Talk:Rambot Challenge 2K6

From Robowiki
Jump to navigation Jump to search

PreChat

Possible candidates:


(copied from GrubbmGait page)

GrubbmGait .. are you up for a rambot challenge? We could find an opponent, let's say the one who GrubbmThree 0.8 had the lowest percentage against (as long as it doesn't give opponents a sizeable advantage) and see who can cause the most damage in 500 or so rounds. We can break down the results on pure ram damage, showing how effective your movement is, and gun damage, showing how effective your gun is in the extreme conditions of short rounds and very close range. (Ugluk v0.9.2b is a rambot.) -- Martin Alan Pedersen / Ugluk

As ramming is suddenly very popular, I also have been thinking about a RambotChallenge2K6. As long as I remember, CassiusClay has been the bot where GT 0.8 has the lowest score against. We could use a real version, but it is also possible to use the TC version which only moves and does not fire back. In fact, we could use the whole set of the TargetingChallenge2K6 (and therefor TCCalc to calculate scores if it also showes bullet- and ram-damage), but I don't know if all those TC-bots are any good as ram-reference. That has to be checked. The rambots themselves do not have to be changed, as their ramming is their targeting. I say, lets ram! -- GrubbmGait

YAY! I've been wanting to work on a rammer for a while now, so I agree! RAMBOTS! --Bayen

So, is it an idea to use the bots from the TargetingChallenge2K6 and pound on them for 500 rounds? It would take a lot less time than the TC2K6, as the rounds are quick and the rambots are fast. These referencebots also provide a mixture of movements, so the generalist would be better than the specialist (well, ofcourse ramming is their speciality). Is it an idea to give nanobots a bonus and everything bigger than a micro a penalty? Lets hear some more ideas so we get consensus and start the dark era of roboRam! -- GrubbmGait

I probably won't be participating, but I would disagree with penalizing for codesize. Comparing bots of equal codesize to decree a bot "best micro rammer" or "best nano rammer" seems reasonable; but I think the challenge should be looking for the "best rambots", not the "best and most efficient ramming algorithm". Your (plural) decision, though =) Just my 2 cents. -- Voidious

I think there should be two different rankings: one for the best rambot overall, one for the most efficient rambot in codesize! --Bayen

    Two things:
  1. My goal is to find the ultimate movement and best gun to compliment it. Codesize isn't a concern on my end, so I am biased, but I doubt there is enough competition to support a breakdown by codezise category.
  2. I wonder about the relevance of trying to ram a target that cannot fire back. Assuming the inactivity timer is reset with ram damage, these battles could be really really long, usually resulting in the destruction of the defenseless opponent. Then again maybe CassiusClay is that good at dodging. It just seems like if two bots can ram CC to death each round they will both score 100%. If it can fire back, you get a measure of how well they deflect targeting as they close the gap (as opposed to GrubbmThree who eats bullets for breakfast).

-- Martin

I think a remark about the WeightClass is enough, no separate rankings necessary then. The challenger is allowed to fire, so it will empty its magazines while closing in on the referencebot. If it hits it often enough with its bullets, it wins. Otherwise it gets disabled and looses. I just have to run a couple of rounds against the TC2K6 referencebots to see if it really is a good set for ramming, but currently I am not able to do so. Maybe later this evening. Your remark about evasive manoeuvering is a valid one though. -- GrubbmGait

Any test of a ram bot should have firing opponents, imho. Without them, it's impossible to measure the predicted success of a bot in the rumble - the rambots will only improve their guns, not their movement, if their opponent doesn't fire. -- Greywhind

Firing opponents it is then. Now about the opponents, 10 is a nice number. My suggestion is to select three flee-ers (like CassiusClay), three close range evaders (like DoctorBob) and some randomish movement bots. But, if you want to see more details like bulletdamage and ramdamage (bonus also?), less bots provide a better overview. Maybe one bot of each type provides enough information. How about the results, standard robocode scoring or should we forget survival and only count bullet- and ram-damage and bonusses. -- GrubbmGait

I believe we should include survival in order to get a more acurate reflection of the rumble. RaikoMicro, FloodHT, DoctorBob, CassiusClay, and HawkOnFire look like good test bots to me. -- Kev

I hope that we can find some opponents who are hard to ram because they can dodge well, and with distinctive techniques, rather than just having a gun that is too tough to beat even if you weren't ramming them. I've been testing against GrubbmGrb and losing badly, though normally I don't fear his gun. -- Martin

I would like to enter the prototype for my rambot just to see what you think of it. It is called UbaRamLT, and it rams using Linear Targeting! I haven't seen anyone else try this yet, so could someone tell me what you think? --Bayen

Many rammers predict the other robot's position in advance. The trick with ram bots is to balance between moving towards where the enemy might go to (to cut the other robot off) and where it currently is (in case the other robot reverses). -- Kev

I've put some possible candidates at the top. I think 4 bots (2 that like to flee, 2 that take evasive manouvres) is a nice number if you also want to know some details like bullet- and ram-damage. I also think that bots that move randomly are not very interesting, as the scoring for most rambots would be roughly the same. Further 500 rounds seems ok to me, if necessary we can always start a FastRamming with 15 battles of 35 rounds to better reflect the rumble. -- GrubbmGait

It seems like a good set of bots and 500 rounds sounds good to me. How would we show ram and bullet damage? Average damage per round might work, but I don't know if we should keep the score "ram damage times two" like robocode does. -- Kev

  • Well, if you reduce the ram damage score you inflate the importance of bullet damage, which strikes me as counterproductive to a rambot challenge. -- Martin

This is an example of the possible layout. Bulletdamage and ramdamage are just the values shown by robocode. -- GrubbmGait

Bot Name Author Size CC bdam CC rdam CC score DB bdam DB rdam DB score HOF bdam HOF rdam HOF score RM bdam RM rdam RM score Overall Score
NanoDeath 2.56 User:Mike Zhang nano 27606 8475 28.99 40818 9657 36.56 49789 8218 47.35 46048 20847 62.50 43.85

How about:

Bot Name Author Size CC bdam CC rdam CC score DB bdam DB rdam DB score HOF bdam HOF rdam HOF score RM bdam RM rdam RM score Ramming Score Overall Score
NanoDeath 2.56 User:Mike Zhang nano 27606 8475 28.99 40818 9657 36.56 49789 8218 47.35 46048 20847 62.50 23.60 43.85

We should probably stipulate that bots whose overall score is not at least 15% from ram damage do not qualify for this challenge. -- Martin

From old wiki

Would you guys consider adding an AntiGravity Bot to the list of references? I'm just curious wether they'd be pinned against the wall or escape the Rammers :-) -- Dummy

Anti-gravity works well against some, but others (notably GrubbmThree) will corall their opponent into a corner and pin them there. The best combination I came up with (given the tools I'd already created) was tangental oscillation (e.g. SnippetBot) with low anti-gravity fields from walls and the opponent's robot. The 'tango' movement throws off targeting and also gives rammers the slip (often) even when already collided with. -- Martin

There are no threads on this page yet.