Introduction

This is my third attempt at producing an acceptable TAS for NES Pac-Man.
A few days ago I submitted a movie #8149: eien86 & mscroggs's NES Pac-Man (Namco) "single level" in 00:44.18 that solves a single level of Pac-Man, under the impression it was the fastest route. I was wrong since Randil had already produced a much faster movie with a better route. Then I took that route and transcribe the pellet order into JaffarPlus to solve the level by having a reward function (r) with the following shape:
r = #PelletsInPathEatenInOrder * 1000.0 - distanceToNextPellet
The PelletsInPathEatenInOrder rewards the bot for every pellet eaten in the optimal path, and not for any other pellet which is out of order in the optimal route. distanceToNextPellet is the manhattan distance from the center of pacman and the position of the next pellet in the route. The result was a movie that's 20 frames faster.
The movie described above was submitted in #8152: eien86's NES Pac-Man (Tengen) "single level" in 00:36.36 and received, just like the first one, feedback regarding its unacceptability. This was because, although the game has no ending, the movie solved only a part of all what the game can offer. Indeed, playing more levels (forgot how many) will bring a set of new challenges as both the Pac-Man and ghosts get faster and the effect of magic pellets get shorter (a metaphor for developing drug tolerance?). Not only does the difficulty ramp up, but also the bonus items change. The last item -- a key, of all things in God's good earth -- repeats indefinitely.
The movie hereby attached incorporates the exact inputs of the previous one, but plays further. And it does so as fast as computerly possible until the very last noticeable change in the game happens: all keys are shown in the screen. And then it beats that level. Nothing new remains to discover, except of course, an eventual score overflow.
I hope this movie now has more chances to be accepted. I am in favor of adding Randil as co-author, if they and the judge considers it fair, since I based the overall route on his previous work.

Comparison Movie

The movie below shows the difference for the first level only between the movie by Randil and mine.

Software + Hardware

Rom Information

Rom: Pac-Man (Tengen)
  • headerless rom hash: SHA1:A34E68372082513209A795786C8EEA493CC2CD14
  • headerless rom hash: MD5:C4AE6CC4E981A8316429572409018DC8
  • PRG (8KB) + CHR hash: SHA1:96C8879B8F0C70803A5F7754D53F8BD60F016163
  • PRG (8KB) + CHR hash: MD5:E6EE06910B926DB6A19F1F1C3A3EB21D

Emulator

  • EmuHawk 2.8.0 (Core: NesHawk)

Routing Bot

  • Bot: JaffarPlus
  • Routing Core: QuickNES
  • Platform: 'The Jaffanator' - AMD Ryzen Threadripper 3990X (64 cores, 128 threads) + 256Gb RAM (Average Exploration Performance: 1.35M States/s)
Re-record count is 53472478070

Darkman425: Claiming for judging.
Darkman425: I did a little legwork with checking. While I was able to use the final board input done with the one syncing without modification, doing that again after had a desync where Pac-Man eats Pinky. That led me to do some RAM checking.
In the RAM the address that tracks how long a power pellet lasts, a big indicator for difficulty, is at 0x008C. It starts at 7 and eventually goes down to 0 with some occasional increases in between for reasons only known to Tengen. At 7 the power pellet lasts 480 frames which would indicate the lowest difficulty. At 0 the power pellet lasts only 30 frames and doesn't seem to go lower, indicating maximum difficulty. At the last round done in this submission, the round before the last one, as well as the two rounds immediately after, the address is at 0 so that would be at maximum difficulty.
Thus the reason that I came to with the desync on input reuse comes down to a simple matter of slight RNG variance. As this submission has a clear and visible goal of clearing a round with all keys on the score area and with RAM watching revealing that the power pellets are at their weakest, I come to the conclusion that this exhausts all unique content.
Accepting to Standard for full completion.
Spikestuff: Wakka Wakka.


TASVideoAgent
They/Them
Moderator
Joined: 8/3/2004
Posts: 15628
Location: 127.0.0.1
This topic is for the purpose of discussing #8155: eien86's NES Pac-Man (Tengen) in 12:02.86
Spikestuff
They/Them
Editor, Publisher, Expert player (2656)
Joined: 10/12/2011
Posts: 6449
Location: The land down under.
That's what I'm talking about! How the shit does your input sequence get that damn lucky to be that repeatable, like the hell? I'll claim this one for pubs when accepted due to the previous submissions.
WebNations/Sabih wrote:
+fsvgm777 never censoring anything.
Disables Comments and Ratings for the YouTube account. Something better for yourself and also others.
eien86
He/Him
Judge, Skilled player (1881)
Joined: 3/21/2021
Posts: 262
Location: Switzerland
Spikestuff wrote:
How the shit does your input sequence get that damn lucky to be that repeatable, like the hell?
Ah, I need to clarify this. The input is the same as my previous submission only for the first level. But then every other level after has a different set of inputs; I had to run the bot for every one them.
Player (13)
Joined: 6/17/2006
Posts: 510
I was about to comment that there was at least one dot sometimes eaten in a different order on some levels. (The 3rd one from the top in the 2nd column.) Very subtle. Great work! I would like to see if something similar could be done in the arcade version too!
Patashu
He/Him
Joined: 10/2/2005
Posts: 4045
A lot of fun near misses in this one, and a human would have no shot at replicating this due to all the frame perfect inputs. I'd also love to see the Arcade version given a similar try. (MAMEJaffar sounds awful to set up...)
My Chiptune music, made in Famitracker: http://soundcloud.com/patashu My twitch. I stream mostly shmups & rhythm games http://twitch.tv/patashu My youtube, again shmups and rhythm games and misc stuff: http://youtube.com/user/patashu
PLANET
He/Him
Joined: 1/3/2018
Posts: 74
Oh, this is nice! I actually enjoyed both of the first levels when they were submitted, but this takes the cake, thank you for going the longer way to make the effort : ) Yes vote obviously : )
Alyosha
He/Him
Editor, Emulator Coder, Expert player (3827)
Joined: 11/30/2014
Posts: 2834
Location: US
Another game tackled with Jaffar, cool stuff. I'm kind of surprised the reward function was able to be so simple, but I guess it helps it was starting from Randil's route already.
Joined: 7/7/2017
Posts: 28
enjoyed watching all the near-misses.
eien86
He/Him
Judge, Skilled player (1881)
Joined: 3/21/2021
Posts: 262
Location: Switzerland
Post subject: Movie published
TASVideoAgent
They/Them
Moderator
Joined: 8/3/2004
Posts: 15628
Location: 127.0.0.1
This movie has been published. The posts before this message apply to the submission, and posts after this message apply to the published movie. ---- [5231] NES Pac-Man (Tengen) by eien86 in 12:02.86