Trackmania TAS - What Now?

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.
Additional content from:
Angecide: / angecide
TmSpeedy0407: / tmspeedy0407
Trackmania: / @trackmania5791
Links to FAQ and vote:
donadigo.github.io/tmtas
donadigo.github.io/tmvotetas
Music:
Dye O - Bloom
Dye O - Interdimensional
www.epidemicsound.com
Marvel83' - Lost Highway
marvel83.bandcamp.com/album/g...

[@spike1613]

3:29 Riolu: 0% legit omegalul

[@averagebeetle]

The fucking cheek of the man holy shit, nice catch

[@wicowan]

yea whatever people say, he's so fucking selfish

[@maxmercurythemm827]

yeah that aged well

[@danielleh.8446]

Hilarios and pisses me off at once

[@nini700]

He knows what he’s talking about 😄

[@rollintm]

Really well put together video, I hope TASing can enhance Trackmania rather than be a detriment to it, there are arguments for both cases. Even if I personally would like to see the tool released, part of me understands why people are hesitant. This "technology" is still relatively new compared to the long time Trackmania has been out there, so perhaps it would be too early to release it now and open Pandora's Box if you get the metaphor.

[@vonix9199]

Pandora's Box is a really nice and fitting example, ye

[@mrstronghands476]

I hope they do a tas one day on d1 platform it's such a great map. You could do so much on the second last jump with bugs

[@Nicklander]

Thank you for this in detail explanation. The video showed that you would love to release your tool to the public, but that you are also very afraid of the potential abuse and the community's backlash. After all, the most qualified person to take the decision is yourself. The security measures you mentioned are great, I'm very impressed by your dedication and thoughts behind it. I believe that a public release is agreeable under those restrictions. My recommendation for you would be to make the tool as safe as possible and prevent any bypasses normal people could exploit. And that's where your responsability as a dev ends. In case a malicious hacker decides to reverse engineer your tool or make his own tool well... I don't think we should make you walk the plank for it.

[@Madinko12]

As a dev, you aren't responsible for how people use your software. Web browsers let you visit illegal or even criminal websites, but they aren't responsible for people visiting such websites or having criminal activities. This software can (and should) be made open-source so that everybody can use it, port it, review it, improve it and find new usages to it. Cheaters are responsible for cheating. It's Nadeo's and the community's resonsibility to do their best to catch cheaters. And anyways, cheaters may be dumb, but sooner or later, they will achieve to make their own tool and use it for worse. And sooner or later, they will be caught. In chess, Stockfish and LeelaChess are both free and open-source engines that can be used to cheat, however they didn't kill chess at all. Instead, they helped a lot of players to improve their playing. Also, they led chess organizations and websites to improve their anti-cheat procedures and this is a good thing: maybe we are now able to catch cheaters that would have cheated anyways and not have been caught otherwise (see also survivor bias).

[@Madinko12]

I may add that if the tool is published without the source code, it requires very little knowledge to reverse engineer it, and even less knowledge to unrestrict it (i.e. remove the offline limitation and the signature stuff).

[@lonesome3958]

​@@Madinko12I mean thats like selling guns for self-defense and then worrying about ppl using them for aggressive acts. I see how you might doubt if the rewards outweigh the risks

[@trabadia]

Very interesting video. Super well made, thanks a lot for your work dona. 9:10 drarker :(

[@bartog1800]

Why are you sad about Drarker?

[@trabadia]

@@bartog1800 he should have done 23.85... he failed at the very end due to a landing bug. He could have hold the world record

[@Cracks094]

I personally think a TAS tool for Trackmania would be able to elevate the skill-ceiling of the best players even further. Just like TAS in other games have led to new Techniques, Routes, Glitches and much more, i think the same could apply to Trackmania. TAS have been frowned upon in the speedrunning community for a long long time but nowadays they work hand-in-hand to reach the ultimate Speed for any game which i think is a good thing. As long as the TAS stay properly labeled i don't see an issue.

[@yesnoreal]

it did make a new glitch!!!

[@DikXyGaming]

I think the standardization of one good secure tool, made by one good guy, is better than all other futures

[@controllerpleb6568]

I really hope your tool gets released. TASing is an important part of speedrunning culture and most games have separate leaderboards for TASing.

[@TabbyVee]

@@missflissen you could always just analyze the inputs

[@casperTM]

Great video, glad you made this!

[@JstAnothrVirtuoso]

Thanks dona for the video. It's well explained. I think the community shifting to tm20 gives us all the more reason to release the tool. Personally, im really looking forward to playing around with it. It should crack open all new possibilities in the game!

[@vonix9199]

perfect 8-cut E02? 🤔

[@lonesome3958]

I cant download the new version (its complex), and so I started pouring in a lot of time in the old version, I would love to see that tool! Its got a lot of potential, but still contains a lot of danger

[@telve2882]

uff great video man, hope we see more videos like this and I hope also we see more about your really amazing tool. really well made man

[@nukegamer5750]

The amount of work that has been put into this video is staggering... Thank you... I really enjoyed it. Keep up the good work. It's seriously well put together. Can't imagine how long it must've taken haha.

[@Flopsaurus]

Thanks for all your work. Really cool to see the fruition of this tool coming through now.

[@ChuckLennon]

Also, very interesting video I came here to thank you for your work and doings, but the video is well-made and the narration captivates you :) good job on this one

[@Nexurl]

Great video, interesting topic and awesome editing, that's nice !

[@Alex-pq4vp]

Great video man, I have play Trackmania casually a helish long time ago but I still remember some of the maps. Thank you for bringing back some memories and teaching about TAS. Pozdrowienia z Polski

[@yoshtm]

Very interesting video, thx for sharing !

[@musti789456123]

oh man what a nice video :o I really like the parts you made from beginning too the end

[@AlexanderB41]

Great vid, hope you release it one day!

[@heliumnetworking5103]

Wirtual vid got me here. First vid I saw on that channel now in deep Nice TAS tool and release glad to see it

[@Angecide]

Great video, hope we someday reach a consensus where your tool can be made public

[@Needformadness2]

No

[@xnamkcor]

@@Needformadness2 Yes

[@cebulak941]

No

[@deleted9821]

What would you get if you had access to this tool? I mean, satisfaction is more at manual getting wr. TAS is like robotic and pointless.

[@Sharpless2]

@@deleted9821 TAS's could be used to find out what is possible after Humans think they maxed it out. This could inspire WR Hunters to push it even further and achieve true Perfection.

[@dimon110388]

Damn, i really don't want this to be abused, but i also wanna see what the limit is on so many maps. Anyway, gj on the tool

[@kass1564]

you finally did what I've been wondering if possible for more than a decade. Congratulations.

[@ivandimitrov2525]

First video for long time without " hit like and subscribe". Thanks for great video

[@Buuuuurd.]

Great video and great work

[@semmu93]

Fascinating video!

[@MazerMP]

3:53 "As of today this rule is being followed and there haven't been any cases of this type of records being passed on as real" welp, that statement aged like milk x)

[@OPEK.]

All the tas runs out now are cool

[@karison4]

Jesteś koxem, nie zmieniaj się

[@mcilrain]

I want to see all the cool stuff that can be done with a Trackmania TAS.

[@ZdympiaZ]

Amazing job, very well done !

[@gugus8081]

Nice vid !

[@tecphin6466]

Water goes brr when you fly over it. Thanks nadeo, very cool. Even tho almost everyone knew this

[@Poutrel]

Great video, very well explained! Although I'd love to see what people can do with such a tool, I totally agree with you on not taking the responsibility for potential cheating. However, I'd still love to see a multi-environment mapping competition like the WirtualBot one! Also, that tool would be awesome to do some reverse engineering about each car's behaviour (for example the impact of the steering ratio on the acceleration and so on...)

[@TmSpeedy0407]

Great video man.

[@mrstronghands476]

Nice video

[@BillyHudson1]

BRB going to make my own tool and release it

[@michal9350]

Pozdrawiam

[@buxxypooh8381]

Hey, love the vid Have you thought about making a better interface for the inputs? Like, some kind of horizontal timeline where you zoom in & out, drag and drop inputs, change input press time etc? Something that would look like the timeline of a video editing software, but instead of video clips, it's the game inputs Taking it even further, why not have some kind of breakpoint feature, where the game is sped up until a specific timestamp, to speed up the input edition process

[@dante8677]

I was looking for a way of tasing tm2020 but man it's hardcore to find something, maybe we'll get to it one time

[@GnuSnu]

Pozdrawiam z Warszawy

[@zoltanfridrich1934]

Very nice video. As someone who hunts nadeo WRs, especially those which are close to limits, it would be very tempting to use this tool and see what is possible. However it would be near impossible to TAS a better run then what human can do on envs like Snow where smoothness of steering matters much more then on Stadium. The only map I would TAS is A-0 to see if 4.38 is possible, my PB on that map is 4.390 so I am curious.

[@ChuckLennon]

I'm kind of desolated to see riolu did cheat, because even if I like him and all, you can't blindly say the rapport isn't concluding. Thanks for your work and for the efforts you put in. A cheater, even a loved streamer, is still a cheater at the end of the day. I would love to continue to follow riolu's career and streams, but I feel like I need him to admit and to stop. I don't know how it could be a trusted declaration but my hope's still there. Once again, thank you for fighting cheat and cheaters, and making this a better game and a better community.

[@wortexTM]

I'd definitely try and do some TASes for Lagoon if it was possible, just for fun, but I see the tool isn't universal :c

[@BigDBrian]

it should be released for sure. With those restrictions built into the application, there isn't really any room for abuse without deconstructing the software. And the people who have the know-how to do so probably also have the know-how to make their own TAS tools.

[@Oscar1618033]

I would like to see an AI optimizing the tracks to see the best theoretical time, and then players trying to reach it

[@sandearcubus9299]

There have been attempts at this. But TM is so complex that real AI based optimization would probably take a long time.

[@Elcanario91]

As an AI engineer I can tell you this is an easy task using reinforcement learning

[@wicowan]

@@Elcanario91 than do better than tas with AI, I'm waiting, but as said above to beat wr, ai would have to spend trillions and trillions of iterations

[@arnaudsm]

@@Elcanario91 I believe Genetic Algorithms make more sense here

[@Al.j.Vasquez]

"Ai" is like a buzzword these days, the implications of making the code for a good Ai for trackmania are big, although it has been done already, it hasn't even come close to being viable for this purpose. I would love to see increments in efficiency as well, but what's possible with TAS is similar to what they have done to Super Mario World, they broke the game badly and it's very interesting.

[@quike]

such a good video, never thought TAS was possible in TM (and i play for almost a decade..)

[@dmkmn4612]

Bardzo dobry filmik jak na taki mały kanał

[@moth.monster]

TASes are cool as long as everyone is honest about what they used.

[@bluestune_1981]

Nice video. Interesting facts about tools. And Sure, there is no fun with these abuses.

[@Kya10]

Your tool seems fine to release immo. With that said, thanks a lot for at least explaining how TASing in TM works :D Edit: Also, the vote has 69% of yes... Just sayin'

[@sirforcer]

I think it's fine so long as there is a surefire way to detect a TAS run. Other games use TAS to great effect, both to find new strategy and push the limits of the game. I think TM could benefit from this so long as TAS runs are showcased separately from normal runs.

[@prototyp60]

Hey, great video! Are you also planning to make one about TMTrackNN? I've been playing with it a lot recently (I'm the author of the only three issues in the repo) and I love how it works. Would be great to see more people experiment with it.

[@donadigotm]

I'd like to do video about TMTrackNN in some time... For now the focus is to finish the case with TAS and put end to all of it.

[@SugoDiGatto]

How did you make the car skin at 13:13? It reminds me of a bug I experienced in TMNF (or was it TMO? Too much time went by, I can't remember), by which if I Alt-Tabbed out of the fullscreen game while painting the car, I would get a glitchy livery. I couldn't replicate the bug on my newer PC, and I hadn't saved the skins, unfortunately.

[@lonesome3958]

"Tschüss" (10:07) Hab mich totgelacht

[@snap_oversteer]

Looking back at this, I wonder if riolu and others involved in the scandal realised what could happen to them when you showed ability to extract and replay inputs and just hoped their cheating wouldn't be discovered.

[@ExodiumTM]

Very nice video, quite interesting Just wondering, can the steer values change to the milliseconds? For example, at 10:45 you show that a 0.01 second difference in a steer value can change the run completely, but can the TAS tool also reproduce steer values to, let's say, 0.001 seconds, since it also shows the 0's at the end of each value? I really don't know if I'm explaining myself correctly

[@donadigotm]

The maximum rate is 0.01 and you can't go lower than that. 10 milliseconds passed = 1 tick ingame and if you wrote something with a non-leading zero then it wouldn't be even accepted by the tool because the time has the smallest step of 10 milliseconds. This is also the case in newer games like TM2/TM2020, they show thousands but these are probably determined differently than the main timer.

[@tonyg1011]

you speak really good english for someone whose first language is not english, but when you talk about how something looks, you don't say HOW something looks LIKE you say WHAT it looks LIKE. if you say HOW you never end the phrase with LIKE. great vid

[@TiagoTiagoT]

Would there be any way that the signature could be steganographically embedded in essential elements of the replay in such a way that both allow for the signature to be detected, and to either detect when the replay has been modified after signing or even outright makes the file not work (or just produce a bad run) if the file is modified after signed or if the signature is removed?

[@Druid_Plow]

I feel that sharing replay files for TAS runs should be encouraged, but only within their own forum's. I understand the hesitation to do so though, at least until there is a way to reliably filter them out of RTA runs to maintain the integrity of the leaderboards.

[@DikXyGaming]

Rly great work, just one question : Is your tool works online ? If yes, what happens with dedimania ?

[@_kapy_]

He answers that at 13:13

[@TiagoTiagoT]

Would the details about how you do the signature be something that you could share, or would saying anything about the signature system allow people to remove the signature?

[@johanneslusk917]

I'm interested, would it be possible to make ai assisted runs where an ai is allowed to change inputs to generate a faster run?

[@darkbelg]

Could you use this for supervised machine learning ? I am thinking if you use the replays of the top 1% to make a ai it might get pretty good. But it seems like having map knowledge is crucial to brake some records. So it will probably only do the trained maps really well. Or the maps that can't be cheated on and you just have to drive fast.

[@tristanwegner]

If the game can validate a run in seconds, this means it can run the physics engine internally much quicker than real time. This would be an amazing tool to train Neutral Networks to play the game, because they need many thousands of runs to learn a technique

[@donadigotm]

Hey, yes, and this is what TMInterface allows you to do. You can hook into the physics engine and run simulations at much quicker rate than a normal rate. There's been already some 3rd party applications for this and TMI itself provides a built-in bruteforce script too.

[@18ScorpiiREAL]

3:33 Riolu: 0% legit how the tables have turned

[@xnamkcor]

Do the physics run at, and synch with, the millisecond timer? I get the feeling if the input resolution is only 1 millisecond, desynchs would occur otherwise.

[@donadigotm]

A physics tick happens at every 10ms and it is completely deterministic. For example, a time of 16.87 would equal 16870ms (the time is always divisible by 10), which is 1687 ticks ingame. There's no additional "millisecond" precision, at least in TMNF. In TM2 & TM2020 the tickrate is the same, but the last millisecond is probably interpolated. To guarantee no desyncs, the tool is injecting itself into the game and directly modifies game memory to achieve this.

[@shantejo]

The next logical step would be to train an artificial neural network that takes a mapfile as an input and calculates the best possible sequence of inputs by itself.

[@corentinz6657]

Which langage of programmation do you use to make the TAS playing the game ? I used python but it doesn't work on games

[@PetrVejchoda]

chaos theory in practice

[@Amprovisation]

The video is great, good job! But just one thing Donadigo "FAQ"(At 15:42 and 15:47)by my knowledge, is french and is said , in english "QNA"(Questions aNd Answers) Person through the screen reading this hope you have a great day

[@Stuntlover-TM]

a tiny bit late bit FAQ stands for Frequently Asked Questions ^^

[@Amprovisation]

@@Stuntlover-TM I have learnt that since, thank you still

[@Stuntlover-TM]

@@Amprovisation haha np, also one of the quickest responses I've ever seen

[@Amprovisation]

@@Stuntlover-TM You're welcome?

[@keiyakins]

So, basically, trackmania players are stuck in 2003, when everyone was freaking out over Morimoto's TAS of SMB3.

[@tristanwegner]

So weird, that other videos talking about TAS Trackmania runs have millions of views, but the origin does not

[@integercyclolcyc]

please do the TAS in TMN race E-4

[@jeesonn]

Do u know how to fix game not compatible?

[@ora_veugle]

*Hi,* all TASing is about TMNF but what about TM² Stadium ? I really want to begin TAS run in TM, but i prefer in stadium I believe that a TAS tool exist for lagoon or other TM game

[@ingmu]

I knew physics in TM is deterministic, but can you explain why are there different results in my track - Art of Apex 3? A video from 2011 is in my channel.

[@donadigotm]

This is simply because the replays steer right at different times. The difference can be really small but the result will differ by a lot.

[@ingmu]

@@donadigotm Thanks for reply. I use keyboard to play. My original thought is, when I press right in the air, it will lead to max steering at landing, no matter which millisecond I pressed it, as long as it's pressed for few seconds. Well, physics proved me wrong, I just can't figure out what exactly happened there.

[@unifiedmongoose7915]

what do you think of the riolu situation?

[@unifiedmongoose7915]

oh wait your the guy who noticed it haha 😂

[@speed7424]

Can you read inputs for TM2 replays? Or only TMF?

[@donadigotm]

Yes, TM2 contains these as well. Either way I'm not interested in newer titles though.

[@speed7424]

@@donadigotm Fair enough, I can definitely understand that desire to purely focus on one game. Very good video btw.

[@azumanguy]

10:07 Tschüss!

[@theepicguy6575]

humans are aspiring to become TAS

[@VyrCZ]

Though what app or mod do you run the script

[@donadigotm]

It's a custom made tool for the game by me.

[@reddmst]

Next steps: 1. Design a headless replay renderer that simulates the entire replay physics-wise but without graphics, just to quickly find the final time given a set of inputs. 2. Develop a genetic algorithm to evolve perfect replays for each track. 3. ???? 4. Profit!

[@donadigotm]

Have you watched my latest video? 1. is already a reality! Now onto the AI side of things... we'll see.

[@reddmst]

@@donadigotm I've just found your channel today, so haven't seen the latest vid until now. Well, that opens up many possibilities for applying optimization techniques... hit me up if you'd like to talk about it - I've got some experience in those ;)

[@HayzerX]

As a long time player I have mixed feelings about its direct release and thank you for not releasing it yet. As with deepfakes, there is serious R&D of the detection programs running alongside it. Once the TAS software is released it yearns for a counterpart and a battle ensues. In short its a disturbance best avoided if you're just a regular TM player. It would be an extra thing to keep in mind (it kind of already is a bit but not that much). IF..... released, it would make sense to release it as a tool behind a barrier. What I'd initially imagine would be that you'd upload a map or a replay file and only get to view the output replay, but not extract it. The TAS replay viewing environment ought to be functional enough to learn from the run (like rewind and replay speed adjustments). But technically even then you could run a separate timer along the replay and get the input data if one really wanted. To try and prevent that, you could add randomized minimally noticable accelerations and decelerations to the replay video. But even then you could probably tweak the inputs according to the location ques... I suppose the next thing would be to flag all TAS times generated this way, so if a submission of a wr is made on mania exchange, its compared to the TAS replay which had been requested in some past. This would require some data storage capacity and might create a target for hackers (but they could be stored encrypted). So much hassle tho :S Perhaps avoid release as long as possible. Or if bored, keep thinking of ways to supersecure its release format and use. The knowledge of someone having done a TAS for TM got me disinterested in making my own. So there is positive effect of you having made it and talking about it.

[@TaxCollector89]

I have come to collect your cheese taxes. 10 pieces will do

[@SomeRandomPinkSpeedrunner]

ayo

[@dmarsub]

I don't see what would stop people from changing the code of your tool and removing the safety features. Am i missing something? I like TAS but it difficult to use in TM safely.

[@troiwgt]

the tool is not source code..

[@Slada1]

@@troiwgt you can modify executables without source code

[@CristiNeagu]

So... what's stopping people from recording a drive with AutoHotKey, then tweaking the recording until it's perfect?

[@shortcat]

exactly my thoughts

[@dimon110388]

Desync

[@CristiNeagu]

@@dimon110388 If this TAS software works by injecting commands based on time, then AHK works too. What is there to desync?

[@dimon110388]

@@CristiNeagu Maybe i'm missing some information about what's possible with AHK, but to my knowledge it's limited to "wait this many milliseconds, then press this button, ..." The "wait this many milliseconds" command however is extremely inaccurate and is not the approach donadigo is taking. He's manipulating the inputs based explicitly on the game's internal ticks (if i understood this correctly).

[@buxxypooh8381]

​@@dimon110388 Indeed I think he's editing the input value in memory read by the game input loop at every tick With a simple macro tool like AHK you'll get the butterfly effect

[@chester_m]

Great software! Could ideas from cheat detection in chess be used to catch cheating with TAS in trackmania? I tend to think TAS should not be released to the public but used by a handful of users known to Nadeo and each other. And have their results released to the public via KZbin.

[@donadigotm]

It's a double edged sword. Making TAS tools public also means that now anyone can inspect and analyze others runs easily. I'm currently working on another project called Competition Patch that is basically an anti-cheat for the game, and support for it is already implemented in TMX. Both projects are designed to coexist together and they both save their own signatures so they're easy to detect. We hope that'll ultimately solve the problem once and for all.

[@chester_m]

@@donadigotm Would be nice if you can figure it out! Good luck!

[@Kataquax]

how about demanding a hand cam for ppl that submit a nonTAS record?

[@PugCuber]

3:29 “0% legit” ok riolu

[@mitchd2214]

Totally would be awesome to see the tool at its limits for the classic maps. Would really suck to see the cheated runs on regular leaderboards though.. With more people building these tools it's only a matter of time before it's gonna be a required hand cam and a live stream or something other...

[@donadigotm]

We are actually in testing a new patch for the game that fixes the issue by detecting whether or not a run had injected inputs, was slowed down etc. We are aiming to set a new rule in TMX that requires players use this patch, at least for the most important leaderboards.

[@mitchd2214]

@@donadigotm That is awesome! Hopefully that will all go to plan there

[@ReiCaixa]

Try to watch this video without audio. Oh, you're leaving already? Bye.

[@Shweetz]

Maybe an idea would be to release the tool not usable with TrackMania but with a copycat game (CloneMania ?) which has same physics but doesn't output a replay that could be uploaded to TMX, you could only get videos out of it. You might get some problems with Nadeo, but if you disable inputs like keyboard and controller so that CloneMania can only be played through an input script, you could argue it's a completely different game ? Clone Hero is a real game based on Guitar Hero after all (although there is no TAS)

[@donadigotm]

We're actually working on something similar but not creating a separate version for TAS but separate patch for Trackmania that doesn't allow you to cheat with macros, slow-downs etc. called "Competetive Patch". If it's successfull, we may change the standard in uploading replays on TMX to require this new patch, at least for the most important categories.

[@Qbe_Root]

I'm not at all involved in the Trackmania community but it sounds like you need stronger verification standards (at least for top/WR runs) rather than private tools. If a closed group of people was able to make the existing tools, other tools could just as well be made by other people and either used for malicious purposes, or published without the community being able to do anything about it. This is especially true when no public tools exist, and making a new tool is the only way to TAS Trackmania in the first place; if your tool becomes public, there will be much less of an incentive to make a new one. But then again, it should never be ruled out that anyone can come in, reverse-engineer the game and/or the existing tools, and circumvent any security measures that aren't part of the game itself.

[@eig5203]

Me and the bois using autohotkey to easily make it

[@det-tn5qf]

Siema

[@flx910w]

So is it a computer driving or is it cheating on the possible values? Didn't get it.

[@donadigotm]

It is a computer replaying already preprogrammed inputs made by human to achieve a certain goal.

[@MaskofPoesy]

When anyone willing to cheat can develop their own tool, screw what the community says, especially the top players, do what you want to do.

[@olsirmonkey]

Id be pretty easy for someone like code bullet, if you've heard of him on KZbin

[@filsekfil]

TAS should be allowed - if people are afraid of cheaters then make better anticheat system - tases are art, tases are life

[@ThePandafriend]

It would be an interesting challenge to develop an a.i. which could perfectly finish runs. Good times are probably impossible on more complex courses though. At least with the current tech. Also you'd definitely need a speedhack, but making that wouldn't be very hard. So personally I'm actually quite interested in that tool. However for this I'd probably need the unrestricted code, which you obviously shouldn't give to a random stranger. Other than that it's nice that you built in the restrictions and I'm pretty sure that it would be fun to play with it, however at the end of the day it might be possible to reverse engineer it. Depending on the current system it might actually help with developing cheats for the current TrackMania title. If you think about it making tools for old games usually isn't profitable and after all the community is small. So I think that it's pretty unlikely that someone will abuse the tool for the old title, let alone reverse engineer it! However releasing it to everyone would probably lead to some players who'd try it out because it's interesting and most would drop it probably soon. I count myself to that group. Other than for TAS it could be nice for videos, but there are probably also not that many creators who'd use that tool. At the end of the day I do think that it's highly unlikely, but possible, that it might cause damage, unless the new game could be affected. However I also think that it's probably pointless to release it and a better choice would be to hand pick users. When someone has an interest which goes further than just messing around he or she should simply contact you.

[@BigBo-Peep]

Hmm making an AI to learn the best course on its own and perform it would be pretty doable. Would be an interesting project....

[@donadigotm]

In fact this has been done already by Yosh here: kzbin.info/www/bejne/l2mloGV6fdelpdk which I really recommend watching. Unfortunately this kind of algorithm or any kind of neural network doesn't work for finding a better time than WR due to the chaotic nature of the physics. A NN wouldn't be able to learn that changing a steer input by 1 step makes a huge difference in what the car does after. Although I do see some hope in using that for filtering out very unlikely solutions.

[@BigBo-Peep]

@@donadigotm That as a really good video, thanks for the recommendation. I think he made 2 significant errors, the first being that generalization isn't valuable here. Therefore, he could cut the track into pieces where all lines would converge, and judge success based on speed at that checkpoint. Creating a machine that can pick apart 1 map should be the key it would seem