CSc 180 -- Spring 2015
Intelligent Systems


2015 Tournament Results - "GasTank Chess"


       **** RESULTS ****
13th ANNUAL AI STRATEGY GAME CONTEST
------------------------------------

The 13th Annual AI Strategy Game Contest took place April 7-9, 2015.
The atmosphere was extremely hectic, with the largest participation
ever in the event, as well as many spectators in the form of students,
faculty, alumni, and previous champions and contestants.  No less
than three former champions were in attendance, including the 2003,
2007, and 2013 winners.  One alumnus showed up with an implementation
of his own, that turned out to be quite strong.

The game this year was "GasTank Chess", which is a sort of reduced
version of chess, with the additional feature that each piece is a
sort of "car" that starts out with a certain amount of "gas" in its
gas tank.  Each time the piece moves, its gas level is reduced.  If
the gas ever reaches zero, the piece becomes dead and can no longer
move.  Capturing, at any time, restores the gas tank level to full.
Additionally, players are allowed to capture their own pieces(!),
which often proves useful in order to fill up a gas tank.  The game
itself proved to be quite tactical and entertaining, often requiring
clever combinations, exchanges, and sacrifices in order to win.

This year there were 33 entrants.  Most programs used minimax with
alpha-beta pruning, and many also used iterative deepening.  A few
used more advanced features such as history tables, killer move
heuristic, and other clever sorting methods to enhance the pruning
and increase search depth.  Two entrants this year implemented
bitboards - a difficult but highly efficient way of generating
legal moves.  One entrant implemented transposition tables with
Zobrist key hashing for even greater efficiency.  Most examined
millions of positions within the 5-second time limit per move.
GasTank Chess has a high branching factor, despite which many of
the programs examined 6 plies deep, sometimes more than 10 plies
later in the game.  About a quarter of the students claimed that
they cannot beat their own program.  This year also included the
widest array of languages used, including Java, C++, C, C#,
Python, and Javascript.

As always, each match consisted of two games, so programs could
lose individual games along the way and still win by amassing
more victories.  The instructor also competed with his program,
named "Buggy".  As the event drew to a close, it became a
closely contested final round with four programs in contention:
"Diesel Wolf" (John Hill), "Alter Ego" (Stephen Ly), "Sneaky"
(Guraj Atwal), and the instructor's program "Buggy".  All four
were written either in C or C++.  In the end, it was a tie for
first between "Diesel Wolf" and "Buggy", with "Alter Ego" in 3rd,
only one point behind the winners.  This was the very first time
that the instructor's entry finished in (or tied for) the top spot.

Many of the programs improved between day 1 and day 2.  Steven Ly's
3rd place finishing "Alter Ego" barely made it through the quarter
finals on tuesday,  then came back on thursday with a much stronger
and more stable program.  But by far the most improved program was
David Grapentine's "Grapes of Wrath".  After spending most of
tuesday awkwardly capturing all of its own pieces, it returned on
thursday fully debugged to win every one of its 12 games in the
lower half of the draw, and securing the coveted "Grand Hamster"
award for the most improved program of the tournament.

Complete results are shown below, including results from prelims,
rounds 2 and 3, quarters, semis, finals, and all consolation rounds:

                                                           Total
                                                      -----------
 1. DieselWolf    (Hill)....f:4-2 s:5-1  q:4-0 p:4-0        17-3
... Buggy         (Gordon)..f:4-2 s:5-1                      9-3
 3. Alter Ego     (Ly)......f:3-3 s:6-0  q:2-2 p:4-0        15-5
 4. Sneaky        (Atwal)...f:1-5 s:3-3  q:4-0 p:4-0        12-8
 5. Up&Down       (Nguyen)....... s:2-4  q:3-1 p:4-0 c:2-0  11-5
 6. jTurtle       (Kerr)......... s:2-4  q:3-1 p:4-0 c:3-1  12-6
 7. Bingo         (HuynhN)....... s:0-6  q:2-2 p:4-0 c:1-3   7-11
 8. Janky         (Clemens)...... s:1-5  q:2-2 p:4-0 c:0-2   7-9
 9. Defection     (HuynhT).......r3:4-0 r2:4-0 p:2-2 c:4-2  14-4
10. iLose         (Russell)......r3:3-1 r2:4-0 p:0-4 c:3-3  10-8
11. Shah          (Parnia).......r3:6-0  q:1-3 p:3-1 c:3-3  10-7
12. VanilaSlushie (Chua).........r3:5-1  q:1-3 p:3-1 c:5-1  14-6
13. Airo          (Belcher)......r3:5-1  q:1-3 p:4-0 c:2-4  12-8
14. Bean          (Ngouyapanh)...r3:3-3  q:1-3 p:2-0 c:4-2  10-8
15. Gas-Navi      (Mihov)........r3:2-2 r2:3-1 p:2-2 c:3-3  10-8
16. D.E.A.T.H.    (Thompson).....r3:2-2 r2:2-2 p:2-2 c:0-6   6-12
17. That'sNmbrwng (Watson).......r3:1-3  q:0-4 p:3-1 c:5-1   9-9
18. Nibbler       (Yee)..........r3:2-4 r2:2-0 p:2-2 c:4-2  10-8
19. C.L.U.        (Malonee)......r3:0-4 r2:4-0 p:2-2 c:3-3   9-9
20. GasChess      (Strang).......r3:1-5 r2:3-1 p:2-2 c:1-1   7-9
21. Cuddles       (Bray).........r3:2-4 r2:4-0 p:0-4 c:0-6   6-14
22. DAY           (Dalisay)......r3:0-6 r2:3-1 p:2-2 c:1-1   6-10
23. GrapesOfWrath (Grapentine)...r3:2-0 r2:2-2 p:0-4 c:8-0  12-6 [GH]
24. KM-180AI      (Mack).........r3:4-0 r2:2-2 p:0-4 c:3-3   9-9
... W.W.B.B.D.    (Ali)..........r3:1-1 r2:2-2 p:0-4 c:2-2   5-9
... Zoopster      (DeLeon).......r3:1-1 r2:1-3 p:0-2 c:3-3   5-9
27. Lamp          (Herold).......r3:4-0 r2:0-2 p:2-2 c:0-2   6-6
28. Ultron        (Gregory)......r3:0-2 r2:2-2 p:1-3         3-7
29. TARS          (Bueff)........r3:2-2 r2:0-4 p:2-2 c:0-2   4-10
30. Secret Robot  (Hall).........r3:1-3 r2:0-4 p:2-0         3-7
31. Tiberius      (Dyer).........r3:1-3 r2:0-4 p:0-4 c:0-2   1-13
32. Phobos        (Jones)........r3:0-4 r2:0-4 p:0-4         0-12
33. AlphaBetaCal  (Tessema)............ r2:0-4 p:0-4         0-8
--  Nitro         (Zepeda)
--  Neo           (Graciano)

A few of the programs also incorporated attractive GUI interfaces,
and some used color in their display to make the game easier to follow.

There were a wide variety of opinions regarding which heuristics work
the best for this game.  Some multiplied the value of each piece by
the amount of gas left in the tank - while others treated the gas as
having a constant value.  Some incorporated center control or mobility,
and others tried to incorporate king safety.  There was no opinion
expressed as to whether it is advantageous to move first or second.

"GasTank Chess" was a great game and it is still far from being solved.

    

2014 results