donadigoPresenting A01-Race in 23.80. After ~50-100k brute-force iterations I was getting pretty sceptical that this time would be achieved with the base run of trabadia's WR 23.83 run. There was simply no time to gain at the end and the search simply didn't yield any results. Recently however, I talked with Adrien how it could be possible to "split" the map into parts and optimize these separately.
I then implemented a trigger system into the script that allows to define custom triggers which act very much like fake checkpoints that the script treats like targets to optimize for. These triggers are useful because they can also act like a fake finish line and can be evaluated much precisely using the position of the car instead. This run used one trigger in the middle of the map, just after the first left turn at ~13s. I ran the script repeatedly to optimize for this trigger which is why you see a pretty big difference already after the first downhill compared to trabadia's run.
After optimizing the first part of the map, I then moved on to fixing the remaining part for the proper line and so that it could finish the map. This fixing resulted in a time of 23.82, which was behind 81, but I knew it only lost time at the end. Surely enough, after brute-forcing the fixed 82 run, the script came up with a time of 23.80 after just 1k iterations.
I'm pretty sure that .7x is possible, but it will require some fiddling again to achieve it.
[TAS] Trackmania A01-Race - 23.80donadigo2020-11-25 | Presenting A01-Race in 23.80. After ~50-100k brute-force iterations I was getting pretty sceptical that this time would be achieved with the base run of trabadia's WR 23.83 run. There was simply no time to gain at the end and the search simply didn't yield any results. Recently however, I talked with Adrien how it could be possible to "split" the map into parts and optimize these separately.
I then implemented a trigger system into the script that allows to define custom triggers which act very much like fake checkpoints that the script treats like targets to optimize for. These triggers are useful because they can also act like a fake finish line and can be evaluated much precisely using the position of the car instead. This run used one trigger in the middle of the map, just after the first left turn at ~13s. I ran the script repeatedly to optimize for this trigger which is why you see a pretty big difference already after the first downhill compared to trabadia's run.
After optimizing the first part of the map, I then moved on to fixing the remaining part for the proper line and so that it could finish the map. This fixing resulted in a time of 23.82, which was behind 81, but I knew it only lost time at the end. Surely enough, after brute-forcing the fixed 82 run, the script came up with a time of 23.80 after just 1k iterations.
I'm pretty sure that .7x is possible, but it will require some fiddling again to achieve it.I created an A.I. to generate racing tracks in TrackManiadonadigo2022-05-30 | Hello everyone. TrackMania mapping is an integral part of the game that provides endless possibilities for players to explore. But can it be automated? In this video I show how I created an A.I. track generator which is able to generate technical tracks that imitate the style of tracks built by humans.
Thanks for watching!Trackmania but its Geometry Dashdonadigo2021-11-25 | I didn't upload anything for 7 months so I decided to upload something. Also TMInterface is out if you have been living under a rock: donadigo.github.io/tminterface.
This is an Openplanet plugin that I worked on a year ago but then I scraped it, for a couple of reasons.
Song: Waterflame - Hexagon Force newgrounds.com/audio/listen/568699[TAS] Trackmania D03-Acrobatic - 13.90 by Drarker & bruteforcedonadigo2021-04-12 | After plastorex's 16.30, it wasn't long before some of the top players started seriously going for attempts on the real D03 map. And it also wasn't long before we got our first cut record by Rollin, then by Hefest and Drarker. All 3 runs were again good material to improve upon with the bruteforce script but people really wanted to see a bugfin instead. First, I scouted the area for which run is the most viable and simply started bruteforcing Drarker's replay and ended up with a strong 14.08. But out of curiosity, I checked if the run could reach the finish directly and after some manual adjustments, the run barely didn't reach the finish. So after the adjustments, I decided to go for the classic approach of throwing the run into the bruteforce machine and sure enough, after just ~700 iterations, the script ended up hitting the bugfin with the time of 13.90.
It was always my dream to see such run come to life with the power of bruteforcing. It's just so exciting when you see a "Found a new record: 13.90" line in the console.
Who would've thought this would go so quickly.YEP Another .80 on A01donadigo2021-03-17 | Continuing the A01 saga, it turns out that it is not that easy to shave off another 0.01 of the .80 TAS run. Simply throwing more iterations at the problem is not enough, as the time barrier is pretty strong. We're entering an optimization stage where it's no more about "going closer to a wall", but more about relations between the speed and the line the car takes to reach the finish. Evaluating only the distance of the ghosts will not work as it assumes that being ahead at any point is actually faster, and as you can see in the video - it is clearly not.
There is also of course the elephant in the room - the blue bug before the downhill - and while I have set correct conditions in the bruteforce script for this to happen, no solutions have been found as of yet.TM Competition Patch - public testing updatedonadigo2020-12-30 | Discord server: discord.gg/hdwV62gxBp Post: donadigo.github.io/tmcomppatch[TAS] E05 60x Respawning Uber vs Demon - Who Will Win?donadigo2020-12-14 | -- EPILEPSY WARNING - this video has flashing images and sped up content --
This is the "we have 60x uber at home" version of this strat which means that it's the same run over and over again for 60 laps, but it respawns at each lap to make the same uber work. Demon is ultra fast on this map, and the uber gain cancells out with respawning each lap, this is why it's barely any faster than the real WR. If we were to do the real thing, we would gain at least ~3 seconds each lap which would equal to about 49-50 min depending on the run.
It was hard to find a uber run that would be done on the map with the multilap block. All uber runs are done on the "one lap" version with the multilap replaced by a start and a finish, which invalidates the run for a proper multilap version. In fact there is no such run afaik. What I did instead is I took Hefest's uber run on a miniuber E05 map, offseted the inputs so that it made it to the uber CP on the multilap version. Now, this alone didn't make the run uber, the setup was there so I bruteforced the uber using a trigger placed at the bottom platform, where the car lands (the uber and landing is still slow btw). After that, I manually drove the rest of the map and then gradually improved each section to achieve a somewhat good run - 53.44. The actual uber WR on one-lap is 52.62 by raizo but I was too lazy to adjust that one.
After having a full one uber lap, I quickly wrote a script to repeat the inputs of the run 60x times in the input file, taking the offset of each lap into account. The footage is not speed up, it's just the tool setting the speed of the game itself. When the car will be respawned, the speed is set back to 0.1x to make sure that the respawn will not desync, because this action cannot be injected directly into the game.
All in all, not that interesting or surprising - it's the same thing 60x after all.[TAS] Trackmania A04-Acrobatic - 5.77donadigo2020-11-18 | Up next in the series of This Is Not Maxed, a 5.77 on A04, again bruteforced on the current WR of WilliamB. It took overall just 3k iterations to get to 5.78 which is a tie with the previous TAS run of 5.78 (not by me) but the run had the advantage of a better line. I threw it again and in 100 iterations, I got the 5.77 which could've been just done manually from this point as well.
Taking this WR as the base run resulted in a more relaxed environment to brute-force in, as opposed to the very strict 5.78 TAS in which there are very few inputs and generally no room to experiment. Two steer values were changed from the initial 5.79 to 5.78. Perhaps a 5.76 is possible with a better start.[TAS] Trackmania D07-Race - 44.88donadigo2020-11-08 | So here we are, with the biggest improvement yet time-wise of -3.85 seconds. The idea of throwing a failed D07 cut into the bruteforce machine was suggested by Edge, which was followed by Hefest sending me this replay of Drarker's cut end fail.
After quickly adjusting the script to discard any finish time higher than 48 seconds, I started the script. In the first 2k iterations it was looking rough, but surely enough, after 3203 iterations the magical Hylis magnetic finish field was broken and the script came up with a time of 46.88. Now obviously this wasn't the end here. Since Drarker's old inputs after the fail were still there, I replayed the run fixing the end inputs manually for a much cleaner finish landing that resulted in a time of 45.25.
After that I also quickly bruteforced the second checkpoint since the run respawns there, so it doesn't affect anything after the respawn at all, improving the end yet again ending with a final time of 44.88. Who would've thought that Drarker's run would've made it into finish if he had only *one* steer value changed. That's like a difference of moving your pad stick 0.01 second slower.Trackmania TAS - What Now?donadigo2020-10-28 | Get to know about Tool Assisted Speedruns in Trackmania - how it all started, how it works and what awaits the future of Trackmania TAS'ing in this video.
Marvel83' - Lost Highway marvel83.bandcamp.com/album/genesis[TAS] Trackmania D11-Acrobatic - 9.42donadigo2020-10-24 | This run was found through the brute-force random algorithm which found this change in a quite short period of time. I'll be honest here, it maybe is improvable to 41 with the current base run, but anything lower would probably require a different approach or search algorithm altogether. No manual changes were done in order to improve this run.[TAS] Trackmania A06-Obstacle 27.62donadigo2020-10-24 | This run was found through the brute-force random algorithm through feeding the script the faster run it generated before. This one can be surely improved a lot, not only at the end but in the whole run too. I'm also feeling that with a longer search, this should get optimized as this is only 0.04 faster than the current WR by oli10. No manual changes were done in order to improve this run.[TAS] Trackmania A09-Race - 24.31donadigo2020-10-24 | This run was found through the brute-force random algorithm through feeding the script the faster run it generated before. Shows how much improvement there can be at the end, and this surely isn't the best you can do here. This particular run is 0.07 ahead of the current real WR. Trinity had a very sick bugslide here which commited for a good base to be improved upon. No manual changes were done in order to improve this run.[TAS] Trackmania A10-Acrobatic - 8.83donadigo2020-10-24 | This run was found through the brute-force random algorithm just after 60 random iterations, which shows how the original sinto's run was very close to .84 and how the "max" can be even lower, if you can call it that. I tried searching for a lower decimal for a longer time, but to no avail. No manual changes were done in order to improve this run.(Outdated) [TAS] Trackmania A01-Race - 23.81donadigo2020-10-22 | New time: youtu.be/IoDpf0Vwgso
The run itself doesn't look like 81 at all, especially when compared to the stellar trabadia's 23.83 run. This replay was achieved by first manually improving the 23.83 to 82, but this was done quite some time ago. The 81 was found through the random brute-force algorithm I was developing, while testing it, it found an 81. Compared to the previous 82, this replay has changed only 4 steer values out of 609, so the replays differ by 0.6%, which just shows how complex TM really is.TMInterface demodonadigo2020-08-14 | ...Inception - my first Trackmania map.donadigo2020-07-04 | Map link: https://trackmania.exchange/tracks/view/568[TAS] Trackmania B04-Acrobatic - 12.47donadigo2020-06-29 | This run was done using different techniques to squeeze out a hundreth out of Hefest's 12.48 run. For the wall entry and exit a brute-force script was used to optimize steer values and the end wall slide + finish were done manually with me driving and manual steer adjustments.