This has been sitting in my user files for a while now. Figured I would submit it. Worst that can happen is that it gets rejected, right?
Objectives
Fastest completion
Heavy luck manipulation
Completes one maze
(Technically) two players
Emulator used: BizHawk 2.10 (Stella core)
Notes
Maze Craze is a simple two-player Atari game. In this game, you have to find the exit in a randomly generated maze. This game has a boatload of customizable options, including adding enemies, changing the maze's visibility, or even placing fake walls. It's worth mentioning that you can increase the player's speed, but this takes way too long to do, so it's not done here.
When powering on the game, the game randomly chooses between 16 starting seeds. By adding power inputs during the first few frames of the run, we can choose our starting seed. After this, the seed is updated every frame. When we reset, the game uses that seed to generate a new maze. The objective of this TAS is to generate the shortest possible maze in the fewest amount of frames.
Quoting Deppy
"I reverse-engineered the maze generation algorithm so that we could search for mazes programatically. We found that pressing the power button 8 times at the start gets us to seed 0x0f00. From here, resetting at frame 95 generates a maze with the seed 0x1b05. This maze can be solved in only 23 steps."
The program Deppy used searched for the top 100 shortest mazes and how long it takes to reach them from each possible starting seed. We believe that this is the fastest one. The source code for this program is available here.
Additional Notes
The improvement for this submission can be found here. I switched to BizHawk 2.10 with the Stella core because it emulates the loading more accurately. In past TASes, I was able to press reset while the game was generating a maze, which simply isn't possible on actual hardware.
Out of the 65 536 possible mazes that can be generated, only 2 of them have the minimal possible length of 19 steps. These unfortunately take a long time to generate. You could potentially do more resets to generate these mazes faster, but at that point, they're not viable. Doing more than one reset is too slow. If you do want to see this however, I made another movie for fun that you can see here.
It's also worth mentioning that this TAS is slower than RTA runs. In RTA runs, people can just reset until they generate a viable maze, since time starts on first input instead.
Special Thanks
I would like to give a really big special thanks to Deppy. Without her work reverse-engineering the game, this TAS would not have been possible.
nymx: Ok, I see that the new files are not getting us anywhere. This situation is weird, because to get the ultimate pattern, you would have to reset to until you get what you want. Because we don't know how long the speed runners perform these resets, this run will suffice. TAS timing starts at power on, but the RTA runs don't. The actual IGT gameplay, may be longer than the human effort...but the overall effort certainly is going to beat a human. I'm just disappointed that desired patterned couldn't be achieved. I'm turning down all the files submitted after my original request...seeing that they all get longer in length.
Accepting.
nymx: Replacing with the 323 version, made with the Stella core.
Because we don't know how much time a "speed runner" takes in producing the best maze, we have nothing to compare against. In my opinion, that kind of human run should be clocked from the very start of the effort. If that was done, then we could certainly solidify a judgement on this. Instead, I'm accepting this because it appears to be out of reach for any human to find the results of this movie. Yes...the IGT part is slower, but this TAS is way above human capabilities.
Joined: 11/14/2014
Posts: 959
Location: South Pole, True Land Down Under
Greetings Walgrey. As you have stated, this run is in fact slower than the WR....which can be found here:
Link to video.
Since this run is so short, I'm extending the opportunity for you to further experiment and see if a faster result can be found. Because the WR times it from the start of the race...we may be looking at a rare situation here that needs to be handled differently. But first...please see if you can find a situation that can be accept without community disagreements.
I'm placing this submission on Delayed, pending your response and/or further optimization.
I recently discovered that if you haven't reached a level of frustration with TASing any game, then you haven't done your due diligence.
----
SOYZA: Are you playing a game?
NYMX: I'm not playing a game, I'm TASing.
SOYZA: Oh...so its not a game...Its for real?
----
Anybody got a Quantum computer I can borrow for 20 minutes?
Nevermind...eien's 64 core machine will do. :)
----
BOTing will be the end of all games. --NYMX
Joined: 11/14/2014
Posts: 959
Location: South Pole, True Land Down Under
Walgrey wrote:
Hello! I have a friend who is looking into this run and helping me with it. Will update if anything comes up!
Curious if you have made any progress on this. FYI...we don't typically allow a status of "Delayed" for long periods of time. If we see progress, usually a month is the most we can offer.
I recently discovered that if you haven't reached a level of frustration with TASing any game, then you haven't done your due diligence.
----
SOYZA: Are you playing a game?
NYMX: I'm not playing a game, I'm TASing.
SOYZA: Oh...so its not a game...Its for real?
----
Anybody got a Quantum computer I can borrow for 20 minutes?
Nevermind...eien's 64 core machine will do. :)
----
BOTing will be the end of all games. --NYMX
I've managed to find one small improvement that saved 28 frames over the submission.
Shortly after, BizHawk 2.10 released with the Stella core. I realized that Stella is way more accurate with its loading. If you play the game on an actual Atari2600, resetting will cause the game to flash colours and take a longer time to generate a new maze. In my previous Atari2600Hawk TASes, I took advantage of being able to reset multiple times in succession to get a good seed quickly, which simply isn't possible on actual hardware.
So I ended up making a TAS that uses the Stella core. The longer loading times meant that you can't advance the seed a lot before losing time. I'm of the personal opinion that the Stella TAS should be used due to its accuracy, but I don't know if that's my decision to make.
I asked a friend to look into the game and so far, she managed to find where the maze is stored in RAM and the RNG algorithm. She's still working on it, and it'll unfortunately take a while (likely a month), and there's a possibility that we might not even find anything better.
I don't think I've been very successful in making a run that's nearly as fast as the RTA WR and for that I'd like to apologize.
Greetings Walgrey. As you have stated, this run is in fact slower than the WR....which can be found here: https://youtu.be/k4VwTVoWFlc.
Since this run is so short, I'm extending the opportunity for you to further experiment and see if a faster result can be found. Because the WR times it from the start of the race...we may be looking at a rare situation here that needs to be handled differently. But first...please see if you can find a situation that can be accept without community disagreements.
I'm placing this submission on Delayed, pending your response and/or further optimization.
Who knows how many resets are necessary (and when they are necessary) to get to that particular maze layout. In my opinion, this is a situation where we shouldn't hold a TASer to beat a human WR when they have different timing procedures. If the human run was timed from power-on instead of race start and had to include however many resets/time delay was necessary to produce that maze, it would likely be slower than this submission.
Hi! I'd like to apologize for the long wait. I wasn't expecting this to happen.
I am happy to report that Deppy and I have found an improvement (although this uses the Stella core and still doesn't beat RTA).
Is it even possible for this game to generate a maze like in the BenS7 WR? Maybe the A2600Hawk core could do that, but then it could be an emulation error. Also, can you both share the program that you made?
Is it even possible for this game to generate a maze like in the BenS7 WR? Maybe the A2600Hawk core could do that, but then it could be an emulation error. Also, can you both share the program that you made?
I have a user file that does an even shorter maze.
Deppy will upload the program to her GitHub soon. I'll update the description when it's ready.
I'm turning down all the files submitted after my original request...seeing that they all get longer in length.
Can I ask why? The file I submitted doesn’t use Stella and takes advantage of being able to reset where you shouldn’t be able to. I’ve explained the gist of this in the user files and description. I’m just confused as to why the original file was chosen since it’s technically not possible to do on actual hardware.
Joined: 11/14/2014
Posts: 959
Location: South Pole, True Land Down Under
Walgrey wrote:
I'm turning down all the files submitted after my original request...seeing that they all get longer in length.
Can I ask why? The file I submitted doesn’t use Stella and takes advantage of being able to reset where you shouldn’t be able to. I’ve explained the gist of this in the user files and description. I’m just confused as to why the original file was chosen since it’s technically not possible to do on actual hardware.
Ok. I have obviously missed a detail here. Let me check this when I get back home.
I recently discovered that if you haven't reached a level of frustration with TASing any game, then you haven't done your due diligence.
----
SOYZA: Are you playing a game?
NYMX: I'm not playing a game, I'm TASing.
SOYZA: Oh...so its not a game...Its for real?
----
Anybody got a Quantum computer I can borrow for 20 minutes?
Nevermind...eien's 64 core machine will do. :)
----
BOTing will be the end of all games. --NYMX
Joined: 11/14/2014
Posts: 959
Location: South Pole, True Land Down Under
Ok...let's review this:
Your original submission (non-stella): 231 frames
*This version is the shortest TAS time. At first glance, it appears that it could not be accepted; however, after realizing how a good maze is produced by RTA runners...this doesn't seem that bad at all. We simply have no measurement on the length of time it takes for them to produce the maze they need for a fast solve. If I made the rules for that run, I would start the timer at power on and reset until I found it. But that is not the case here.
First WIP (non-stella): 203 frames
*This is the kind of improvement that I was originally looking for. During this time, I still hadn't got a grasp of this game's uniqueness to treat it differently. Plus, I think the upcoming "WIPs" confused me and overshadowed this one.
Second WIP (Stella): 414 frames
*Yes...this is longer, but not necessarily a problem. Having a more accurate core is certainly desirable, but in this case is not required. We have many cases of this right now with other consoles. For instance, QuickNES. In my experiences, I have chosen a slower operating and more accurate core for TASing Tetris 2, even though QuickNES would have cut my TASing time down to less than a year from the 2 years and 4 months that it took me. So we think alike in that regard. Because this version doesn't look sloppy, this would certainly override a less accurate core.
Third WIP (Stella): 323 Frames
*Excellent improvement. Doesn't beat the non-stella version though, nor the IGT part of the human speedrun...but it doesn't matter. This is the accurate core of the bunch. Plus...let's remember the reset time that we can't compare against, done by runners.
Fouth WIP (Stella): 44748 Frames
*I get the point that we are looking for IGT results here, but the entire run is way too long and certainly has the potential to beaten, if other techniques are found. Its been a while since I've seen a submission that strictly went for IGT, as most movies are now TAS timing. I am not considering this version, but it is mice to have though to see what is possible.
Now I have to convey, that I've not been too organized since BizHawk 2.10 has been released. I've basically had to move around to other machines, since the one that I primarily use is a Windows 7 machine...which 2.10 is not functioning for me. That, plus two other TASes have gotten me mixed up a bit.
Out of the entire list here, I can upload the Third iteration. This includes the fastest version for the most accurate emulation. But do note...it is not an automatic rejection for using a lesser accurate core.
Do you agree to using the 3rd WIP (323 frame version)?
I recently discovered that if you haven't reached a level of frustration with TASing any game, then you haven't done your due diligence.
----
SOYZA: Are you playing a game?
NYMX: I'm not playing a game, I'm TASing.
SOYZA: Oh...so its not a game...Its for real?
----
Anybody got a Quantum computer I can borrow for 20 minutes?
Nevermind...eien's 64 core machine will do. :)
----
BOTing will be the end of all games. --NYMX
Yes, we agree to using the 323f TAS.
One last thing to note, Deppy found a faster way to get the shortest maze. It's still 43 seconds, but I figured it was worth mentioning here.
https://tasvideos.org/UserFiles/Info/638733772675487075
