Submission #8541: Winslinator's A2600 Checkers "Game 3" in 12:40.03

Atari 2600
Game 3
(Submitted: Game 3)
(Submitted: Checkers (1980) (Activision) [!].a26 USA)
BizHawk 2.9.1
45543
59.9227510135505
3293
PowerOn
Submitted by 74FalconX on 8/19/2023 1:15 AM
Submission Comments
  • Note: Sorry, no encode that removes loads this time. Instead, I submitted alongside this a TAS that plays on the easiest difficulty with significantly shorter loads so check that out instead if you want to watch something quick.
Objectives
  • Uses hardest difficulty
  • Intends to obsolete the site's current publication for this game
  • Genre: Board
About this version of Checkers
This game uses official checkers rules which, most notably, means that jumping is always madatory: if a player has the option to jump, he/she must take it, even if doing so results in a disadvantage for the jumping player.
Umm... white always moves first though?? That's not the official rule but ok.
Difficulty Choice
Game 3 (expert) is the hardest difficulty and was chosen here. At Game 3, the manual claims the computer will take 2-4 minutes per move, with an entire game lasting 2+ hours. Here are the stats on the computer's thinking time during this run:
  • Total: 11 minutes, 42.36 seconds
  • Average: 36.97 seconds
  • Maximum: 2 minutes, 3.20 seconds
  • Minimum: 0.22 seconds
16.104-second improvement and new cursor tech This improvement was possible thanks to new cursor tech being found. If you input Up+Left+Right or Down+Left+Right, this will cause the cursor to move two squares to the right. Normally, the cursor can only be moved in single diagonal increments so this can essentially be used as a "double cursor move" in certain situations. What's even more curious is when these inputs are performed on one of the rightmost squares on the board, the cursor will wrap to the leftmost square one row above, which can save several cursor movements. Each cursor movement saved equals 3 frames saved. In all, a cursor wrap was added to move 6 which saved 6 frames and a double-right was added to move 13 which saved 3 frames. On a final note, I was unable to find an analogous strat which made the cursor move two units left or wrapped from the left to right side of the board, so no timesaves in that department.
One of the consequnces of saving these 9 frames is that it changed the RNG, leading the computer to do something different on move 13. It moved 20-24, which isn't surprising as this was the computer's 18th move in the previous TAS. What really surprised me was the computer's next move. Instead of going 2-6 like I expected, the computer went 13-17 which is not only a horrible move, but also an amazing RNG pattern which I failed to find when making the previous TAS. It significantly simplifies the rest of the game, as all remaining moves the computer must ponder are now more forcing, causing the computer to think less. It also reduced the total number of moves played in the game from 21 to 19, leading to about a 16-second time cut.
Routing
Routing a game like this where the main thing to optimize is the computer thinking time is particularly difficult, as the "best" move is seldom the one that'll save you the most time. Lots of trial and error will be required. There are two good rules to follow in general however, which are:
  • Don't make the position complicated. More complicated positions and more move options greatly increase AI thinking time.
  • End the game in the least moves possible that does not complicate positioning. In the case of checkers, that would be done by either trading many pieces as early as possible, or trapping an opponent's pieces.
Well, trading a whole bunch of pieces early turns out to not work in our favor. Once all the front and most of the second-row pieces are traded, the AI begins to think for a VERY long time as the large separation between the two opponents' pieces means suddenly there's an extraordinary number of outcomes to consider.
In this TAS, we used a trapping strategy. What's effective about this is you're progressively reducing the complexity of the position as AI pieces become trapped (options to calculate for them are basically nil) and the player's dominance of the board becomes stronger. It will also ultimately reduce the total number of moves played in the game. The main thing to watch out for will be to not let the AI undermine your trap from the other side!
RNG
There is some randomness to computer moves when more than one move is found to be about the best, and will choose one of up to two different moves which can change based on when the player's previous move is confirmed.
Game Summary
MoveWhite Red Load TimeCommentary
1 22-18 11-161:02.20 Old Faithful opening. Out of all the opening moves, black thinks for the least amount of time
2 18-14 10-170:00.36 Getting rid of some nuisance red pieces that would mess up our trap later
3 21-14 9-18 0:03.19 Forced captures also usually means short thinking times for red
4 23-14 8-11 1:09.22 In the end, white has control of the center and now we can go for our trap
5 26-22 16-201:43.53 Many options here. This move forces red to not think as much and traps a checker on the right
6 22-18 6-9 1:23.01 Part of the trap complete... looks like red is trying to go around the left side...
7 25-22 9-13 2:03.20 Let's make sure red doesn't try to slip by.
8 30-26 11-160:59.01 Many options here. This move forces red to not think as much and clump checkers on the right
9 26-23 7-11 1:06.35 A diagonal of 5 white checkers; now we can move in front of these without getting jumped over
10 24-19 1-6 1:12.13 Time to get moving along the diagonal highway!
11 19-15 3-8 0:27.07 Key losing move for red. Now if our planned trap unfolds, red will have 5 trapped checkers!
12 28-24 6-9 0:14.54 Our trap will be set with the next move. Time to focus on the left side of the board
13 24-19 20-240:05.69 But first, red decides to make a break for it on the right. There will be no chance of that!
14 27-20 13-170:04.97 Red's move was a bit surprising to me. In the previous TAS, the computer moved 2-6 but this time moves 13-17 which is SIGNIFICANTLY worse as we are immediately forced to do a double jump.
15 22-13-62-9 0:02.87 It is now a win for white no matter what red does
16 29-25 9-13 0:03.92 Putting a checker on position 25 ensures the red checker on position 13 will not slip by
17 25-22 5-9 0:00.43 Red runs into a forced capture
18 14-5 13-170:00.45 Red runs into a forced capture
19 22-13 0:00.22 A 9-5 win on "expert"!

nymx: Claiming for judging.
nymx: Good work on the cursor movement exploitation. As with your "Game 1" run, this run will now exist as a "Game 3" branch.
Note to publisher: Please fix the branches to reflect the category being played...as this game will be running with two branches now. Thanks.

despoa: Processing...
Last Edited by despoa on 8/22/2023 9:12 AM
Page History Latest diff List referrers