FreeGameDev

Hello,

I worked on a TAS for SuperTux ( https://en.wikipedia.org/wiki/Tool-assisted_speedrun ).
I explained with more details here: http://www.speedrun.com/st/thread/hcxzk/1#bi5so
Video of the first unoptimised version here: http://mcmic.haxx.es/speedrun/supertux_tas_sub18.webm

Code and record files here: https://github.com/MCMic/supertux/ (record files are in data/TAS/levels/world1 )

Do not hesitate to help by sending me record file which are better than the ones in the github project.

Excellent, and I look forward to seeing it optimized; would you mind the code changes being merged back into the main version? Then people would be able to use the Windows nightly for speedrunning.

How do you change these files? It is just a random_seed number.

Mathnerd314 {l Wrote}:Excellent, and I look forward to seeing it optimized; would you mind the code changes being merged back into the main version? Then people would be able to use the Windows nightly for speedrunning.

My code is way too ugly to be merged back upstream.
My goal was to have code in only a few places and only using the input system, so that:
- people could easily read the code change and see that the TAS does not cheat
- be sure that the TAS only does things you could do with a gamepad, as it uses the input system.

I hardcoded the chosen path (going down at bifurcation) and some other things.
So no this code is not made to be merged upstream.

brmbrmcar {l Wrote}:How do you change these files? It is just a random_seed number.

No it’s not, they are binary files, you need to edit them with a binary editor. I use okteta.
You have the random seed indeed, and then you have blocks of 6 bytes for each frame, saying if each button is pressed or not. The buttons are ordered as LEFT RIGHT UP DOWN JUMP ACTION. (And it seems a lost of disk space to use whole bytes to store 1 and 0s but whatever).

I'll take a look then.

Very cool run. I hope to see more TAS runs like this, but even faster and better

Have you tried a forest speed run?

Hi, I'm OscillatingOcelot, one of the top speedrunners of this game.

The TAS can go from 17:49.6 down to a staggering 17:03, theoretically. This is by reducing the time spent not going right, as well as using a trick called "ICG" (instant ice cube grab) to throw at enemies to skip certain cycles. Many of these have been done RTA. Some ILs are even faster than the TAS, sometimes by seconds. This shouldn't be the case.

We can also save a lot of time by doing "boosts" where tux's hitbox is brushed against other hitboxes, and subsequently pushed forwards a few frames. These add up over the course of the run a lot. You can save about 0.1-0.3s per level by doing these.

There's also a faster Yeti kill we can do, resulting in an In-game-time of about 4.85s. There is also a potential way to make this 4.51s, but we are still waiting on one of the runners who was able to get this consistently to send us his demo files. He hasn't been on in 6 months.

There are some levels that can have staggering cycle skips, kaizo jumps, and all sorts of useful tech. I will be working on a TAS that does all of these. Once i've recorded all the levels, I will use MCMic's tools to play them all together. I look forward to what can be created. I may need help on getting MCMic's tools to work, but for now, i have figured out how to record demo's and play them back. This should keep me good for a while

Again, the TAS theoretical time is 17:02 or 17:03... I doubt sub 17 is possible but I will give every level my absolute best, even if it means trying a hundred RNG seeds on 23rd Airbourne to get the most optimal boosts. The current RTA World record is 17:11.816 by surbus (otherwise known as Buryusadu) and I am in second place at 17:12.734

Our times both dont go for some setupless ICGs. The TAS will.

New TAS in 16:51.37: https://youtu.be/hQrwOPMJ5-g