Sanguijuela/Evolution History

From Robowiki
Jump to navigation Jump to search

Sanguijuela 0.8

Download

Released: 27 April 2008

Codesize: 748

What's new

  • Implementation of the deceleration acceleration trick when changing direction.
  • Faster in some other special cases.


Sanguijuela 0.7

Download

Released: 20 April 2008

Codesize: 588

Ratings Date: 27 April 2008

Rumble Position Points Battles
General 263 1647.54 2419
Mini 108 1669.87 1374
Micro 68 1660.69 988

What's new

  • Bug fix. When going to the center, stop if the discovered host is not correctly aligned.
  • Internal restructuration (Code rewrite).

Comments

Some codesize left and better performance than evolution 0.6.


Sanguijuela 0.6

Download

Released: 10 April 2008

Codesize: 724

Ratings Date: 20 April 2008

Rumble Position Points Battles
General 262 1641.37 2043
Mini 109 1663.45 1038
Micro 67 1656.76 711

What's new

  • Rollback to linear movement and targeting. The branch of evolution started with 0.4 is over.
  • Bug fix that prevents Sanguijuela losing a turn when turning to face the enemy at the beginning.
  • Go to the center of the territory until there is no host.

Comments

Better results than evolution 0.3 but Sanguijuela is still 7.66 points away from GrubbmThree.


Sanguijuela 0.5

Download

Released: 2 April 2008

Codesize: 635

Ratings Battles: 3108

Rumble Position Points
General 1631.88

What's new

  • Choose each round between Linear (90%) and Circular (10%) movement and targeting
  • Change the calculation of the best initial Radar direction using Smash method by Starrynte. My code got the same results but with more codesize.

Comments

It was not good idea but I wanted to try it. Looking at the comparison between 0.3 and 0.4, Linear is better in a 76% of the conflictive cases. If I choose between Linear (90%) and Circular (10%) strategy then

p(ok)= p(s=L)*P(ok|s=L) + p(s=C)*P(ok|s=C) = P(ok)*P(s=L) + P(ok)P(s=C) =0.76*0.9 + 0.24*0.1 =70.8

So Sanguijuela lost 5.2% of the conflictive cases but I thought that adding this variability in the strategy could be interesting. The 0.3 (only Linear) has still better performance.


Sanguijuela 0.4

Download

Released: 24 March 2008

Codesize: 727

Ratings Date: 1 April 2008 Battles: 2388

Rumble Position Points
General 296 / 616 1619.14
Mini 116 / 306 1650.26
Micro 72 / 209 1645.37

What's new

  • Change Radar from WideLock to NarrowLock.
  • Change linear movement and targeting to circular. The predicted position of the enemy use 8 as velocity for the movement and 11 for the targeting.

Comments

Involution. After some testing I thought that a circular movement and targeting could be better than linear but, in general, it seems to be false. The previous version appears and disappears misteriously in the RoboRumble so I could make these statistics.

Comparison for v0.3 Linear and v0.4 Circular


Sanguijuela 0.3

Download

Released: 8 January 2008

Codesize: 749

Rumble Position Points Battles
General 1638.62 4326

Ratings Date: 18 January 2008 Battles: 2024

Rumble Position Points
General 253 / 604 1624.67
General PL 279 / 604 612
Mini 108 / 302 1643.79
Mini PL 122 / 302 338
Micro 67 / 203 1645.21
Micro PL 75 / 203 236

What's New

  • Now Sanguijuela doesn’t ignore the walls in the linear movement and targeting.
  • It doesn’t bite before it is close to the enemy. (350 pixels).


Sanguijuela 0.2

Download

Released: 6 January 2008

Codesize: 743

What's New

  • Code reduced to 743. Now Sanguijuela is a MicroBot.
  • Sanguijuela ignores the walls when moving and targeting.

ToDo

  • Done v0.3: Don’t ignore the walls and mantain the micro category.

Comments

I have tested Sanguijuela's ramming performance (V0.2) with the archived results of RamBotChallenge2K6 (two years ago) and it takes the 3rd place.

Bot Ramming Percentage Overall Score
MaxRisk 0.6 24.65 46.39
SledgeHammer 0.22 21.42 45.26
Sanguijuela 0.2 21.86 44.11
ARAMtocles 0.3 20.65 44.05
NanoDeath 2.56 23.60 43.85


Sanguijuela 0.1

Download

Released: 5 January 2008

Codesize: 869

Origin

  • Linear Ramming and Targeting.
  • It calculates the best direction to start scanning.
  • It chooses ramming with the anterior or the posterior suckers depending on host's starting position.