

Computing power increases in power and/or drops in cost at a staggering rate. Perhaps the first few increments of CCAs will not have a “big picture” type software agent, but I think a few iterations in they will. If we are talking about off the shelf commercial processors, it simply will not be prohibitively expensive to give a multi million dollar CCA an air management AI. Smaller stuff - more disposable UAVs and expendable decoys - can be directly controlled by the CCAs.

Drones will continue to see AI improvements, but the thing no one ever looks at for some reason is the ABMS. Once the secure beyond-LOS TDL's have been solved (largely already true, if lacking dedicated hardware and a "clear" commitment across branches to the exact pathway [SIMPLE, Link-22, -16, etc] at the moment), the main advances can come away from the limitations of the SWAP-C and pricetag of the platform.
Spent a lot of money planting seeds for ABMS, IBCS, Pit Boss, Maven, etc. Finally reaching benchmarks to harvest that crop.


Now they just need to figure out who on the network needs to see what, reliably and quickly.

SAM.gov

sam.gov

 

[quellish]

Drones will continue to see AI improvements, but the thing no one ever looks at for some reason is the ABMS. Once the secure beyond-LOS TDL's have been solved (largely already true, if lacking dedicated hardware and a "clear" commitment across branches to the exact pathway [SIMPLE, Link-22, -16, etc] at the moment), the main advances can come away from the limitations of the SWAP-C and pricetag of the platform.
Spent a lot of money planting seeds for ABMS, IBCS, Pit Boss, Maven, etc. Finally reaching benchmarks to harvest that crop.


Now they just need to figure out who on the network needs to see what, reliably and quickly.

SAM.gov

sam.gov

Yes, and the first "NGAD" hardware to be fielded is actually an ABMS pod fitted to MLP in the next year or so.

 

[Training_Dummy]

Yes, and the first "NGAD" hardware to be fielded is actually an ABMS pod fitted to MLP in the next year or so.

Ive tried for too long on a weekend to look up what MLP means in this context…. I know Pegasus is getting ABMS as part of CR1 but am i missing something obvious but non-official with “MLP” apologies and thx in advance

 

[NMaude]

When I read "MLP" the first thing that comes to my mind is "Mobile Launch Platform".

 

[quellish]

When I read "MLP" the first thing that comes to my mind is "Mobile Launch Platform".

My Little Pony. KC-46

 

[siegecrossbow]

Computing power increases in power and/or drops in cost at a staggering rate. Perhaps the first few increments of CCAs will not have a “big picture” type software agent, but I think a few iterations in they will. If we are talking about off the shelf commercial processors, it simply will not be prohibitively expensive to give a multi million dollar CCA an air management AI. Smaller stuff - more disposable UAVs and expendable decoys - can be directly controlled by the CCAs.

You still need to harden it against EW and possibly microwave weapons going forward. Probably will lag state of the art civilian process for some time.

 

[_Del_]

Ive tried for too long on a weekend to look up what MLP means in this context…. I know Pegasus is getting ABMS as part of CR1 but am i missing something obvious but non-official with “MLP” apologies and thx in advance

Tankers Likely the First Aircraft to Receive ABMS Upgrades | Air & Space Forces Magazine

USAF’s aerial refueling fleet could be the first platform to adopt new technologies developed under the ABMS effort.

www.airandspaceforces.com

Tis a joke.

If the Pegasus new pod or otherwise has JREAP-C terminal (previously demonstrated), then it can use military or commercial networks sending TDL info over internet protocol. Anything connected to the internet on Starshield/Starlink can function as part of the mesh network. Both (or multiple) nodes can send and receive data over IP and share that with any platform/node they can reach otherwise. Those nodes can share with anyone it can reach on the network, and so on. Which means all this sharing can be beyond line of sight even if not everyone has JREAP-C.

This is how someone else and I otherwise butted heads earlier, because the Link-16 data can absolutely be sent over IP networks. And the node does not have to be so enabled by hardware. It can be a standard TDL node. It just needs to connect to a gateway JREAP-C node and then it will be shared like normal with everyone over standard TDL, and/or everyone connected to the network with JREAP-C gear (who then also share like normal).

One JREAP-C equipped CCA can communicate from a dozen CCA drones in it's sight to an off the shelf F-16 if it goes through Starshield (or any other network) to be shared at the other gateway. Everything is a repeater in the Link-16 network. If you can get data over the commercial network to another gateway, say a tanker, there is no reason it cannot share with platforms half a world a way (latency could be an issue until they get more hardware and bandwidth up there and stop using commercial networks).

 

[Josh_TN]

You still need to harden it against EW and possibly microwave weapons going forward. Probably will lag state of the art civilian process for some time.

I am not aware of any directed energy weapon that remotely approaches the envelope of an AIM-120.

 

[Training_Dummy]

My Little Pony. KC-46

LOL thank you. I saw the Vance / MLP stuff in one the searches and thought it can’t be thought.

 

[icyplanetnhc (Steve)]

That says 18% over F119. Nice, and a significant improvement.

But I repeat, even a 25% increase in efficiency over the F119 will not give the range they're talking about in an F-22 sized airframe.

You can also have a considerably higher fuel fraction, both by increasing fuel capacity and reducing weight. Relaxing some of the demanding but rarely used maneuver parameters from the F-22 can save quite a bit of weight. And as is, even the F-22's fuel fraction isn't quite what was originally intended due to cost saving measures during EMD that deactivated certain fuel tanks.

Note that the 18% range improvement for the F-22 refers to the application of AETP/NGAP technology for a Raptor-specific powerplant (i.e. enhanced F119), not the direct application of the NGAP engines (XA102 and XA103) themselves.

 

[F119Doctor]

You can also have a considerably higher fuel fraction, both by increasing fuel capacity and reducing weight. Relaxing some of the demanding but rarely used maneuver parameters from the F-22 can save quite a bit of weight. And as is, even the F-22's fuel fraction isn't quite what was originally intended due to cost saving measures during EMD that deactivated certain fuel tanks.

Note that the 18% range improvement for the F-22 refers to the application of AETP/NGAP technology for a Raptor-specific powerplant (i.e. enhanced F119), not the direct application of the NGAP engines (XA102 and XA103) themselves.

It also might not be a 3-Stream engine. It could just be injection of AETP/NGAP technologies in the F119 architecture, like using the F135 ECU core module, modified to fit the F119 internal interfaces.

 

[Scott Kenny]

Yes, and no . It is software and it will be airframe agnostic defined only by mission set. Airframe capabilities incl. available weapons are only one of the variables. If you can run GPT at home, then most probably you can put such model on the CCA airframe.

Which still takes a big chunk of volume, weight, power, and cooling in the CCA. And that costs money. So does armoring the processors to military standards.

If the Pegasus new pod or otherwise has JREAP-C terminal (previously demonstrated), then it can use military or commercial networks sending TDL info over internet protocol. Anything connected to the internet on Starshield/Starlink can function as part of the mesh network. Both (or multiple) nodes can send and receive data over IP and share that with any platform/node they can reach otherwise. Those nodes can share with anyone it can reach on the network, and so on. Which means all this sharing can be beyond line of sight even if not everyone has JREAP-C.

This is how someone else and I otherwise butted heads earlier, because the Link-16 data can absolutely be sent over IP networks. And the node does not have to be so enabled by hardware. It can be a standard TDL node. It just needs to connect to a gateway JREAP-C node and then it will be shared like normal with everyone over standard TDL, and/or everyone connected to the network with JREAP-C gear (who then also share like normal).

One JREAP-C equipped CCA can communicate from a dozen CCA drones in it's sight to an off the shelf F-16 if it goes through Starshield (or any other network) to be shared at the other gateway. Everything is a repeater in the Link-16 network. If you can get data over the commercial network to another gateway, say a tanker, there is no reason it cannot share with platforms half a world a way (latency could be an issue until they get more hardware and bandwidth up there and stop using commercial networks).

Latency will still be an issue when you're bouncing up to satellites and back down. It's 15 light-seconds from Earth to Moon, 15 seconds back down. It's over half a second to the usual comms satellites and back down, and roughly a second and a half to bounce off 3 satellites to reach the far side of the planet.

500ms latency is no bueno in combat.

 

[ncc81701]

Which still takes a big chunk of volume, weight, power, and cooling in the CCA. And that costs money. So does armoring the processors to military standards.



Latency will still be an issue when you're bouncing up to satellites and back down. It's 15 light-seconds from Earth to Moon, 15 seconds back down. It's over half a second to the usual comms satellites and back down, and roughly a second and a half to bounce off 3 satellites to reach the far side of the planet.

500ms latency is no bueno in combat.

Typical latency over Starlink with a local node (Ground Station for example) is on the order of 25-50ms (can be pushed even lower) . The whole point of starlink and starshield’s LEO constellation is to shorten distance to the satellite and reduce latency.

 

[Scott Kenny]

Something as ubiquitous as LLM AI that has been eating garbage all its life is dangerous and unnecessary. Aircraft AI will be given narrowly defined data that has been tagged and cleaned by humans and used for specific tasks to give automation a facsimile of intuition: recognizing new EW wave forms and countering them with ECM, recognizing new target types or targets that are masking their EO/IR signature with camo or light emitters, and of course the software agent(s) that actually fly the CCAs. There is little reason not to install one on manned NGAD as a fallback; it could handle surprise situations while the aircrew focused on managing a wider network of unmanned aircraft.

Vetted data sets are not cheap.

Carefully vetted data sets are doubly so.

 

[NMaude]

It's 15 light-seconds from Earth to Moon, 15 seconds back down.

1.5 light-seconds not 15 light-seconds,

 

[Josh_TN]

Which still takes a big chunk of volume, weight, power, and cooling in the CCA. And that costs money. So does armoring the processors to military standards.



Latency will still be an issue when you're bouncing up to satellites and back down. It's 15 light-seconds from Earth to Moon, 15 seconds back down. It's over half a second to the usual comms satellites and back down, and roughly a second and a half to bounce off 3 satellites to reach the far side of the planet.

500ms latency is no bueno in combat.

The LEO satellites in orbit by 2028 will have significantly shorter latency and lag time is less significant if the platforms are mostly automated. C&C should be more general - setting behavior and mission, or possibly operating area and target set.

But there would still need to be some kind of graceful degradation mode in case satellite contact was lost.

 

[Josh_TN]

Vetted data sets are not cheap.

Carefully vetted data sets are doubly so.

Never the less they will be made (more likely already in process). The advantages offered are too great to ignore.

ETA: a dataset for a very specific purpose would also be several orders of magnitude smaller than a LLM. For instance an EW software agent largely just needs to know everything on file about friendly and opponent wave forms and how to jam them. Chances are that data is already stored in a highly organized and vetted form.

 

[_Del_]

Latency will still be an issue when you're bouncing up to satellites and back down. It's 15 light-seconds from Earth to Moon, 15 seconds back down. It's over half a second to the usual comms satellites and back down, and roughly a second and a half to bounce off 3 satellites to reach the far side of the planet.

It's about .001 seconds to a LEO satellite like Starlink, plus whatever distance it travels on network. Not the biggest concern.

 

[Kat Tsun]

Since the LLMs are starting to show collapses due to them "eating"/analyzing LLM-created bullshit, I doubt that we will see LLMs on military aircraft.

Not important for parsing battlefield orders versus making Wikipedia 2. Combat LLMs would have their own training data sets.

 

[quellish]

I doubt that we will see LLMs on military aircraft

Large Language Models are statistical models of language that are specific to natural language processing. LLMs do not make decisions or have the ability to reason. There is not a good use case for a natural language interface on a military aircraft.

 

[ubiquitous08]

LLMs are especially bad at dealing with new situations outside their training data so are especially vulnerable to novel tactics.

 

[Josh_TN]

LLMs are not what the USAF is working on. Their flavor of AI is more focused and different with a specialized data set. USAF AIs do not need to answer help desk questions or pass the bar exam.

 

[NeilChapman]

Nope, far from dead. As of early summer it was very much alive. USAF wants to produce small numbers at a time and constantly iterate. 3-6 years for each airframe iteration, with 8-16 year service life.

Each iteration could be made by a different contractor. They want to get away from “winner take all” and use CCA to incubate new contractors.

Rapidly iterate...

Something we can thank SpaceX (Musk) for.

 

[NeilChapman]

A blockade would be imposed the moment the PRC decides to invade Taiwan and attempts to send PLA forces across the Taiwan strait.

What would that look like?

 

[NeilChapman]

CCA bullshit can't replace a plane like the F-22, it is like fighting a jet with piston engine. Musk is saying bullshit, he want to sell is starlink service for the drone fleet , but soon the conflict will be in space, to shot down the CCA fleet shoot the Starlink satellite detonate a nuclear weapon in orbit an bye bye the marvelous drone fleet. There is nothing better than a human in a real fighter , they must look at the Ukraine war carefully with no dominance in the air no win. Musk is a genious for Space but know nothing about military aviation.

A couple of thoughts...

DoD is already purchasing a separate version of the Starlink service. Spacex was the only satellite system that the russians weren't able to jam. It would make sense that the DoD version would be hardened.

Setting off a nuclear weapon in space would be politically catastrophic for the CCP.

Space is really big and SpaceX has a lot of satellites.

 

[NeilChapman]

This is the problem. None of this should be surprising considering DoD has been stating this for years, yet the "news" outlets that are most often read and cited here do not report on them.

Frankly, most of our forum members do a better job of reporting the news on these topics than those sites do. Which... is why... those places... keep using material from the forum.




Hold my beer....


View attachment 751162

How deep is the magazine?

With AI, and iteration for both design and manufacturing techniques, you can have faster, cheaper, and better. This is something else that Musk knows how to do. This is not something that you get when you have an accountant running an aerospace company.

 

[NeilChapman]

What do we need NOW and the next ten-fifteen years? Not what can we deliver in fifteen years time. The catch is if you commit to an extremely expensive and capable NGAD right now, you might be caught holding the bag in fifteen years when the technology delivers even further. The expensive NGAD that is redundant and worse tying up assets you could have used to field the new technologies.

That's where the talk about "re-imagining" and punting on NGAD commitment is coming from, and where the small, fast, programs delivering cheap aircraft with short life cycles generates from. They aren't being completely feckless. We're on the dawn of truly revolutionary technology, and noone wants their balls in the vice because we emptied the treasury for a very fancy and redundant aircraft that missed the revolution.

Yes

I would argue the time frame is 2-5 years. The PRC needs to see an unencumbered US economy growing at 3+ percent per year, rapid progress with CCA, and a complete transformation in munitions production.

It was a full two years before the US economy was able to get on a war footing after 1941. And that was when the country built its own machine tools. The best that can be done today is to convince them that's the direction that they're going.

 

[NeilChapman]

Regardless of AI the manufacturing/logistics aspect remains the same. Capability = cost so the more capable the drone the more expensive they are to build and logistically support. Meaning you won't be able to afford absurd numbers of them in the air. 100,000 dirt-cheap quadcopters or kamikaze drones aren't going to do much of anything when they run out of power and fall into the ocean. The distances needing to be flown, the projected levels of air defense and enemy EW support, that all requires some minimum level of capability that AI isn't going to fundamentally change the balance of. Designs like the MQ-28 and MQ-58 are probably the bare minimum that would be useful in an offensive sense that aren't just a one-way expendable munition. But of course, you need other platforms providing them the much of the sensor data and coordinating with them to perform successful sorties.

I'm not suggesting quadcopters are the answer.

But Taiwan is not very far from the northern Philippines, nor Yonaguni. There are lots of islands in the Ryukyu island chain.

 

[NeilChapman]

Might be surprised on the range for a Valkyrie. It won't lug 2000 lbs that far down range, but if you can by 10 or more of them that far for the price of one X-47B-style UCAV, I'm not sure the math leans in the direction of the more capable platform.

In war, a one way trip makes a lot of sense.

 

[quellish]

Rapidly iterate...

Something we can thank SpaceX (Musk) for.

No, they have been very clear that they are following the model of consumer cell phones.

They want to have a hardware platform that allows them to rapidly solve problems through software, and iterate the hardware as rapidly as feasible.

 

[NeilChapman]

We can basically fix this issue until like 2050+ ish without increasing taxes on anyone making below 150k a year. Also optimizing/reforming the big three SS, medicare/medicaid would help a bunch too, there's a lot that can be done.

From what I remember Trump/GOP in 2016 proposed a pretty solid tax reform idea but dropped it in the final bill, they could bring it back again this time and that would actually go a long way and allow way more openings for

The big three don't affect the budget. They were already paid for by our taxes we used to pay directly into them. The problem is congress likes to loot them to give tax cuts to the wealthy and the wealthy don't want to pay it back. They can keep their greedy paws off of my earned benefits that I already paid for.

This is way off topic.

Let's wait and see what they're able to do.

 

[NeilChapman]

Worth noting that with the widespread adoption of the F-35's worldwide, designers are tailoring new weapons within this constraints. Many/most munitions in the next ten-fifteen years are going to be designed around that size-constraint. In spite of that they will probably increase, not decrease, the ability to deliver whatever effect downrange be it AAM, ARM, cruise-missile, ASh, EW, etc

The next fielded combat aircraft won't have to size themselves to carry 4 AMRAAM internally to deliver that equivalent effect downrange, or perhaps it is sized to deliver that effectx10 or more with a smaller bay than the F-35, as one example.

Let's hope that comes with a cost reduction as well. These munitions are ridiculously expensive. Has anyone determined what they currently cost per kilo?

 

[NeilChapman]

No, they have been very clear that they are following the model of consumer cell phones.

They want to have a hardware platform that allows them to rapidly solve problems through software, and iterate the hardware as rapidly as feasible.

Well, that's necessary if you have an airframe requirement of eight thousand hours. Not as necessary if you optimize for production speed and lethality.

 

[NeilChapman]

That's not very likely in regards to the PRC as it is dealing with intractable demographic and economic issues that are getting worse.

I agree. The problem with autocratic regimes is that as they get more unstable, they create problems to take the focus off their own shortcomings. The United States must move quickly to achieve it's objective of maintaining peace in the region.

Getting its fiscal house in order and modifying how DOD does business seems to be part of the plan of the incoming administration. I, for one, hope they're successful. Magazines need to be refilled. Airframes and munitions need to be manufactured and fielded orders of magnitude faster.