Duration: 0:10

388 views

9

Matt: Well, this is an odd one for me. I started working on a lv01ox demo somewhere around 9-12 months ago, honestly so long ago that I can’t say for certain exactly when, and have chipped away at it on and off over that time. I realised that if you gently crashed into the west wall before beginning the final eastern wallrun to the final door, you’d be in a good position to save a frame. It’d be tight but doable. The curve to the final wallrun is really annoying, you want to be south enough that you don’t crash into the eastern wall in that final corridor while approaching the wallrun position, but being too south tends to mean you have more south momentum which means that you need to dedicate more of your movement to removing this southern momentum while building eastern momentum, which makes it hard to cross the threshold for having enough momentum for the eastern wallrun. Travelling south for a further frame would have made everything easier, but there was simply no frame to spare.
Anyway, I threw a lot of time and computational power at this. When Rocky and I first developed a tool to run parallel brute forces this was the first project I turned it loose on and I ran a LOT of tests, one of which was over 58 billion input combinations over 13 frames, which means over 683 years of gameplay was simulated for that brute force alone, which was just one test of many. It was still a pretty good learning experience for me because having this new brute forcing tool has forced me to re-assess my understanding of what a reasonable brute forcing range is and the types of ranges that can now be performed with it (hint, 58 billion input combinations is not reasonable), and has also in turn helped my understanding of how much optimising curves can affect things, so it was still worth it in that regard, even though there was a lot of time ultimately wasted for no result. I also edited some stripped down map01 maps to run brute forces on that were similar enough to the actual map01 but with every unecessary thing, vector and linedef removed so that brute forces ran faster, which was also beneficial to experiment with.
Why do I say this was an odd one for me? Well, basically all of that was completely scrapped because in an incredibly short period of time Rocky got a better position than me leading up to that final wallrun. Relative to the length of the run this was the most time I have ever put into a TAS demo, and Rocky took a slightly different approach and made better progress than me in, like, a day. Definitely pretty humbling. His position before curving to the wallrun approaching the door was about 1.013 units better than mine and that’s what it took to make this work, which should give you some indication of how close this was to NOT working. Anyway, we got the wallrun, I slapped the rest of the run together and bam, 4.34. Massive thanks to Rocky for having a go at this, I’ve been wanting to see this done for a pretty long time now.

Rocky: Thanks to Matt for bringing this demo up. Slapped my old 4.37 nomo for the beginning, then brute forced the .000000 again, without any inputs below a value of 50 (only used turn values in the brute force). Took a few rebuilds of how exactly to turn while along the wall, but eventually I got what I think might be a damn near perfect momentum value on the last frame of the .000000 wallrun. Then it was a matter of getting a good collision on the right wall, which took some more rebuilding. Got really close to being able to wallrun (14.94 x momentum versus 15.0 required), which Matt then got with some brute force. He then built the rest. Really happy with this :D