User talk:Pedersen/Banzai!

From Robowiki
Jump to navigation Jump to search

I have to disappoint you, no rambot will ever be able to outscore GrubbmGrb. Its gun is to strong at close ranges, and it also hits when the rambot is homing in. Even my own rambot scores a measly 38%. But if you don't try it, you're sure you never beat it, so please try. -- GrubbmGait

My victory shall be all the sweeter for it! I frequently use GrubbmGrb as a litmus test for bugs in Ugluk. If I lose, I didn't code it right. If I win, it's probably performing as expected. -- Martin

After 4 hours of very frustrating test results I figgured out that my wave hits were recording against the wrong object, so my scores were very similar to the WaveSurfingChallengeBotA results no matter what combination of segmentations I tried. After a few more hours testing, I am only using 2 of the 6 segmentations I've coded, though oddly the unsegmented gun performs pretty well. Banzai! shares the gun with Ugluk, and though I threw in some movement tweaks along the lines of previously unthrilling attempts, I think the gun adjustments / fixes are the core of the 110 point rating jump, bringing Banzai! v0.3 in line with the rest of the ramming population (MaxRisk has ~1650).

Banzai! v0.4 only lost a few points, but I didn't really make any changes, aside from the revamp of Ugluk's physics subsystem. Banzai! v0.5 features the return of the fully segmented bearing offset gun as well as some bug fixes. Banzai! v0.6 has new virtual gun statistics segmentation. v0.5 was in Ugluk mode for some opponents. This has been reverted.

I can't download Banzai v0.5. Please, check URL to bot. -- DemetriX

  • I just uploaded it to my host, but after reading why I pulled 0.5, I'll switch it back to 0.6a. -- Martin

I have been reluctant to update Banzai! following major overhauls in Ugluk's code (they share a common codebase), but since I've had a break from coding (bots) now is as good a time as any. -- Martin

Well, I decided to make a non-firing rambot and a non-firing ram-avoidance bot. Watching the battles, I saw the avoider consistently zooming away from the rammer across open ground. So I added some speed tracking and found the avoider had an average speed of ~7.5 while the rammer had an average speed of ~5.75. Clearly something in my throttle is awry. I have been adjusting my maximum speed to allow me to make tight turns without overshooting, but when my angle adjustment should be minimal, I should be hitting top speed. -- Martin

Well, the 130 point rating difference amounted to changing a 1 to a 4 in the ramming movement method. If my memory weren't muddled from months of absense, I'd know exactly, but I think it correlates to how hard the movement pushes the robot, and 48.0 is working far better than 18.0, even though there are no other factors (e.g. anti-wall gravity) at play. It's too early to call it, but it will be nice to have a ram bot that finally beats the "average" mark of 1600. -- Martin

I thought I had disabled all opponent-specific behaviors from Banzai! (only present since he shares a large portion of Ugluk's code) but I've tested the latest release version against one of the suspicious results (i.e. 90%+) and he is clearly not ramming. I really want a baseline to see how the movement system changes affected my ramming ability. Still, while 0.7.1's rating is artificially inflated, he still demonstrates the effectiveness of ramming (for better or worse) as a strategy against opponents who are difficult to beat in any other fashion (such as the top bots). Rather than being satisfied with a 26.4% score vs. Dookious, for example, I can bump it up to 36.8% by ramming him. It may not be a big deal in the grand scheme of things, but it sure feels right. =) -- Martin

I can't download pedersen.Banzai_0.6a (http://home.comcast.net/~kokyunage/robocode/ugluk/pedersen.Banzai_0.6a.jar). -- DemetriX

  • Apparently I cleaned it out with a bunch of older tanks. It is back up again. -- Martin
There are no threads on this page yet.