VR Designs

After improving the Soviet artificial intelligence (AI) I am now turning my attention to the German AI. In this AI log we’ll take a look at how it is doing right now.

The German AI is basically not doing so bad and in the end managed to get one of the three major objectives. It is already in quite good shape, especially considering that is facing more Soviet units than a human player would be fielding. However as discussed in the video above I feel it is missing something. I’d like to call it a certain je-ne-sais-quoi… Its a bit of that edge that made commanders like Guderian and Manstein so effective.

But what is that edge? If I drill things down I think the issue with the German AI is it needs to take calculated risks at the right spots. The spots are where their four ‘Panzergruppen’ are and the risks are that these ‘Panzergruppen’ need to advance more while leaving some Soviet units in their rear. Furthermore the german Infantry divisions need to accept that sometimes higher casualties can be acceptable if it leads to capturing essential terrain or opening up the road for those ‘Panzergruppen’.

Sounds like a tough lesson to teach :) When I have a new German AI ‘Guderian’ version running I’ll report back here if I succeeded.

Best wishes,
Vic

Remember the previous Soviet artificial intelligence (AI) losing cohesion and Moscow quite early? After some big changes it has really acquired more fluidity and backbone. Not perfect yet, but much better!

The soviet AI is basically winning the game now. Admittedly against the German AI but soon we’ll ship the next BETA version to the testers and see how this improved Soviet AI will fare against some brighter humans.

Next up I am going to analyse and improve the German AI because it could use a bit a of extra punch and speed in some places and a bit of caution in others. Expect a new AI development log on the German AI next week!

Beta testing is well under way and one on my big tasks is to get the artificial intelligence (AI) in to tip-top shape. I thought it could be nice if to take the readers along for the ride in some regular AI dev logs.

Current results from the beta group playing versus the Soviet AI were positive, but some more experienced players had a bit more mixed results. Some of them managed to crush the Soviet AI just a bit to easily.

Let me show you a video of an AI-to-AI game which mimicks an experienced human German player versus a regular AI:

The way I code and improve the artificial intelligence is by making analysis of its current behaviour and then designing a plan to improve its behaviour.

Analysis of current behaviour:
-The Soviet AI in the North is losing to much troops before the Germans reach the defensible swamps and forests around Leningrad. Because of this the AI does not manage to keep the line later on.
-The Soviet AI in the Center almost manages to stop the Germans, but it might have made its stance just 2 or 3 turns to early. Furthermore it seems to hold the southern part of the center front (Bryansk) quite well, while giving to much terrain on the key highway Smolensk-Moscow. Eventually this leads to an early fall of Moscow.
-The Soviet AI in the South gets a lot of armies in this playthrough but it doesn’t manage to hold the Dnjepr-line as well as it should.

Plan for improving the AI:

Considering everything the Soviet AI needs to:
1. On a few exceptions (Riga, Odessa, Pripjet) retreat as many troops intact to a defensible line. Which is roughly the Leningrad forests – Smolensk – Dnjepr line.
2. Better understand (like in the center) where the enemies main drive is and move the brunt of it forces there. Especially if it concerns a direct attack on Moscow.
3. Hold the Dnjepr line at all costs. This means holding it to the hex and making aggressive counter attacks against bridgeheads while they are still small.

Cannot make this log to long, since I have to get back to coding :)
But I hope you join me again in my next AI log to see if I got the Soviet AI to toughened up as intended.

Best wishes,
Vic