Jump to content
Sturgeon's House

Using computergames and simulators to collect data and evolve AFV designs.


Recommended Posts

I am not sure if this should be posted in the software section or here. So I chose this section since I found it the most relevant. 


But on point: Collecting data. 


AFV designs are based on data right? These are usually battle reports, which some times may be misleading, but has helped shape AFV design. 


For example:

*The frontal 30 degree section of a tank will take 50% of the hits. Thereby, lets up armor the front 30 degree to resist the threat the vehicle is designated to stop, which in the case of a MBT, is AT-rounds.
*The turret is going to be hit the most. That's why we are going to armor it the most. 
*The UFP and the Lower section of the turret is hit the most, therefor let's have them as the most armored parts. 


So I was wondering, does developers of games like Steel Beasts collect information on the areas the AFV has been hit, the range the shot was fired from, and the type of ammunition used?

This information could be given/sold to AFV developers for them to further enhance their designs. 


The data could be separated into two groups:



The military group would be data from actual AFV crews training in a simulator, while the Civilian group would be data collected to anyone playing the game outside training, where skill, proper training, or seriousness may lack. A example here is how WT players will often pick the KV-2 for shits and giggles simply because it has a giant cannon, disregarding how useful this actually is. 


Further, the two groups could be separated by country, as well as if they are conscripts or not. 


To increase the variety of data, the civilian version at least could include a "deathmatch mode" and the usual game modes where battles are often and intense. 



A more experimental way could also be used. Using neural networks, we could have a AI evolving to adept perfectly to the opposition. This could be done via Steel Beasts. You simply load up a scenario, and hook either one AI to all of the AFVs or even one AI for each crewmember.  After several "evolutions" the AIs would have figured out the perfect way to take out the enemy opposition.   This would also be a excellent way of finding bugs in the game, since neural networks will use any advantage they can. 


Hook this up with a AFV building program, with limiters of course, the neural network could also develop the perfect design to counter the threat. Repeat this over several, if not all scenarios and it could come up with a lot of interesting designs. 


But note, I am not saying we should let a neural network design a AFV all by itself, it should rather serve as a guideline or a reference for what AFV designers can improve.


Here is a example of neural networks:




Link to comment
Share on other sites

You're going to find longstanding parameters that have been a part of "improving" armor since it's conception.


 1-See first.

 2a-shoot first.

 2b- don't miss

 2c- kill the target with as few shots as possible.


 3- Be able to shrug off sufficient incoming fire in order to accomplish most of the above.


 4- Be able to get where the bad is, or where it is not, with reasonable speed and reliability.

Link to comment
Share on other sites

In terms of optimising your system from the ground up, this has certainly been tried in a number of contexts and the results are often... less than useful.


A quick example: it has been a common trend for a long time now that gamed-out designs for warships (or fantasy spaceships) tend towards large numbers of identical, disposable, defenceless ships packing as much firepower as possible. The reasoning is simple and elegant: Lanchester's laws apply very well in the open spaces of the ocean (and, of course, space) and damage to any ship invariably takes it out of the fight. So the best way to make a winning fleet is to spread your weapons very thinly across a number of platforms, thus turning the loss of each into a linear loss in capability instead of an exponential one.


But in real life these sorts of approaches almost never work. Why? Partly because of all the complexities that the model cannot properly capture: interaction effects between units, non-linear gains in efficiency of sensors, non-linear losses in command-and-control as numbers increase. But mostly because of factors outside of the simulation entirely: economics, politics, social factors and so on. The end results of these infinitely more complex processes, run through the game theory of thousands of intelligent actors, is ultimately the reason why fleets tend towards a mix of capital ships and smaller platforms rather than a uniform sea of missile boats.


Finally, there is the every-present issue of GIGO and un-examined wrongness in the underlying assumptions which power the model.

Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.


  • Create New...