Actions
Engagement
  • Author Author: tariq.ahmad
  • Date Created: Date Created
  • Last Updated Last Updated:
  • Views 1061 views
  • Likes 3 likes
  • Comments 7 comments
Related
Recommended

Atari Dragster World Record - Part 2: Testing with Software -- Episode 321

image

element14's The Ben Heck Show

Join the Ben Heck team every week for amazing hacks! Watch them build and mod community-inspired projects using electronics!

Back to The Ben Heck Show homepage image

The Learning Circuit
sudo Sergeant
See All Episodes

 

 

Ben is joined by Todd Rogers, owner of the longest held video game world record, whose legacy has been put into jeopardy by statistical models that show that the best possible score is 5.57, not the 5.51 that Todd recorded. Adding intrigue to the story, Todd not only submitted a Polaroid picture in 1982 of the 5:51 that’s in question.  He's also performed a 5:51 score at the 1982 Consumer Electronics Show in Chicago followed by another live performance of a 5:51 score at the Electronics Things Show in Detroit. 

 

 

Episode 317: Atari Dragster World Record - Part 1: Building the Hardware

 

There hasn't been a man vs. machine battle like this since Luke turned off his targeting computer and trusted his feelings to hit a pinpoint target that blew up the Death Star. This isn't the first time that Todd Rogers 5.51 score has challenged computer models.  After he submitted his Polaroid picture of his 5.51 score in 1982, Activision, who released Dragster as their first title and the first third party game ever, called him wanting to know how he shifted his Dragster comparative to their computer model.  He was then invited to the 1982 Consumer Electronics Show in Chicago in June of 1982 where he performed his second 5.51 score in front of a live audience. They then invited him to the Electronic Things Show in Detroit where he again, for the second time in a live event, performed another 5:51 score.

In a previous episode, Ben made a hardware device that hooks up to a real Atari and allows you to look at the RAM and make decisions based off it.  Today, they’re going to use Todd’s information and plug it into Ben’s system so it can repeat the kind of gameplay he did back in the 1980s, so if they give it the numbers to try it will do it every time it can always start on the right frame, a human only has a 1/16 chance of doing that, and it doesn’t get tired or have cramped hands, making it the ultimate video game playing machine. Ben goes over the basis of his analysis system.  He’s programming a PIC32 using Microchip’s MPLABK IDE. Whenever the RAM is written to an interrupt is called on the PIC32 where it copies the contents of the RAM buffer into memory so it copies RAM into RAM.   The other thing that happens is it looks for a change in the frame number.  The second RAM position is the current frame counter, so if they see that number has changed from the last version of itself, they do all of their logic code.   The next thing he looks at is disassembly in the stella emulator.  They’re going to trace the frame counter and count how many machine cycles occur.  He then takes out an oscilloscope to show you what that looks like.

The main loop of the program cycles around to check whether the RAM changed.  That’s always going, but when there is an interrupt with the RAM, when there is new RAM, the time between the RAM changing and the program actually saying it should analyze the RAM, that can change.  Now that he knows he’s analyzing things with time to spare, he can proceed with his coding knowing that fact. Ben’s code looks for a change in the frame counter and that’s when it does all its logic. The frame counter and the player count are actually inverse of each other.  The player count is even while the frame number is odd.  This is important because you want to push the reset button at just the right time so that player 2, which is player 1 in code, has the first time increment, not player 1.  Now that he knows how that works he can be sure to push reset at the right time, so that the player that they are trying to test has the best chance of getting the best time. The system is doing 5.61 over and over, so Ben is going to put a button menu system onto his board set.

Ben takes all the single, simple variables and puts them into a menu system that will appear over the serial terminal, so he can make tweaks without having to flash the code every time.  He diagrams the menu on a board.  This will allow them to dial in the functions even faster than they are now. Todd sits down with Ben to advise him on how to make adjustments to gameplay in order to get 5.51.  As they inch closer to the number, it's time to test Omnigamer’s spreadsheet together.  Omnigamer is the speedrunner who theorized that a 5.51 may not be possible.

 

Disclaimer

  • in reply to fbx

    I definitely understand your view.  That said I think there’s a danger in holding up Case A as proof of Case B.  Each case needs to stand on its own IMO.

     

    There is definitely clear indication that a Part 3 is coming.  I think a conclusive conclusion as to what is possible on Dragster will come soon enough.

  • in reply to pacmantab

    @ pacmantab,

     

    My problem with Todd and why I don't trust him on this or anything else was after the Barnstorming debacle, where his record on that stood for decades on TG without ever notifying them that they didn't have the record correctly entered. Even when made aware I was challenging the validity of it, he continued to 'go along' with the record being legit. It wasn't until I proved that it was impossible that FINALLY it got taken down after the TG ref covered for Todd by confessing he never witnessed it and that it was merely a clerical error. Then later more than just myself got concerned over his Atari Donkey Kong score of 15 million. There were actual videos he had taken of that, but I wasn't allowed to view them. Another ref viewed them and decided there was shenanigans in the videos themselves, and actually took the record down as a result. When that happened, Todd's fans gave that ref a REALLY hard time, accusing him of being a jealous hack and other such nonsense. Still another impossible record he got was on skyjinks, where in order to get a time score on it, you have to bank left and right to clear markers. I demonstrated with stopwatch that a run where I intentionally only banked 10 times compared to Todd's 'official' 17 bank record was SLOWER. That defies physics for a 10 bank flight to be slower than Todd's 17. So you can see, it's not just this one controversial subject of Dragster with Todd. There's a laundry list on several games. A lot of people have suggested it's because Todd has unique or modified hardware, but as you know, that would still invalidate all his records because it's not a fair playing field.

  • Point well taken.  I remember seeing Yee on Starcade - that arcade game show - where he made note of his score.  That said Todd being willing to come on Ben’s show may suggest there was not, nor now, any intentional shenanigans.  He could be just as eager to get to the bottom of this as anyone - regardless of where the trail leads.  In addition, not every player gets ”hearsay” from the actual creators of the game.  Would there have been the chance of a favorable hardware glitch?  Possibly.  Whether any glitch was known at the time is another question.

     

    To the topic at hand; I had another suggestion for a test, which expands on my previous suggestion:  Would it be possible to run Omnigamer’s simulation with Players One and Two simultaneously?  It would be interesting to see if it always works out to a tie....

  • In regards to the claim he got this record three times in 1982, hearsay doesn't count these days, even with Twin Galaxies. This is because of hardware hacking/glitching, that can go undetected on a live audience. For example, if Todd had an unfair advantage with his hardware, all he'd need do is simply take it with him to these shows, bust the record in front of a live audience, and nobody would question it. Now that things can be scrutinized in detail, Todd runs the risk of being caught red-handed exploiting a glitch or hardware variance (neither of which are allowed for logging a record in a given category). So Todd's only recourse is to never submit himself to scrutiny on a new record. You gotta remember that 1982 was the same era of Jeff Yee getting "over 5 million" on Pac Man, which netted him TV show guest appearances, expensive gifts, and even acknowledgement from the President of the US. It was only discovered some years later that Jeff had been flat-out lying after it was determined a perfect score is over 3 million (not 5 million). Here again, witnesses vouched for Jeff.

  • If you can get the button mapping onto the PIC automatically, this would work well with a Genetic Algorithm. I know that sounds like deep AI voodoo, but they're pretty simple. The goal is to let the machine figure out how to get the best time and then see if you ever get anything better than 5.57.

     

    * Create a series of bytes which maps which buttons to press at a specific frame of time

    * Randomly create 100 such byte strings. You can also feed in the different 5.57 times from Omnigamer’s spreadsheet.

    * Have it automatically run each of those strings. Grab the final time out of the Atari's memory.

    * Throw out 70% of the worst times

    * Create new set of 100 strings out of the ones that remain. There are many different approaches here, and it's where things get interesting. For instance:

    ** Take two strings and cross them. For instance, the first element would be taken from the first element of string A, then the second from string B, the third from string A, and so on

    ** Take the first half of string A and the second half of string B

    ** Mutate by randomly changing some of the bytes on a string

    ** Randomly create entirely new strings

    * Now run the next generation. Repeat until satisfied.

     

    There's lots of parameters to tweak here. Perhaps you throw out 50% of the worst times rather than 70%. Maybe you take an "elitist" approach, where the top 5% are left unchanged for the next generation.

     

    Then turn it on, let it run, and leave for the night. Might even take several days or a week.

  • Nice episode.

     

    I liked the look at the timing with relationship to the code.

    Those of us who did real time programming back in the day spent a lot of time doing that kind of analysis.

     

    DAB

  • Just curious.....what would happen if you fed the inputs into Player two?