Garm

What's special about it?

Up to now Garm is quite simple, it only uses one GFGun and WSMovement and saves no data. The only special thing is its accuracy, the enemy's GF1 and GF-1 is calculated with precise prediction and the botwidth is considered in the Gun and Movement code.

How competitive is it?

It just entered The2000Club and is ranked around place 20 int the RoboRumble. Its quite competive in the PL-ranking with rank 18 too (with disabled tick waves).

Great, I want to try it. Where can I download it?
https://designnj.de/roboking/Krabb.sliNk.Garm_x.x.jar

How does it move?
It surfs the first two waves, using some kind of TrueSurfing.

How does it fire?
One exact GF-Gun without virtual waves (Segmentaton is based on Komarious).

What does it save between rounds and matches?
Gun and movement data between rounds, nothing between matches.

How does the melee strategy differ from one-on-one strategy?
No meele stategy up to now.

Can I use your code?
Not jet.

What other robot(s) is it based on?

Current gun Segmentation is loosely based on Komarious.
Using "Fast Iterative Wall Smoothing" (By Voidious) and "Non-Iterative Wall Hugging" (By Simonton)
The rest is unique.

What's next for your robot?

First of all i need to improve its speed and clean its code.
Garm was originally designed to be a 1on1 melee and team bot. I'll update its code structure to reach this aim.

Versions:

Comments:
Whoa, close hit near Seraphim there, nicely done, might still beat it too. --Chase-san

Yea, i think i got you :) And GresSuffurd is only a few steps ahead! --Krabb

Now the three of us can race for The2000Club. Good luck ! Still lose against 50 bots, but ranked higher: I must be doing good (like always) against the less-than-average bots ;-) -- GrubbmGait

I´m good vs top bots (like always) :D Lets see who is the first one in The2000Club! --Krabb

Aww, do you ahve to make a competition out of it =)? Garm and Seraphim are bot closed source, but more then that I am currently working on a complex wall distance method (that should be alot faster then the current one at any rate). -- Chase-san

And now I am completely lost: in both TargetingChallenge2K6 (85.42 - 78.75) as in MovementChallenge2K6 (53.02 - 42.98) Garm is far ahead of GresSuffurd. Yet you are ranked 40 points lower in the roborumble. I only score better against HOT-bots, but thats only a 10% or so. Ideas someone? -- GrubbmGait

I´m a bit confused too :/ I can think of three reasons:

The gun and/or movement might only work correctly with one bot sitting still.
You might be better in 35 round battles.
I think the reason for the low score vs HOT-bots (and may be its "bad" ranking) is his distancing code. If the enemy is really close, Garm is forced to change direction whether there is a bullet or not.

I´ll check the 35 round thing first (its easier to check, but i don't think it's the reason), lets see --Krabb

It could be a combination of other things, in the TC and the MC one of the bots doesn't move, in the rumble they both normally move about. The challenges are just a benchmark of how good the components of your bots are. Doing better in the challenges means you will do better in the rumble, but it doesn't mean it is directly related to the rumble rating. Meaning a bot with a lower rank in the challenges could have a higher rumble rating. --Chase-san

Don't forget energy management, don't underestimate distance, and consider little strategy elements as well :-) -- Kawigi

Hey, those were points I had in my original 2 page speech before I summerized it into that. I guess great minds Do think alike. However I edited them out because thier value overall was less described by that statement. --Chase-san

To complete the List: Bumping into walls can increase your MC score. (Has your great mind considered this too? :P)--Krabb

A location-related "off by 1" problem in the wave systems for WaveSurfing or targeting could be hiding in the challenges since the position of one bot doesn't change; a similar problem with the absolute bearing would be half as much of a problem as it is in a real match; and your segmentation could behave differently if you're segmenting on distance instead of bullet time (not to say one is better than the other in the end...). Bullet power management can definitely be worth a few points... And in my experience, the MC doesn't even come close to completely benchmarking a movement's abilities - distancing, angle control, quick adaptation, and complimenting your targeting are all things that are largely overlooked. -- Voidious

My bad ranking in comparison to its challenge results was (as I assumed) related to the distancing and DiveProtection code. It failed in keeping garm at distance and screwed up the movement statistics. It turned out that Krabby2's old and simple solution works still quite well :) --Krabb

Looks like 0.7d is gonna make it! (+6 points 500 battles). Keeping my fingers crossed. -- GrubbmGait

Yea looks nice!! How is the wether in Holland? My school closed today because of "Kyrill" :) But it's still a normal, windy day here. --Krabb

It looks quite bad here, at 17:00 I have to drive home (160 km) so I hope I will be back in time for my competitiongame of darts. -- GrubbmGait

Ouch, that were two fast tweaks and now you're ahead of me. No guarantee for the victory though. It's time to overthink my segmentations and trying to understand CurveFlattening. -- GrubbmGait

It was fun to kick you out of the top 20 with that two small but efficient updates :) But my next step is more comprehensive and may be not that fruitful. Additionally I'll be on a class trip to Denmark for one week. Your chance to recapture your long-established rank. --Krabb

My bot is so funky I donno where to begin fixing/updating Seraphim, and building a new one would probably take far to long. --Chase-san

Could someone do me a favor? I get some missed scans with Garm when robocode runs minimised. It only happens when Garm skips turns. I wonder if this is just a problem on my pc. Garm would print something like "84: Missed Scan (9)" to the console. One "Missed Scan (2)" in the beginning is normal. Any 1.4x robocode version with Garm 0.8j should do the job (It doesn't happen with old robocode versions).

I could fix this problem whith a different radar(one from Radar), but I might not be the only victim :) --Krabb

Could this be a wrong setting for your CPU constant? Garm is a little bit of a SlowBot (not bad compared to some though). -- Skilgannon

Yeah, what's your CPU constant? Dookious is on the slower side, but still doesn't skip turns (on the default CPU constant of current Robocode versions). And is Garm really a SlowBot? I haven't checked lately, but I really didn't think so. It's not really fair to compare WaveSurfers with PrecisePrediction and GuessFactorTargeting to MicroBots when it comes to execution speed... I can try testing it for you - do you print when you skip a turn, so I can check? -- Voidious
- My CPU constant was 19, now i set it to 400. That should solve the problem :) Yes, Garm prints when it skips turns. --Krabb
Random sidenote: By "default cpu constant" do you mean the one you get if it was never calculated, or the one you get if you calculate it. Because the default in robocode is 400 milliseconds/turn, but on my machine it generally calculates to around 20. -- Simonton
I mean the one that Robocode sets if it calculates - it's 18 on my MacBook (2 GHz Core Duo). Back with 1.07, which set it to 1 (!!) on all modern machines, I'd set CPU constant to 100 just so nobody would skip turns. (The 1 ms was too small to measure accurately, so everybody would skip some turns in that setup.) -- Voidious
Note that the CPU constant for RR@Home clients only gets calculated when you start Robocode (not roborumble!!) for the second time. Mine is 21 btw. -- GrubbmGait
I just noticed your comment, Krabb, about the low CPU constant for the RR@H. I've been leaving 1 core of my 4 core work machine (2.4GHz Core 2 Quad) running the RR 24/7. When I ask it to calculate the cpu constant, it's around 10ms per turn... so perhaps it's my fault that Garm is having this problem. Sorry. Tim Foden
Never mind! I removed a code part which prevented Garm from crashing after skipped turns in v. 0.9r. If you use a new robocode version for your roborumble client your CPU constant should be right automatically. --Krabb
What's more, as from 1.5.4 the system granularity of the milliseconds no longer has any effect... all timing is done in nanoseconds which are always accurate. Just to be safe and taking worst case scenarios, make sure you're doing your regular work at the same time as running the "Reset CPU Constant" on your machine, but it shouldn't make much difference, being a 4 core. -- Skilgannon

Version 0.8k on 11th place, 2045 points (152 battles). That hurts . . but you deserve that jump. I just hope that that last few points do take some more time. -- GrubbmGait

Just a lucky streak :D --Krabb

Should have followed PEZ rule #1, "Assumptions are bad and assertions are good." I think one line, which seemed futile, is my undoing. By the way, i'm a bit confused, do you say commented or uncommented for "//setXYZ();"? In german it means "auskommentiert" :D --Krabb

Commented means it gets ignored by the compiler. So "//setXYZ();" is commented. It's called this because the main use is for adding comments, ie. explanations, to your code. And I thought PEZ's rule #1 was "Don't work on movement and targeting in the same release". =) -- Skilgannon

Hehe, he said that here. I think "comments are bad" is the next PEZ's rule #1 --Krabb

I think PEZ has a lot of rules, tied for #1. =) They are generally all good advice, though! -- Disciple of the Way of PEZ

I just noticed that Garm is pretty strong against the top bots, with nice fat PBIs against Ascendant (+9!), Phoenix, Dookious, Lukious, etc. You may have the makings of a PL champion on your hands here. =) --David Alves

Looks like you found the path back, awesome PL score. Now stop being nice and crush those weak bots. Hey, just compared Garm and GresSuffurd, and there is a lot red and green in the 'Diff' column for bots with almost the same rating. Just indicating that both of us have room to improve a lot. We still have about 50 weeks left to get to 10th/2050 -- GrubbmGait

I'm pretty satisfied with the 5. PL rank :D But I still don't beat my first WhiteWhale SandboxDT. At least there is no more red in the "% score" column at the details page! Hmm I suspect its not that easy to stop being nice, its one of my characteristics :) --Krabb

I have a question about .9d: what does "faster" binsmoothing mean? -- Simonton

I minimised the "window" width in the algorithm (It's a normal distribution with a variance of 1). I previously calculated the smoothing for the interval [bin-15;bin+15]. Now it ranges from bin-4 to bin+4. I also tried a faster algorithm (with no ^x calls), but it performed worse. In addition a lookup table is used to speed things up.--Krabb

Very nice rating jump! You need to add some time-since segments to your surfing though, because you are being hit quite badly by pattern matchers (Waylander, Smoke). Very good PL release, it seems. -- Skilgannon

Good hint, I already have some time-since-velocitychange segments, but they could probably be tuned. The general movement is more or less ready :) The only thing left to do is a flattener and some segmentation tuning, but the PL rank is good enougth so far :D The next step is a gun inprovement, it is still really horrible vs simple bots. --Krabb

I found time-since-decel to be more effective, time-since-lateral-direction-change also worked quite well. But I think the most effective was distance-last-10. =) -- Skilgannon

Hu? "distance-last-10" is the average distance during the last 10 ticks? --Krabb

That would be like nowLocation.distance(locationTenTicksAgo?). I use distance-last-8 in my gun. -- Voidious

Sounds good :) --Krabb

You're getting much to close to that 2050 (or was it place 10). Time to put the sparse time I have into GresSuffurd again instead of TwinDuel. Good work though! -- GrubbmGait

Looks pretty good, he? :) But maybe some points are due to different roborumble-client versions :/ Should we say top 10? You wouldn't have a chance otherwise. --Krabb

We could rephrase our goal: Who is the first to knock out CassiusClay and put him outside the top-10. Strange to see that proud ex-number 1 holding on with its fingernails to that tenth spot. Just wait till I release my next version, you'll be surprised (and I am affraid I will too) -- GrubbmGait

Seraphim 2 is still in the early beta's, it won't make your contest unless I decide to hammer it out all in a few hours like I do my homework assignments. EVen then, the original never even made the 2000 club. --Chase-san

This latest release is right on CassiusClay's heals... your top-10 aspirations may be fulfilled sooner than you thought! - Skilgannon

Yes, that >20-Point jump from 2055 to 2077 was great! Finally all the hard work paid off :)

GrubbmGait, It seems like i can't push CassiusClay out of the top 10 by my own. I need a mate in order to get him from place 9 to 11. Would you mind helping me with this? Somebody else is also welcome :) --Krabb

Seems you have done it without my help ;) Congrats on reaching the top-10 way before I came even close. I must say that after > 3 years of Robocoding, I kinda lost my interest, although you never know. Someone introduced me to SeaFight?, an online pirate game, and that is what I have been doing the last 2 months. I only have time to do one 'game', so my focus shifted to be a pirate. Good luck on your hunt to become the best German robocoder. -- GrubbmGait

I probably take a robocoding break too. I'll finish Garms melee compatibility and may be add some basic team play support. But I'm tired of tuning all the small parameters. May be I get the "best German melee robocoder" but probably not "best German robocoder". Winning the race to top 10 was all I wanted to do :) (After loosing the previous two^^) Take care of yourself as a pirate :P --Krabb