[Project] Turbosliders Bots

[power79]

As we have no idea when Ande is going to release new version and if we even get better AI driver's(pitstops and fair play driving) for that, I decided to start my own AI project for Turbosliders. It's a long project and I am still just testing stuff but at least I have something to show at this point. I don't expect anyone to be that interested for this project but maybe sharing stuff here motivates me to get things done faster.

At the moment my biggest obstacle is probably getting the bot to drive fast enough. Right now I have to slow it down because otherwise it drives off the road. Ande is probably not going to release his AI engine algorithms to me so I just have to figure something out. But for now I think I just focus on other things and improve this part later.

Anyway, the basic concept seems to work ok. I don't have much to show yet but here is a 3 lap "race" (TSE_Veloccio) where bot driver drives first a normal lap and then makes a pit stop at second lap. First lap follows(or tries to) MikeNike's AI lap line (with much slower speed) and then at the pit lap(s) it follows driving lines by me, which are much worse as I just wanted to stay on road while recording and didn't focus on making a good lap.

BotTest_001

Maybe you laugh watching this, but I have faith that I can make this work. Next step is probably to make more bot driver's on track and make them avoid collisions (fair play driving).

[Whiplash]

Oh man, this is a great shit I was already insisting that Ande spend some time on making bots smarter (going to pit, be fair etc.), because that way TS would become somehow rounded of. He told me that the main problem is to teach them to go to exact place where pit is, but he promised to do his best. I wish you could explain better how you did the trick and I 'll inform him about it.
You probably made separated driving line for the pit laps?

[power79]

You probably made separated driving line for the pit laps?

Yes, pit entry and leaving pit are using two separate ai laps. Finding the pit spot I used a shortcut and just marked the coordinates where I want the car to stop. Later I try to do it more automated. I think it can be done by checking the pat image for pit terrain and taking the coordinates from there. But right now Im just testing so I use easiest/simpliest possibly solutions for things.

[Whiplash]

O.K., then keep working on it, and on the end we 'll inform Ande about what you've done.

[Rendy Andrian]

And don't forget an algorithm to make the AIs drive to the desired line in case of different start and start/finish lines!

Like this one actually. I've already posted this but idk, no one understood probably...?

[power79]

And don't forget an algorithm to make the AIs drive to the desired line in case of different start and start/finish lines!

Like this one actually. I've already posted this but idk, no one understood probably...?

That's probably because AI cars search the closest AI point and in your case its before the start line. Problem is that AI can only see the AI lines so it doesn't know that there's normal road straight ahead instead of obstacles,grass etc. So making that kind of start work would require analysing the whole track and not just rely on the AI lines. I had something like that in mind but its way too complicated to work at this point. Other possibility is to make separate AI lines for first lap. TS doesn't do this with airecord but I could make it from tsr file same way my TSR_Reader software makes AI lines from recordings. But even in that case the cars would follow the first position car's line so it probably wouldn't work the way you want.

[Whiplash]

Yep, the only solution is to make a separated first lap line and then force all cars to use it in first lap and never again. I'll mention this to Ande at some point.

[power79]

Forum seems pretty dead again so let me post an update. Now here's a video of full Veloccio race(37 laps) with fuel,damage and 20 bots on track. Bots choose randomly from two simple pit strategies before race. No collision detection yet so this is a ghost race. They use one AI lap for normal laps and 2 laps for pit stops. Ai lines for pit laps are still really bad so that costs them some time. Also changing start fuel is not yet implemented so they start with full tank without much fuel tactics.

BotTest_002

Finding pits is now automated, but my method is probably not good enough for Ande. And I don't even know if it works well at other tracks. But for these tests I think it will do.

Bots are reasonable competitive with full tank but without fuel/damage they are about 5% too slow. You can see in video that they don't improve their lap times much during race. I try to work on that. Also they make way too many mistakes.

But again, the concept seems to work so maybe this ends up being something fun.

If I would know anything about Punaball then probably bots for that would be possible too, but I leave that to someone else(Ande).

[Whiplash]

Quite fun to watch. With collision included, bots would make mess in pit. I hope you 'll be able to solve that. Anyway, ghost bots will be very useful for practice still.

[power79]

New race video with some improvements:

- Bots have now 10 ailaps to choose from randomly for every new lap. Thanks to Mike Nike for providing nice ai lines. Some laps are of course better suitable for bots than others. And pit laps are still quite bad as those lines are from my shitty driving.

-Bots are faster and make fewer mistakes. By sticking only with best laps there could have been even less mistakes, but I thought that diversity of driving lines is nicer to watch. Without fuel/damage they still lose about 1sec/lap to Ande's AI cars and about 1.5 to Mike Nike's original time, but I am satisfied by the improvement at least for now.

-Bots now choose startfuel from 30-60 range. Pit strategies are not optimal or even very logical sometimes. I didn't spend much time for those. Some of the cars even go to pits second to last lap. So there's still bugs to fix too.

Winner's time was 26:42 so not quite top level perfomance and it's still a ghost race but not so laughable driving as in my first test attempts.

Next I try to build some very basic collision avoiding system and perhaps make bots race against me. Don't know how much time it takes but stay tuned for new videos.

BotTest_003

[power79]

Small setback as I noticed that you can't really have my bots driving when TS is in fullscreen mode. TS probably reserves too much resources or something in fullscreen mode so my bot software can't function efficiently enough. But in windowed mode stuff work ok.

Here is me driving full race in Veloccio against 19 bots with "fair play 1.0". They are not smart or anything but at least they are aware of other cars, so it is not a total chaos on track.

Would be nice to get feedback if someone is watching these videos. Anyway, I think next I focus on making the bots faster. They are using Mike Nike's AI lines so they should be a lot faster than this.

BotTest_004

[Mike Nike]

woah..what a photo finish ;D...damn that b*tch blocking you to the sand, few curves before the end ..you woulda caught them

to your bots: cant rly tell right now, how good your work was - sometimes it looked like people try to let someone pass or weirdly try to defend (while not rly needing to), sometimes they were diving to overtake, sometimes it looked like the slow down bigtime without touching each other, lap 10-12ish or so, some cars after/while the circle (or that was just my pc being warpy).
all in all, it felt like a good simulation of a random race, kinda like in a f1 manager game. just...way better. except that you cant chose the strategies of course.
on the other hand, one time someone made a mistake, short before the curcle...and 5-6 cars in a row drove into the car in front without eving trying to overtake. that looked like typical "poor" bot behaviour.

all around, looks promising.

[Whiplash]

Hah, I'm actually already amazed by how good they are. They finally have a brain - good job!
I would just like to see them more aggressive at some points. I mean, sometimes instead of taking the overtake, they decide just to slow down. But most of the time, they are really good. Ande could let as having ability even to tweak their aggressiveness, not just the speed.
Btw, I already informed him about your tests and he seems like interested to work on this either.

Also, I would like to see a video with Sliders on Sandycur or on Rockfore.

[power79]

Thanks for the comments.

Yes, bots need to be more intelligent. Right now they don't care about cars that are behind them. Overtaking needs adjustment too. They try to decide if there's enough straight to get pass the car in front. In corners they are more careful as changing the driving line is more dangerous. It is also bit tricky to now how much road space there is on side of the cars. I think there's also some bugs still in the code.

I try to improve those issues. But I also need to get more speed to those bots.

At this point I am not sure if sliding cars would work well with my test code. All non-sliding should work in theory but I think I continue testing with F1-premium and TSE-Veloccio for now.

[power79]

I found some critical flaws from my fair play system so I basically build everything from start. So let me present you another full race video. Bot system could now be called: "Rough Play 1.0". There's lots of adjustments to do and probably million bugs or so but this forum is so damn dead at the moment that I thought posting something for you to watch.

I wasn't even close to winning this time. I have just been running simulations lately so lost touch to driving I guess.

Bots are quite rough at this point but not totally blind. They can only "see" one car in front of them which is the car they have smallest estimated impact time with. So when there's more cars packed together there's lots of collisions. Specially in the first lap. One car didn't even finish the race. As you see in video, sometimes it happens that after having an accident bot mysteriously starts driving backwards for a while and might get stuck to something. I guess you could call that rage quitting.

They also collide in pits but that's on my long list of things to fix. Anyway if you care to watch the video(or some parts of it) I could again use your analysis and suggestions.

I have to start summer break from this project soon, but maybe there's still time to get one update next week, can't promise anything. Whiplash could also try to get some new words from Ande, wasn't it last summer when Ande got lot's of progress with the beta? Maybe he finds some time for it this year too. This community seems to really need some action. Even Szymek seems forgotten his rally cup stuff? I thought he would organize test rally in June but I guess not.

Anyway here's a TSR video of me fighting with(losing to) 19 mean bots:

BotTest_005.tsr

[Whiplash]

Well, I didn't see them as too rough there actually. Somehow they look like more "humanic" even with all the "mistakes" they do. And, as I saw, whenever they pushed you out of track - that was your fault.

The only thing that is still unnatural is their decision rather to slow down than to overtake a car that is fucked up in front. And of course, those pit bugs need to be eliminated.

About the game's deadness:
I'm not going to bother Ande again any time soon, cause I already talked to him last week. I believe he'll not forget about us.
I'm more disappointed by dede and T.
And I'm mostly spending time on creating cars for the new version. I guess that's the only area that I'm good at. Unfortunately I don't know how to code anything. Otherwise TS 3D would be already launched 5 years ago together with multi-medial website and centralized ranking system.

[Tijny]

I'm more disappointed by dede and T.

How come? I don't see why it's our responsibility to do your shit. I haven't been interested in anything but Punaball for literally years now, so yes, excuse me if I don't feel like dancing to your tune.

As with all games that aren't updated in over 6 years, it's only logical that the overall activity is in decline. It's a miracle the community survived as long as it did. So sure, you can blame me for my lack of interest, I blame myself for that too, but I honestly don't see anything I could've done that would be a long-term solution to the problem.

[Whiplash]

I'm not blaming you. I just can't expect anyone else to stop the game dying other than you two.
The website is totally separated from the game for too many years now. For me it's just weird you can just watch it and doing nothing about it.
If you're busy, I respect that. But if that's not the case, please just stop saying you're helpless over and over again. New version will change nothing - even if it by some miracle become 3D.

Of course, you can say: "Well, no one pays me here, so I can't be responsible for anything". Sure, but that's one more reason for me to be disappointed.
And, once again, I'm asking nothing for my self. If others can't enjoy something, then I can't too. If you're having better ideas than me - put them on table and I'll be the first one who will support you.

[Tijny]

As for the website, of course it's been outdated for ages now, but I am not a web developer and I do not have the know-how to build a new one, especially not by myself. That's not me being helpless, that's me being realistic. And sure, a new version will change nothing now, but when potential new players see that the last release was six years ago, that's going to put them off.

[Whiplash]

Then dede sux.

Anyway, release day can't be a problem for a new players since they can't see easily what it is. I don't believe they explore the "doc" folder right away. Btw, 2008 year isn't that far still.
The game can be refreshed by the website features that will finally care about people's activity. For example, you can finally do some experiments with the scripts and see how they influence the activity. You don't need to build anything on the site except the easily noticeable link that will lead us to the simple ranking. And...of course...experimenting with the scripts doesn't mean experimenting with the cars (F1-TSB, Trololo etc.) - like you did last time. If you want a serious activity, you need a serious competition.

[dede]

Hello there

I've disappeared for quite a lot of time and this is not a comeback message. I haven't understood how power did this, but it looks great!

As for the website, I feel that I will never have time again to take care of it. I'm available to give access to any other webmaster in order to make a new version / develop new stuff.


Ciao in the servers sometime, maybe

[Mike Nike]

@dede
if we'd have a potential programer like power, me or whoever...

how many hours of learning a) PHP b) HTML c) CSS d) other languages
you would say he needs to invest to learn it from the beginning...in order to be able to maintain the ts page and be able to develop similar tracks and statistics subpages?

[Whiplash]

@dede
if we'd have a potential programer like power, me or whoever...

how many hours of learning a) PHP b) HTML c) CSS d) other languages
you would say he needs to invest to learn it from the beginning...in order to be able to maintain the ts page and be able to develop similar tracks and statistics subpages?

Mike, I love you.

[Rendy Andrian]

Just watched the Bot Race #4 where the AIs brake whenever they're too close to each other and not forgetting the fact that they can pit too. I was like "Damn, how could you do that?" and then a Le Mans-style race pops out on my mind, saying "Hey, can we do races like these, Le Mans style? "

Maybe that this will be a good cornerstone for "AIs' characteristics"...

[dede]

@dede
if we'd have a potential programer like power, me or whoever...

how many hours of learning a) PHP b) HTML c) CSS d) other languages
you would say he needs to invest to learn it from the beginning...in order to be able to maintain the ts page and be able to develop similar tracks and statistics subpages?

I think everything is pretty easy. Hardest part might be to understand my code probably. I would have some difficulties too, I guess