AI failings in 1.3 beta (and earlier)

Bug reports for the different beta versions of transcendence.
Post Reply
User avatar
Atarlost
Fleet Admiral
Fleet Admiral
Posts: 2391
Joined: Tue Aug 26, 2008 12:02 am

AI failings in 1.3 beta (and earlier)

Post by Atarlost » Sat May 31, 2014 3:00 pm

The AI has been having trouble since the introduction of the new maneuver system.

It's leading by too much. The way the AI is sloppy about accuracy the default and flyby AIs and possibly all AIs except chase naturally shoot ahead of the target because of their circling. They focus too much on circling and do not properly line up their shots.

Last night pm912 discovered that it is unable to account for the difference between rotationAccel and maxRotationRate, and gets worse with more facings.

pm912 and I also believe that the default AI tries to point its nose at its target while trying to fire its weapon ahead of the target.

In addition to the obvious issue with rotationAccel I believe the default and standoff AIs need to drop into the working chase AI temporarily when they're having trouble hitting and the flyby AI needs to just strafe in and out of range, preferably turning so as not to appear to be attempting to ram and not try to circle. The attempts to circle very negatively impact performance with slow firing or counter based fixed axis weapons.
Literally is the new Figuratively

george moromisato
Developer
Developer
Posts: 2849
Joined: Thu Jul 24, 2003 9:53 pm
Contact:

Re: AI failings in 1.3 beta (and earlier)

Post by george moromisato » Sat May 31, 2014 4:16 pm

I've definitely noticed some of this too, but I can't quite pin it down. Any additional info you can provide would be helpful.

Either way, I'll work on it some more.

PM
Fleet Admiral
Fleet Admiral
Posts: 2202
Joined: Wed Sep 01, 2010 12:54 am

Re: AI failings in 1.3 beta (and earlier)

Post by PM » Sat May 31, 2014 7:02 pm

AI assumes max and accel Rotation are the same, and can stop rotating instantly as before in previous versions. A ship will rotate toward its ideal facing towards its target, then after it faces the target, it spins past its desired facing (due to momentum), then the AI compensates like a pendulum. AI ships may wobble while trying to face its target. This is especially noticeable on ships with high max (over 9) and low accel (2 or less).
Download and Play in 1.7 Beta...
Playership Drones v7 (Beta): Acquire and command almost any ship in Stars of the Pilgrim or Eternity Port.
Drake Technologies (Alpha): Updated for 1.7! More hardware for combat in parts 1 and 2!
Star Castle Arcade: Relive classic arcade gaming in a new Transcendence adventure!
Godmode v3 (WIP): Dev/cheat tool compatible with D&O parts 1 or 2.

User avatar
Wolfy
Fleet Admiral
Fleet Admiral
Posts: 5342
Joined: Tue Feb 05, 2008 1:10 am
Location: Somewhere in the Frontier on a Hycrotan station, working on new ships.

Re: AI failings in 1.3 beta (and earlier)

Post by Wolfy » Sat May 31, 2014 7:48 pm

At least that was the correlation I was able to observe with the named ships that pm912 was testing, and he drew the same conclusions.
(shpOrder gPlayership 'barrelRoll)
(plySetGenome gPlayer (list 'hycrotan 'nonBinary))
Image
Image

kay.py
Militia Lieutenant
Militia Lieutenant
Posts: 139
Joined: Mon Dec 28, 2009 10:33 am

Re: AI failings in 1.3 beta (and earlier)

Post by kay.py » Sun Jun 01, 2014 4:55 am

I assume this is what is making the meth enforcers hilariously bad pilots? Either that or they've been sampling the merchandise...

PM
Fleet Admiral
Fleet Admiral
Posts: 2202
Joined: Wed Sep 01, 2010 12:54 am

Re: AI failings in 1.3 beta (and earlier)

Post by PM » Sun Jun 01, 2014 1:35 pm

Not just Meth Enforcers, but also ships like Repentents and Ares Chasms and Tundras... or any ship with max of 18 and accel of 2 or less.

Speaking of Chasms and Tundras, shouldn't those two have lower max, say 9 instead of 18? They used to spin no faster than the pre-1.3 Sapphire or Wolfen, but now they spin as fast as a Sandstorm or Hornet. They are heavy gunships like the Centurion, which has max of 9.
Download and Play in 1.7 Beta...
Playership Drones v7 (Beta): Acquire and command almost any ship in Stars of the Pilgrim or Eternity Port.
Drake Technologies (Alpha): Updated for 1.7! More hardware for combat in parts 1 and 2!
Star Castle Arcade: Relive classic arcade gaming in a new Transcendence adventure!
Godmode v3 (WIP): Dev/cheat tool compatible with D&O parts 1 or 2.

george moromisato
Developer
Developer
Posts: 2849
Joined: Thu Jul 24, 2003 9:53 pm
Contact:

Re: AI failings in 1.3 beta (and earlier)

Post by george moromisato » Tue Jun 03, 2014 5:09 am

I found a bug which caused the AI to ignore rotation acceleration in its calculations. Fixing the bug fixes some of the maneuvering problems with Repentants, etc. [They no longer swing wildly from side to side.]

User avatar
Wolfy
Fleet Admiral
Fleet Admiral
Posts: 5342
Joined: Tue Feb 05, 2008 1:10 am
Location: Somewhere in the Frontier on a Hycrotan station, working on new ships.

Re: AI failings in 1.3 beta (and earlier)

Post by Wolfy » Tue Jun 03, 2014 5:22 am

Ok, looks like my hypothesis was right - good to see the AI works now!
(shpOrder gPlayership 'barrelRoll)
(plySetGenome gPlayer (list 'hycrotan 'nonBinary))
Image
Image

george moromisato
Developer
Developer
Posts: 2849
Joined: Thu Jul 24, 2003 9:53 pm
Contact:

Re: AI failings in 1.3 beta (and earlier)

Post by george moromisato » Tue Jun 03, 2014 5:07 pm

Found and fixed another AI regression: when aiming at a freighter/capital ship with a large exhaust trail, the AI was treating the trail as part of the ship for purposes of computing its angular size. This made the ship seems bigger than it really was, which made the AI think that it could hit the ship, when in reality it could not.

User avatar
Atarlost
Fleet Admiral
Fleet Admiral
Posts: 2391
Joined: Tue Aug 26, 2008 12:02 am

Re: AI failings in 1.3 beta (and earlier)

Post by Atarlost » Tue Jun 03, 2014 10:55 pm

george moromisato wrote:Found and fixed another AI regression: when aiming at a freighter/capital ship with a large exhaust trail, the AI was treating the trail as part of the ship for purposes of computing its angular size. This made the ship seems bigger than it really was, which made the AI think that it could hit the ship, when in reality it could not.
That explains why the AI fired behind some ships and ahead of others. Now we just need a fix for the latter issue.
Literally is the new Figuratively

george moromisato
Developer
Developer
Posts: 2849
Joined: Thu Jul 24, 2003 9:53 pm
Contact:

Re: AI failings in 1.3 beta (and earlier)

Post by george moromisato » Wed Jun 04, 2014 12:19 am

Atarlost wrote:
george moromisato wrote:Found and fixed another AI regression: when aiming at a freighter/capital ship with a large exhaust trail, the AI was treating the trail as part of the ship for purposes of computing its angular size. This made the ship seems bigger than it really was, which made the AI think that it could hit the ship, when in reality it could not.
That explains why the AI fired behind some ships and ahead of others. Now we just need a fix for the latter issue.
I believe it is the same problem. The code used the incorrect ship size and *centered it* on the ship center. So the ship looked bigger both in front and behind.

NMS
Militia Commander
Militia Commander
Posts: 374
Joined: Tue Mar 05, 2013 8:26 am

Re: AI failings in 1.3 beta (and earlier)

Post by NMS » Thu Jun 19, 2014 10:31 pm

I recently observed a very odd battle at the main screen in 1.3 b2. An EI500 was just drifting through space, not turning or thrusting, but it did fire. The enemy, a Molotok, was carefully not aiming at the freighter, such that every single dragonfly shot missed completely, regardless of the range and direction. However, it was firing its MAGs at the freighter using its swivel.

Post Reply

Who is online

Users browsing this forum: No registered users and 1 guest