This is probably the game I would like to see a TAS of most of all games. This new (?) technique could make the TAS even more spectacular than I ever could imagine.
yep, I knew about that technique - and had factored it into my routes.
I've pretty much given up on the bot idea - it just wasn't beating the best non-tool assisted times.
Instead I'm going to try to make a Gimp (like photoshop for linux) plug-in so that I can draw the moves it should do (green for left, red for right) and have it calculate and draw the path it's taking over the map - so I can see the route and adjust it from above - with the bot taking care of easily automated things like when to hit accelerate, charging the 'boost counter', hopping the moment before hitting dirt and luck manipulation.
I have made a gimp-plugin before, so I'm sure this is possible.
ETA: some time next year (or the year after that...) :P
that's a bit sad Huffers, but stay assured you have our support to keep working on this great project, whenever you will have enough free time to do it. reading your (programming) efforts is highly interesting, and I hope that you will produce a publishable movie in a near future.
anyway, were your best times that far from the WRs ? could you just please present some of your results ?
I never sleep, 'cause sleep is the cousin of death - NAS
Joined: 6/13/2006
Posts: 3300
Location: Massachussetts, USA
As you wish.
Super Mario Kart, 150cc, Bowser, Mario Circuit 1 complete!150cc SRAM for those that don't have it.
Hey all! This is a joint effort by me and Huffers. No bots were harmed (used) in the making of this course. Human TASing FTW!!!!
I know that everyone has been desperately waiting; the wait is over!
Lap times:
L1: 9"46
L2: 8"36
L3: 8"35
L4: 8"25
L5" 7"64
Total time: 42"06
Ok. time for a great deal of explaining.
After a few days of noodling around with the memory addresses and physics of this game, I started working on it. I got stuck on the second mushroom, I asked Huffers for help, and before you know it we're co-authoring this. we'd be sending back .smvs back and forth like every 30 frames or so. the switching off between playing happens so much in this course that not even I can tell who's behind the wheel. We went through hell and beyond to figure out all the tiny physics of this (most notably the probability of the mushrooms; see below), and we're looking forward to completing this run. This game has sat alone for too long, and the both of us are really motivated with it, especially since we completely raped the first course.
Here's the logistics of everything you'll see.
Mini-boosts were used a lot. Charging them up both on turns and on straights, by alternating left/right. they proved really helpful in gaining that extra bit of speed, and the invaluable driving over dirt.
Long boosts: the meat and potatoes of the speed. the first part is the mushrooms.
In our searching around,w e found a memory address that shows the value of the item in the box; the instant that the box turns red. we suspected that items in boxes are predetermined, and indeed they are. the items are 100% dependent on what position in the race you are (of course we were always first) and the in-game timer on the clock. landing one frame later did not matter, it was purely the timer. we extensively tested this, and it vastly affected the run. much of the difficulty came in manipulating the mushrooms. the odds, based on real time 2/32 (which Huffers thinks is crap) to get 5 mushrooms in 5 laps like we did is 1/1,048,576. It was even tougher when TASing, because you have to either slow down or do something faster in order to get the perfect timer number when a mushroom would appear. "Mushroom Famine" as we dubbed it. that's why the 5 courses all have vastly different times.
We considered using Toad, because it aesthetically looked like he goes faster and he can make tighter turns, but our eyes were playing tricks on us. He is slower, his boost is no faster, and we can deal with Bowser's inferior turning.
The luck manipulation to get mushrooms is so tough, it explains some of the "sloppy" movement, i.e. not driving perfectly around corners and slowing down sometimes. and not landing on the first possible box. EVERYTHING was done for a purpose.
My test runs had me long boosting off of the wall in the youtube video. This was a bad idea, because while I could mini boost across the dirt, I was actually going too fast as I got towards the item boxes, meaning that manipulation would become near impossible. I couldn't test every box/in-game time. so we boost off of the pair of pipes :-). great care was taken to just barely bump the pipes to not slow down at all.
holy hell did we go past the other karts. poor Koopa was lapped three times. most of the others were lapped twice or more.
we've also found a movie with infinite boosts in certain tracks, which we will utilize in the future. it requires the yellow speed strips, so expect that in future courses.
This was very difficult to optimize, and very painstaking in getting those mushrooms. We hope that our WIP is exactly what everyone was hoping it to be.
Enjoy! you all were very patient in waiting on progress; you deserve this lovely minute of kart madness.
If anyone has questions about anything, please feel free to ask.
PS:
how do you change the author title in the movie file? the name of the file is correct, but in the Snes9x select movie thingy it only lists me as the author. is there a way to change that?
We've also been discussing using soft resets in order to shorten the time. We're planning to reset at the first possible frame to skip the receiving trophy cinematic. does this work/feasible/good idea? I still have my record reset patch from my DKC2 102% run.
edit: oh, and Huffers: beat your 42:18! got ending time of 42:06.
I just watched your wip. Great job Comicalflop and Huffers!! I don't know the game physics any more than the next guy, so I doubt I can really help. I can offer moral support, though :D.
Wow, that was very impressive. I did have a few questions about inconsistent turning but I suppose it was due to the luck manipulation on the mushrooms.
I think Maximus is busy adding this feature to his movie editor, but if it won't be completed in time, you could always make a new movie with the correct description and paste the entire movie in that new movie (although his does reset the rerecord count).
In any case, I'm looking forward to the final result :).
Joined: 6/13/2006
Posts: 3300
Location: Massachussetts, USA
yes, luck manipulation of mushrooms dictated all our movement. We're kind of slaves to the game this way, and not the other way around, which is unfortunate. the odds for getting feathers are better than mushrooms, so that'll help for the feather routes. and the courses with speed strips makes things so much easier.
Maximus will probably be done before us; there's 19 more courses to go, some easy lap skipping, however most very difficult. and if that doesn't work, I can probably copy and paste the input into a new movie file, and spend a few hours pressing rerecords keys like a madman until I reach what the actual rerecord # is. or it can be hexed into the file. there's ways I'm sure.
Wow, that was very impressive. I did have a few questions about inconsistent turning but I suppose it was due to the luck manipulation on the mushrooms.
It did look a bit sloppy, but if we'd gotten there any earlier we wouldn't have got a mushroom...
By the way, the chance of getting 5 mushrooms in a row like that is exactly one in 1,048,576.
We beat the world record non-tool assisted time (45"94 by Florent Lecoanet) by 3.88 seconds.
Excellent work. I'm really looking forward to you blazing through the rest of the tracks. (Particularly eager to see the double bounce shortcut in GV1 done 5/5, and what the greatest victory margin will be.)
The Mushrooom manipulation issue explains the wider cornering during the first-lap longboost, I suppose. Unfortunate that it looks like it can be improved. It's so close to being sub-42, though.
I'd really like to see VL1 without feather abuse, but that's for novelty reasons more than anything else.
Keep it up!
Andreas
Joined: 6/13/2006
Posts: 3300
Location: Massachussetts, USA
Huffers and I have been discussing, and we're going to split up to test something out. He's going to continue on Donut PLains 1 with Bowser, and I'm redoing MC1 with Toad to see if he is a superior choice, due to better handling and potential better boosting. I've discovered a source that says Toad has a faster extended boost, and that Bowser has the slowest, so maybe my eyes weren't deceiving me. I also have a sneaky suspicion that the long boost that I did in the final lap, where I activated the mushroom earlier instead of right in front of the pipe, could be the fastest way of doing it, so I'm heading back to verify that.
don't know if this means that MC1 will be completely redone with Toad or not. but this is a good case where having two authors lets us divert to find this stuff out. I'll post the results on Toad usage and if we'll switch to using him or sticking with Bowser; and expect Huffers to keep going with Bowser in DP1 to work out the kinks of the lap skips.
edit: Huffers got 24:94 on DP1, the WR is 57"98. He's going to try and improve it, and I'm going to give Toad a go tonight.
Joined: 6/13/2006
Posts: 3300
Location: Massachussetts, USA
yes. all 4 cups, 20 races. we might record resets to skip the cinematics of receiving the trophy at the end of each cup, although if worse comes to worse that cutscene is not as painful as the MK64 one.
and I've decided to throw in input animations going to the music in between the races.
Joined: 6/13/2006
Posts: 3300
Location: Massachussetts, USA
Edit: gah, beat me by 18 seconds.
I always suspected that hopping to turn tight corners works in SMK just like in MK64, and it did hold true. I'm curious if Toad can do this faster, but first I have to see if he even should be used.... hm, I should get on that. =D