Will development of good AI routines be beneficial for developing an UAV?

[Feuerfalke]

Different goals and different problems.

In contrast to the real world, a plane in a game flies in a virtual environment. Everything put into the game is calculated and therefore for terms of simplicity transfered into the AI-routines as necessary. No need to adopt to unknown territory or obstacles.
A simple example: When Flak fires at an AI-plane, it can be noted by simply setting a variable, telling the AI it is under fire. Another example is the spider-sense in IL2 that warns an AI pilot as soon as a fighter is in a given distance. In the real world, the AI would have to be able to interpret sensors to even identify such a threat and out of the masses of information gained from the sensors and that is a difficult task.

On the virtual battlefield, the AI also tries to incorporate human behavior. SoW will even incorporate fear, fatigue, aggressiveness, skill, etc. That's exactly the points you want to eliminate for autonomous RL planes.

[Bobb4]

Originally Posted by Feuerfalke

Different goals and different problems.

In contrast to the real world, a plane in a game flies in a virtual environment. Everything put into the game is calculated and therefore for terms of simplicity transfered into the AI-routines as necessary. No need to adopt to unknown territory or obstacles.
A simple example: When Flak fires at an AI-plane, it can be noted by simply setting a variable, telling the AI it is under fire. Another example is the spider-sense in IL2 that warns an AI pilot as soon as a fighter is in a given distance. In the real world, the AI would have to be able to interpret sensors to even identify such a threat and out of the masses of information gained from the sensors and that is a difficult task.

On the virtual battlefield, the AI also tries to incorporate human behavior. SoW will even incorporate fear, fatigue, aggressiveness, skill, etc. That's exactly the points you want to eliminate for autonomous RL planes.

You know this site is going down hill when we move from talking IL2/SOW to UAV's

[Feuerfalke]

Lol

[Skarphol]

Weeelll, there isn't too much to talk about on this site anyway, reading the same repeated questions over and over again is boring and futile, and besides; there is a lot of people with insight into aviation hanging around here, so I gave it a go.

I guess you guys are right about different problems and different solutions with UAV development and AI developent.

I'm really curious about the AI in BOB. I think IL-2 must have had some of the better AI behaviour in a flightsim, but I haven't spent much time on other flightsims, though. Just som AEW and MSCFS..

Skarphol

[Igo kyu]

Originally Posted by Skarphol

I'm really curious about the AI in BOB.

Me too.

I think IL-2 must have had some of the better AI behaviour in a flightsim, but I haven't spent much time on other flightsims, though. Just som AEW and MSCFS..

Skarphol

Maybe, but they are really poor at turn fights. In a turn fight, the guy behind drives, the guy in front keeps turning until the guy behind gives up. The Il*2 AI don't do that, if they are in front they change turn direction straight across the following planes fire. The guy in front has an option between blacking out, and dying, the guy behind has an option between blacking out and losing the chase, so the guy in front can turn harder.

[zapatista]

Originally Posted by Skarphol

Weeelll, there isn't too much to talk about on this site anyway, reading the same repeated questions over and over again is boring and futile, and besides; there is a lot of people with insight into aviation hanging around here, so I gave it a go.l

dont pay any attention to the moaners, its an interesting topic

afaik the artificial intelligence routines used in autonomous unmanned vehicles are based on neural networks first studied in simple biological organisms/creatures, and trying to get an engineered robot to even replicate some of those basic behaviors is extremely complex

if you simply need a drone like that to map a certain area, it is not to difficult. the same principle is already used in agriculture where you can have farm equipment (like tractors, harvesters etc..) cover a particular field while guided by gps, and the equipment is fully controlled by an onboard computer, it is being used like that and does work well

the problem that arises with a drone having to move in 3 dimensional space (not that much harder) and that it has to analyse and "understand" what it sees and then respond to unpredictable events is the hard bit. to determine what events are a trigger is part of the problem, but the processes involved in the human brain assessing what it sees in similar situations is incredibly complex, and hard to copy effectively with artificial intelligence routines so far