[Home]GrubbmThree

Robo Home | Changes | Preferences | AllPages

GrubbmThree

Author

GrubbmGait

Description

GrubbmThree is a bot which only focus is ramming. Clean, simple and not very sympatic.

Great, I want to try it. Where can I download it?

micro.GrubbmThree: http://www.robocoderepository.com/BotDetail.jsp?id=2444

How competitive is it?

march 2005: As it is higher ranked than mz.NanoDeath and radnor.RamRod, it looks like it is the best rammer around.
feb 2006: The crown is disputed by Kev's MaxRisk.

How does it move?

Ramming, it tries to intercept the opponent by following its own bullets.
The movement code really is just two lines: Turn to where I shoot, and move ahead.

How does it fire?

It uses an iterative CircularTargeting gun and tries not to shoot into the walls.

How does it dodge bullets?

It eats bullets for breakfast.

How does the melee strategy differ from one-on-one strategy?

Ever seen a successfull melee-rammer ? ?

Where did you get the name?

Surprisingly it is my third bot.

Can I use your code?

It's included in the jar under the RWPCL.

What's next for your robot?

Nothing until it is rammed from the throne.

What other robot(s) is it based on?

Default framework of micro.GrubbmOne.

History

MICRO
20060216: v 0.9 -- Put aiming-point always within field, less energy preservation.
Rank Mega:174 -- Rating:1635 -- PL:178
Rank Mini:70 -- Rating:1673 -- PL:66
Rank Micro:40 -- Rating:1679 -- PL:42
Comment: it seems a draw with MaxRisk 0.5

20050316: v 0.8 -- Do not shoot at the center of the enemy, shoot at the outer hull.
Rank Mega:135 -- Rating:1637 (down to 1629) -- PL:133
Rank Mini:59 -- Rating:1675 (down to 1662) -- PL:61
Rank Micro:37 -- Rating:1683 (down to 1670) -- PL:35
Comment: Prediction is less far in the future and therefor a bit more reliable. NanoDeath is outranked (again) and now for good.

20050130: v 0.7 -- Refactored CT-gun (integrational -> iterative) to win 100 bytes and still have improved near-wall performance.
Rank Mega:145 -- Rating:1630 (down to 1623) -- PL:145
Rank Mini:63 -- Rating:1659 -- PL:64
Rank Micro:40 -- Rating:1671 -- PL:38
Comment: Nailing the opponent when hitting the wall is more effective than try to shoot through the wall.

20050127: v 0.6 -- Improved radar, removed sharp-turn.
Rank Mega:166 -- Rating:1606 -- PL:182
Rank Mini:74 -- Rating:1638 -- PL:85
Rank Micro:44 -- Rating:1650 -- PL:55
Comment: Sharp-turn is too predictable and slows down the hunt.

20041219: v 0.5 -- Back to CT-gun, make sharp-turn sharper, fire also when angle between 10-20 degrees.
Rank Mega:169 -- Rating:1604 -- PL:186 (Down to 1600)
Rank Mini:73 -- Rating:1616 -- PL:-
Rank Micro:? -- Rating:1634 -- PL:-
Comment: Firing when close enlarges the fire-angle.

20041218: v 0.4 -- Changed CT-gun into averaged HOT-CT-gun.
Rank Mega:203 -- Rating:1563 -- PL:-
Rank Mini:95 -- Rating:1573 -- PL:-
Rank Micro:65 -- Rating:1588 -- PL:-
Comment: Not a good idea . . .

20041202: v 0.3 -- Changed bpower/distance ratio, made sharp turn less sharp.
Rank Mega:180 -- Rating:1593 -- PL:-
Rank Mini:87 -- Rating:1588 -- PL:-
Rank Micro:56 -- Rating:1607 -- PL:-
Comment: Improvements against the small bots (which was not needed) and no improvement against the large bots (which was what I needed).

20041127: v 0.2 -- Solved sharp-turn bug, removed 10-tick wait at startup.
Rank Mega:180 -- Rating:1592 -- PL:-
Rank Mini:87 -- Rating:1582 -- PL:-
Rank Micro:56 -- Rating:1591 -- PL:-
Comment: Extended tryouts do not give better results, NanoDeath keeps the crown.

20041126: v 0.1 -- Initial version, just ramming and shooting.
Rank Mega:194 -- Rating:1577 -- PL:-
Rank Mini:92 -- Rating:1567 -- PL:-
Rank Micro:61 -- Rating:1573 -- PL:-
Comment: A bit less than expected, due to 2 small bugs (I hope).


Comments, questions, feedback:

GrubbmThree has a higher ranking than mz.NanoDeath and radnor.RamRod, does this mean that this is the best rambot around? I know it is not fair to compare a nano with a micro, but at least I will be the best micro-rammer. --GrubbmGait

Hmmm... Why isn't BiggerDeath? ranked anymore?

And I haven't seen a successful melee rammer, but DoctorBob is pretty close to that - it's quite good at melee and also very aggressive (by staying close to one enemy, you stay away from all the others!) -- Kawigi

I know BiggerDeath? is active on the RobocodeLittleLeague, but I've never seen it on the RoboRumble. Checking its results of 20050312 it does not seem to perform better than either NanoDeath or GrubbmThree, but there is only one way to be sure: the author should let BiggerDeath? loose in the rumble :-)) -- GrubbmGait

This ram bot certainly gave me trouble off and on for a while (depending on my overall rating). I think I've finally found the key though. With other ram bots I was able to use a combination of wall and robot repellant fields to maintain distance, and with a little movement mirroring I'd escape chase like a hockey puck. Not so with GrubbmThree though. This one pins you in a corner and sticks it to you. I had to redesign my bot a bit to allow me to throw energy management out the window against rambots. At first I just rammed right back, which gave me a little over 50% of score. That still wasn't enough, so I did some more research and found a winning combination. There are some other rambots out there, but this is the only one I fear. -- Martin / Ugluk

Yeah, this is a mean rammer. For me to really avoid it would probably mean a special anti-rammer movement instead of my current anti-rammer adaption of my BlindMansStick driving. What saves CassiusClay is the power of its gun it seems. If GrubbmThree had a good gun it would become really dangerous. Maybe we should try with the Aristocles gun? It's small and I think we could tune the segmentation to take advantage of the special situations the target is forced into by the ProvocativeMovement of GT. -- PEZ

I have shrunk GrubbmThree somewhat and put in the Aristocles gun. Alas I know nothing about GuessFactor guns and segmentation-tuning, so PEZ, if you have the time, could you look into this? It also uses the bulletpower calculation of Aristocles as that is part of the distance-segmentation. I am curious if it will be better than GrubbmThree, as the movement just follows the targeting. Kev already proved that better rambots are possible. -- GrubbmGait

Does it work as is? If so you should try it in the rumble. I won't have time to look into it until tonight earliest. But I will do that of course. What I imagine we should tune is the distance segmentation and if possible also add reverse a wall proximity segmentation. -- PEZ

May be its good to segment on yourself too! The enemy bot reacts on your movement and this movement seems to be agressive :) I tried this without success in Krabbys GF gun, but please give it a try! --Krabb

That's the thing with Aristocles' set of segmentations. It captures quite a lot about itself. Especially if we add reverse wall segmentation. -- ~~~~

You can find it under the name wiki.micro.ARAMtocles, source is ofcourse included. It has 20 bytes free and a victory-dance, so enough room for some tuning. Now no-one can escape anymore! -- GrubbmGait

The comparison details doesn't look too promising yet. But it will be fun to see if it can be made to work better than the original gun. -- PEZ

Hehe, but it looks like this beast will force ABC to visit his anti-rammer strategies. =) -- PEZ

I don't have anti-rammer strategies! It looks like my redone movement sure needs some... -- ABC

BlindMansStick movement isn't ideal against rammers. At least with my implementation I have serious problems making my bot behave "as it should" close up. One thing that definately helps is shortening the stick when facing a rammer close up. This makes you vulnerable for slamming into the walls instead though. I have anti-ram on my todo-list, but I have no real ideas about what it means... -- PEZ

Aristocles' gun fires with power 3.0 when closer than 360, is this the intention? Aramtocles really needs some tuning indeed, but that could be expected with RamRod as example. As for anti-ram movement, slowing down a bit (f.e. to 4.0) could help you making a sharp turn and avoiding the walls. Rambots usually are not good at making sharp turns. -- GrubbmGait

What baffles me (due to a lack of research) is why GrubbmThree doesn't have a higher rating. Ram bots wouldn't be such problem bots if their difficulty were felt all around. Is it just that they tend to bring everyone to the same level and are 'just as easily' dealt with by low ranked bots as by high ranked ones? I guess low ranked bots are getting a lot more bullet damage against ram bots than they are used to, as well as unintentional ram damage, so it balances out the damage taken and round losses. -- Martin

Why should it have a higher rating? The supposed mean of the ratings is 1600 if I remember correctly. Ramming in to your opponents makes you trouble for the good bots and easy pray for the bad. Relatively that is. 50% is what you can hope for unless the opponent is doing something anti-ramming or having some ramming weaknesses. I think that's bound to keep you around the 1600 mark unless you do something really clever. And I still think adding 1600 to the rating is silly. -- PEZ


Robo Home | Changes | Preferences | AllPages
Edit text of this page | View other revisions
Last edited February 20, 2006 11:22 EST by GrubbmGait (diff)
Search: