Aircraft-AI: Why can't full-fidelity units be AI-controlled, they should ... once

[Worrazen]

Hi

There was and perhaps still is that conception regarding full-fidelity aircraft not being able to be AI-controlled because mainly it would take so much CPU horsepower that you wouldn't be able to play ... hmmm, perhaps so, perhaps not quite...

How technically true is at all anyway? Is it just a DCS thing and would need an engine upgrade?

Either way, it really should be attempted to try to make a special full-physics AI units, albeit not having them have to control a full-fidelity cockpit, but the full physics and flight modelling, and other things.

I think the AI's right now don't work by emulating a human using input controls to control the aircraft, but perhaps in the future it kinda should, to simulate all the things a real pilot would deal with in a cockpit, while at least there should be pilot sight simulation on top of what we have now. AI's are most likely cheating in this case now by being able to see through all the % of their cockpit glass at all times, probably it's not even just glass and they can see through their cockpit nose into the forward direction, hopefully the sight cone is circular/rectangle tilted upward.

So I think the AI code is interwined with the do-this-do-that interfacing with the specific physics model logic, and doing a full-fidelity AI would likely require basically making AI version for it fron scratch, and simulating an actual pilot and the inputs would I think be significantly harder and perhaps not even optimal for how to do an AI in programming efficiently, this is just speculation but I think this would be a better reason and real limitation of why there's no full fidelity AI's.

Why do I sound like I'm dismissing the CPU processing requirements ... well because there's a little fact that at least to my knowledge nobody has think of ..., err rather I should say nobody has pointed out, and  that is, what prevents us from just setting a limit on one single full-physics AI at a time, in a single mission, or perhaps 2-3, depending on the system's hardware, but just one would make a big difference for where it would be most benefitial, dogfights.

In pure dogfights, you don't have 500 full-physics AI's running around, the hardware should be able to deal with just one (1), if we somehow make it over the other possible limitations I mentioned.

There! Solved???

Edited June 2, 2023 by Worrazen

[twistking]

ED is already working on an advanced FM for AI. It will not be based on the full fidelity FM, because - honestly - that would be crazy.

Current AI FM can barely called a flight model by today's standards. You can make it a million times better and still only use a fraction of what a full fidelity FM needs for processing power.

I'm not sure to what extend they want to also simulate AI factors like perception, workload, stress... i'd already be happy when AI can't see through clouds anymore...

Edited June 3, 2023 by twistking

[Kang]

I don't see much point in simulating all the 'inner workings' of solely AI units, really. There is no harm at all in using a whole lot of shortcuts that can simulate that perfectly well with much less strain on the computing resources. The trouble isn't that the approach of tabulated flight models was inherently bad, it's more that some AI units just have utterly ridiculous values in those tables.

[Exorcet]

20 hours ago, Worrazen said:

How technically true is at all anyway? Is it just a DCS thing and would need an engine upgrade?

Not only will it be more CPU intensive, the AI has to be coded to intelligently fly. You'd increase CPU workload and ED's literal workload by making them code 1:1 AI that flies as the player does. It also wouldn't work because AI just isn't that smart, so it would probably fly poorly.

There is zero reason to do what you're asking because 99% convincing AI can be done with less demands on the hardware and the developers. Just let ED finish their AI changes and GFM and we should be good.