Jump to content

Petrovich "AI" or Not?


MAD-MM
 Share

Recommended Posts

After looking the ATGM Video dont see much like "AI". Looks more like a "Interface" to control the foward Cockpit? What does Petrovich when the Player doesnt know where the enemy Tanks are placed? Nothing when he is not in search Modus and the Player dont know where to look at?

 

Under AI i understand artifical more responsive self Acting Petrovich with Spotting and call out for Targest and Engaging the most dangerous Target in the List then a UI where i am tell him what to do and when?

 

It work ofcourse but i this the future AI?

 

 

Once you have tasted Flight, you will forever walk the Earth with your Eyes turned Skyward.

 

[sIGPIC][/sIGPIC]

9./JG27

Link to comment
Share on other sites

This is just very early access. At least we will have the functionality of using ATGMs. It's quite intuitive, but I agree that Petrovich should be able to spot targets by himself and give some basic callouts. It's early days yet, I have asked a very similar question in another thread, hopefully it's coming at some point in the future. 

  • Like 3

Specs: Win10, i9-9900KF@5Ghz, 32GB DDR4 RAM 3200XMP, 1 TB M2 NVMe SSD, KFA2 RTX3090, VR G2 Headset, Warthog Throttle+Saitek Pedals+MSFFB2  Joystick. 

Link to comment
Share on other sites

Did you read the quikstart?
If I under stood correctly, it will give you lists of targets, and you can order him to rescan, and the target computer will show you where to turn for him to b he able to track target, or something like that.
I'm really looking forward to try it out!

Sent from my MAR-LX1A using Tapatalk

Link to comment
Share on other sites

If Petrovich can spot / call out targets that would be great.  But other than that, I really don't think we need a lot of fancy pseudo-AI functionality in the Hind. I think a streamlined UI that gives us simplified control of the other cockpit functions will do just fine.

 


Edited by cow_art
  • Like 5
Link to comment
Share on other sites

2 minutes ago, cow_art said:

If Petrovich can spot / call out targets that would be great.  But other than that, I really don't think we need any fancy pseudo-AI functionality in the Hind. I think a streamlined UI that gives us simplified control of the other cockpit functions will do just fine.

 

I agree personally I am not really interested in the jester thing for the hind. 

  • Like 1
Link to comment
Share on other sites

I'm more concerned about Petrovich becoming super human. i.e. being able to

- instantly detect and classify all units in an area regardless of LoS, smoke or lighting conditions

- never miss a shot even while maneuvering (withing limits)

- always fire at the perfect range to hit

- do other stuff not shown in video where "AI" can easily cheat (e.g. calling out SAMs or spotting units).

 

I really don't want him to perform better than a somewhat competent human player.


Edited by Blackeye
  • Like 5
Link to comment
Share on other sites

11 minutes ago, Blackeye said:

I'm more concerned about Petrovich becoming super human. i.e. being able to

- instantly detect and classify all units in an area regardless of LoS, smoke or lighting conditions

- never miss a shot even while maneuvering (withing limits)

- always fire at the perfect range to hit

- do other stuff not shown in video where "AI" can easily cheat (e.g. calling out SAMs or spotting units).

 

I really don't want him to perform better than a somewhat competent human player.

 

Alexander has clearly stated before that Petrovitch will act as a human operator which means with a percentage of errors and missed fires...😉

  • Like 2
Link to comment
Share on other sites

17 minutes ago, Blackeye said:

I'm more concerned about Petrovich becoming super human. i.e. being able to

- instantly detect and classify all units in an area regardless of LoS, smoke or lighting conditions

- never miss a shot even while maneuvering (withing limits)

- always fire at the perfect range to hit

- do other stuff not shown in video where "AI" can easily cheat (e.g. calling out SAMs or spotting units).

 

I really don't want him to perform better than a somewhat competent human player.

 

 

That's right! I want a real AI that

  • cracks jokes about, and references the most common Stanislaw Lem novels
  • smokes cigarettes so neither can see anything in- or outside even in clear weather
  • Drinks heavily and then ejects when he sees a scary face in the rearview mirror
  • Sings loudly, but surprisingly on-pitch, an old, sad russian hymn that breaks your heart but unfortunately drowns out the radio

If Pet can hit the side of a barn, heck, that's bonus.

 


Edited by cfrag
  • Like 10
  • Thanks 2
Link to comment
Share on other sites

2 hours ago, Razorback said:

Alexander has clearly stated before that Petrovitch will act as a human operator which means with a percentage of errors and missed fires...😉

 

Couldn't find the statement but I feel a percentage dice roll isn't really the same as proper limitations, e g. a fixed  5% (or whatever number) chance to miss a shot regardless of how violent you maneuver is very different from a human having to compensate for that.

 

I don't know how detailed or well ED can/will implement those limits but I hope it's going to factor this stuff in rather than go for the all-seeing all-knowing all-capable AI with a (simple) failure chance on top.

 

Edit: Found something

Quote

To make a lifelike AI, we did an observation model based on academic study done. It takes into account contrast and angular size to figure out the probability of this object being spotted. Between that there’s a priority dependency on experience, and we’ve added that too into the spotting model. The AI won’t see everything like the AI on other modules, and will be a lot closer to humans, same for the scanning patterns. In other words, they’ll pause to look at a certain area before proceeding. That’s already working, but we’re still refining it.

 

Sounds promising.


Edited by Blackeye
Link to comment
Share on other sites

49 minutes ago, Czechnology said:

i imagine he's going to have a level of spotting, but frankly i don't want him to have a crazy high level of spotting. I don't want to lean on the AI to find targets for me.


Why not? I would try to burden the AI as much as possible (if he can handle it). Let Petrovitch do his half of the job, It is a two-seat helicopter and the less I need to do the better 😎
      

  • Like 4
Link to comment
Share on other sites

 

I do hope it'll eventually automatically call out the type of target, bearing and approximate range.

 

It'd also be nice if it occasionally made mistakes to "Tank 320 degree approximately three kilometres" [...] "correction tank 320 degrees is IFV"

It'd add a lot to immersion.

 

1 hour ago, Czechnology said:

i imagine he's going to have a level of spotting, but frankly i don't want him to have a crazy high level of spotting. I don't want to lean on the AI to find targets for me.

 

If I'm carrying all of the extra weight for a second cockpit and crew-member - I'd like to actually benefit a bit from it. When not guiding missiles the primary role of the forward crew-member is as an observer... so I think we differ on this.

 

 

3 minutes ago, Schmidtfire said:

Why not? I would try to burden the AI as much as possible (if he can handle it). Let Petrovitch do his half of the job, It is a two-seat helicopter and the less I need to do the better 😎

 

The way I see it - it doesn't make sense for people to constantly complain about a lack of a missile warning system for aircraft like the Mi-8, but to reject having AI crew take on that role... it is realistic... and they've built in realistic delays to spotting.


Edited by Avimimus
  • Like 3
Link to comment
Share on other sites

I think we need to remember that the co-pilot/gunner is using an optical device with x10 magnification(what is the magnification?). The target acquisition is solely reliant on the MK1 eyeball. Moving targets will be easier to see due to dust, reflection and the very fact the eye has a remarkable ability at detecting a moving object. When things become stationary, it becomes extremely difficult to see targets, especially if it’s parked in the shadows, tree line, weather etc. I think Petrocitch’s ability should take this into consideration. 
 

What I would like to see is the ability to observe an area of possible armour locality and observe that for a period of time. For those of you that have no experience being around armoured vehicles (in real life lol), especially MBT’s of that era, is when they first start up, there is a plume of blue/grey smoke plumes that floats up, giving away their location. This is something we would do on excessive in Germany in the 80’s and 90’s, observe potential enemy positions and spot the dozen or so rising smoke plumes coming out of the forest. Then call in a simulated air/artillery strike. 

  • Like 1
Link to comment
Share on other sites

The major thing I noticed about the Petrovich pilot AI (in the first video showing it - the ATGM video) was that it was oscillating the helicopter in pitch quite a bit. I found this a bit nauseating.

 

What do you think - would it be better for Petrovich to be less accurate in speed and altitude keeping in order to be more stable in attitude (note: attitude not altitude)? Would that be more realistic?

Link to comment
Share on other sites

15 hours ago, Blackeye said:

I'm more concerned about Petrovich becoming super human. i.e. being able to

- instantly detect and classify all units in an area regardless of LoS, smoke or lighting conditions

- never miss a shot even while maneuvering (withing limits)

- always fire at the perfect range to hit

- do other stuff not shown in video where "AI" can easily cheat (e.g. calling out SAMs or spotting units).

 

I really don't want him to perform better than a somewhat competent human player.

 

 

No one want's Petrovich to be superhuman. Currently he is much, much inferior to a human co-pilot. You have to tell him where to look and shoot, or you have to tell him where to fly. Basically he is little more than either a targeting pod, or a basic autopilot mode, depending on which seat you're giving him commands from. He is getting some voiceover lines, which is nice, but I would argue that for him to be a proper co-pilot he should at least be able to call out targets that he can spot and obviously some incoming threats, and wait for it: If he can see them. 


Edited by Lurker
  • Like 2

Specs: Win10, i9-9900KF@5Ghz, 32GB DDR4 RAM 3200XMP, 1 TB M2 NVMe SSD, KFA2 RTX3090, VR G2 Headset, Warthog Throttle+Saitek Pedals+MSFFB2  Joystick. 

Link to comment
Share on other sites

AI sounds fancy, but usually it's just a lot of if-else statements. 

In the case of Petrovich the most important function is "does he hit or miss witht he shturm".

 

I think you could quite easily mimic human imperfections, if you are able to put some numbers on them.

Put a number on the "steadiness" of the flying, and you can feed that number to an RNG and make Petrovich's hit-chance sort of a diceroll.

Completely simulating his movements of the targeting sight would be more complicated and would lead to almost the same effect. 

 

Spotting may be a bit more tricky, as this is already a problem for static units in the game with their godlike spotting abilities. 

But also here you could use an RNG, that takes into account range and lighting conditions; rather than giving the list of all units in a 5 km radius.

So in principle you could have a spotting chance of 100% when LoS is there, minus a penalty for range, a penalty for lighting and a penalty for steadyness of flying. 

 

  • Like 1
Link to comment
Share on other sites

24 minutes ago, Lurker said:

No one want's Petrovich to be superhuman.

 

Of course not - yet a lot of the other AI in DCS certainly has superhuman feats. Whether it is a physics defying flight model or ground forces with perfect SA.

 

Quote

Currently he is much, much inferior to a human co-pilot. You have to tell him where to look and shoot, or you have to tell him where to fly. Basically he is little more than either a targeting pod, or a basic autopilot mode, depending on which seat you're giving him commands from. He is getting some voiceover lines, which is nice, but I would argue that for him to be a proper co-pilot he should at least be able to call out targets that he can spot and obviously some incoming threats, and wait for it: If he can see them. 

 

I think spotting/calling is planned.

 

And I agree that making (tactical) decisions all by himself would be great and actually push into "AI" territory. However those things are quite easy to make up for as pilot, so being inferior in that regard is not a big issue and expected to some degree.

The "menial" tasks OTOH like spotting and shooting is something that humans struggle with and where the AI is cheating to some degree, which can easily lead to a situation where a human pilot with a cheating AI is superior to a team of two humans.

  • Like 1
Link to comment
Share on other sites

3 hours ago, malcheus said:

AI sounds fancy, but usually it's just a lot of if-else statements. 

In the case of Petrovich the most important function is "does he hit or miss witht he shturm".

 

I think you could quite easily mimic human imperfections, if you are able to put some numbers on them.

Put a number on the "steadiness" of the flying, and you can feed that number to an RNG and make Petrovich's hit-chance sort of a diceroll.

 

 

I disagree with this. Let me explain it as simply as I can. Would you like it if your Targeting pod arbitrarily missed every once in a while based on a % chance, simply because RNG? I agree that during hard maneuvers it should be very difficult to keep the sighting system "locked" on a target, but as far as I know that is already in the sim and Petrovich uses the system to it's limits, just like a human would. Just because we haven't seen it doesn't mean that Petrovich can't miss, I'm sure he can and will miss based on how well we fly and hard we make it for him to keep sighting the target correctly. 

  • Like 2

Specs: Win10, i9-9900KF@5Ghz, 32GB DDR4 RAM 3200XMP, 1 TB M2 NVMe SSD, KFA2 RTX3090, VR G2 Headset, Warthog Throttle+Saitek Pedals+MSFFB2  Joystick. 

Link to comment
Share on other sites

56 minutes ago, Lurker said:

 

I disagree with this. Let me explain it as simply as I can. Would you like it if your Targeting pod arbitrarily missed every once in a while based on a % chance, simply because RNG? I agree that during hard maneuvers it should be very difficult to keep the sighting system "locked" on a target, but as far as I know that is already in the sim and Petrovich uses the system to it's limits, just like a human would. Just because we haven't seen it doesn't mean that Petrovich can't miss, I'm sure he can and will miss based on how well we fly and hard we make it for him to keep sighting the target correctly. 

I wouldn’t compare a targeting pod to some ones eye ball and what is basically a pair of binoculars integrated to a missile system from 40- years ago. There will be a miss rate based on a moving helicopter, they weren’t very stable platforms back then and a MBT trying to break the world land speed record across uneven terrain. 

Link to comment
Share on other sites

18 minutes ago, Rabies said:

I wouldn’t compare a targeting pod to some ones eye ball and what is basically a pair of binoculars integrated to a missile system from 40- years ago. There will be a miss rate based on a moving helicopter, they weren’t very stable platforms back then and a MBT trying to break the world land speed record across uneven terrain. 

 

You completely missed the point I'm trying to make.

 

Also, a "pair of binocluars" integrated to a missle system...." Yeah, it's so horrible that for some reason they kept using it for 40 years! Please, provide some evidence of how bad this piece of hardware is, go ahead, We're all ears. 


Edited by Lurker

Specs: Win10, i9-9900KF@5Ghz, 32GB DDR4 RAM 3200XMP, 1 TB M2 NVMe SSD, KFA2 RTX3090, VR G2 Headset, Warthog Throttle+Saitek Pedals+MSFFB2  Joystick. 

Link to comment
Share on other sites

5 hours ago, Blackeye said:

 

Of course not - yet a lot of the other AI in DCS certainly has superhuman feats. Whether it is a physics defying flight model or ground forces with perfect SA.

 

 

I think spotting/calling is planned.

 

And I agree that making (tactical) decisions all by himself would be great and actually push into "AI" territory. However those things are quite easy to make up for as pilot, so being inferior in that regard is not a big issue and expected to some degree.

The "menial" tasks OTOH like spotting and shooting is something that humans struggle with and where the AI is cheating to some degree, which can easily lead to a situation where a human pilot with a cheating AI is superior to a team of two humans.

They should not give it the Name "AI" in the first glance, the Hind is not released so far but we will see how good the Petrovitch Scan Modus can detected stuff behind Trees and Obstacels that would not be seen by a Player because they partelly Coverd.

Think implement good AI that's is plausible is a hugh Task and Money, that would be something like this:

 

° Target is standing still ~ detection -30% because tracking a non moving Targets is difficult for human Eye

° Target is partelly Coverd and Camo Skin, Spotting -80%

° Target is producing Dust while Driving +30 %  (Dust brings Shells)

 

Some other Infantry Fighting Simulator tryed this allready, flying there with the AI is absolut annoying because either the AI seem's to ignore Stuff that can be seen by the Player, or spotting Targets that will be not seen by Human ever because standing in the middle of a Forrest.

 

That think will be positive suprised that will ever come so far in DCS for working good AI, but higly doubt this Petrovich will much evolve  as usefull AI or not only a UI.

 

Once you have tasted Flight, you will forever walk the Earth with your Eyes turned Skyward.

 

[sIGPIC][/sIGPIC]

9./JG27

Link to comment
Share on other sites

2 hours ago, Lurker said:

 

I disagree with this. Let me explain it as simply as I can. Would you like it if your Targeting pod arbitrarily missed every once in a while based on a % chance, simply because RNG? I agree that during hard maneuvers it should be very difficult to keep the sighting system "locked" on a target, but as far as I know that is already in the sim and Petrovich uses the system to it's limits, just like a human would. Just because we haven't seen it doesn't mean that Petrovich can't miss, I'm sure he can and will miss based on how well we fly and hard we make it for him to keep sighting the target correctly. 

 

If a real life targeting pod has 100% hit rates, then I want the targeting pod to have 100% hit rates in the game.

And weapon operators, I want to perform as closely as they can to human operators.

I have no idea how well the real Petrovich's performed in the field, but I know that the existing AI's in DCS vastly outperform humans in terms of observing and aiming.

 

As for operating within the limits of the sight, I think you can get a god-like performance while still operating the sight within its limits if you let an AI control it. 

Basically we have only two relevant axes, pitch and yaw, that the operator has to compensate for.

Humans will take some time to detect movement on an axis, and then always over- or undercompensate to some degree. Not due to the limits of the sight, but due to the limits of human cognitive and motor function. 

An AI can always compensate perfectly within a fraction of a second (as you can see in targeting pods).

 

Again, I don't know how real-life operators performed, so perhaps the human limitations are negligible, but I don't want a superman in the front seat that always hits, no matter how rough we are flying. 

Link to comment
Share on other sites

Russian equipment throughout the Cold War was of poor quality and lacked the technology when compared to NATO. This USSR doctrine of mass production resulted surplus material and the prolonged use of that outdated material for a considerable period of time, post collapse of the East German border and the USSR. It’s only been in the last 20-years or so that Russia has invested heavily in upgrading its military capability and technology to be on parr with Western nations. 
 

The limitations of the-optical sighting system of the gazelle and Lynx helicopters during the Cold War era was of concern and was often discussed in the crew rooms. Concerns being, the ability to acquire targets through a simple optical device in a combat environment of an approaching USSR armoured division. The one thing the USSR was fairly good at was their mass of air defence systems and the Lynx had to try and get with reasonable proximity to optically identify and engage these defences prior to engaging the armour. This feat was challenging as it involved the aircraft being within detectable and engagement range of the air defence system such as the ZSU 24-4. 
 

During my time in a British Army aviation unit during the Cold War, I had the good fortune of being able to visit a large aircraft hangar in Germany, that allegedly “didn’t exist” at that time, and had hands on opportunity to climb into multiple USSR aircraft types, air defence systems and armoured vehicles. The visit reinforced our knowledge at that time that USSR technology and aircraft design and weapon delivery systems was decades behind and NATO’s only true concern was the shear number of USSR MBT (1000’s) that faced us in Europe. 

 

I don’t know if you fly, but if you do, take a pair of x10 or x12 binoculars with you, place a dot in the centre and try to maintain that dot on a vehicle that’s 3 km away during forward flight. And maintain it there for 45-60 seconds, this will take into account target acquisition, identification and missile flight time. 

  • Like 1
Link to comment
Share on other sites

2 hours ago, Lurker said:

I disagree with this. Let me explain it as simply as I can. Would you like it if your Targeting pod arbitrarily missed every once in a while based on a % chance, simply because RNG?

 

Well not RNG... but RNG+Contrast? Yes!

 

Honestly, there should be some situations where the Shkval's automatic tracking system fails (e.g. due to intervening obstacles, low contrast, lighting conditions) and one has to manually correct to keep the cross-hairs over the target. It is unrealistic for targeting pods to be perfectly reliable! I want to sometimes have to go manual!

Link to comment
Share on other sites

 Share

  • Recently Browsing   0 members

    • No registered users viewing this page.
×
×
  • Create New...