[Project] Turbosliders Bots

[power79]

I agree with those points. I tried to improve it a bit by adjusting the parameters. They make more mistakes but maybe it's better now overall. Most of the mistakes are due to bad speed adjusting when changing driving line but there are probably also bugs as the system has gotten quite complicated over time.

I would like to get rid of most of the collisions, but seems impossible. Side by side driving is also problematic as I can't really come up with system of which car should give in at those situations so they also add to the roughness.

I reversed the starting order to create some more action. So slowest drivers start at the front. I don't think it makes much difference with the laps I am using, but I think these races are getting almost interesting to watch. But that's probably just me, hehe.

BotTest_010.tsr

[Whiplash]

Fix the download link.

[power79]

Problem is not the link but my web page. You notice that previous tsr links don't work either. I hope it gets fixed. I changed my internet connection recently (same ISP) but I was told that I can keep my home page.

I am not home until next week so I can't post the recording to any other place as I don't have access to the file.

[pablinho15]

Good job power

[Whiplash]

Yes, power, this is much better now.

[Rendy Andrian]

I had to say that I enjoy both races. Didn't knew that there was a get together from one of the drivers in the middle of the first race posted, which I think it is a nice touch. I can assume that there's somekind of a parameter in which a bot will make a mistake somewhere in the race (read: Mistakes percentage). For the second race, I saw quite an action, even though there were some instances where bots collide another's back while they could slow down a little bit more to avoid damage.

Cannot wait for the next recording! I wish the next recording will feature damages.

[power79]

Didn't knew that there was a get together from one of the drivers in the middle of the first race posted, which I think it is a nice touch.

If you mean situation where cars get stuck behind slower car then that's not intentional but just a bug where car's can't find a way to pass the slower car for some reason.

I can assume that there's somekind of a parameter in which a bot will make a mistake somewhere in the race (read: Mistakes percentage).

No, all mistakes are unintentional. If I make them drive a ghost race then there's not much mistakes beside some smaller ones that come from bots not being able to perfectly follow AI lap driving line. Most mistakes in these recordings come from passing attempts and bad speed control during them.

For the second race, I saw quite an action, even though there were some instances where bots collide another's back while they could slow down a little bit more to avoid damage.

I know this problem, and I try to fix it.

Cannot wait for the next recording! I wish the next recording will feature damages.

I replaced damage with TyreWear and I don't see much point to have also damage as it would complicate the bot's pit strategy which is already quite difficult to optimize.

I might create one more race recording if I can improve the driving enough for it. And maybe at some point, June perhaps, I am able to host a network test race that's open for everyone. But can't promise anything.

[Rendy Andrian]

If you mean situation where cars get stuck behind slower car then that's not intentional but just a bug where car's can't find a way to pass the slower car for some reason.

No, I was referring to the first race's recording where one car went too wide that it went off course at one point in the middle of the race.

[power79]

This is best I can do right now. There's still lot of things I don't like about these bots, but I think they are improving.

BotTest_011.tsr

And here is short 30sec clip about me standing as an obstacle to a bot at different parts of the track. Not perfect, but have to do for now. I edited video to only include "high lights" so you don't need to do any fastforwarding.

ObstacleTesting2.tsr

[Whiplash]

I think it's actually hard to see how smart those bots are when that car is picked. Even humans can't be fair with it.
So, please, try to apply those bots on this tack and car. You don't have to use pit (fuel and damage).

Track
Car
AI lines

[power79]

Ok, here's the result:

BotTest_012.tsr

Clearly more challening as the bots are designed for fast takeovers, which isn't usually possible with this car. I don't know if I should continue to work with this car or with Prime? Not sure if I can make F1-NG work as my bots can't really see that far to the future as would probably be required for them to drive properly with such a slow turning.

[Whiplash]

Well, nobody would play with this car I guess.
I thought it would work better with current bot settings, but obviously you would need more tweaking than I thought.

[power79]

I have now implemented speed control for passing situations so now cars are much better at staying on track. And same method seems to fit both F1-NG and F1-Prime. Of course some adjustments and improvements are still needed, but I am ok with these results so far.

I made three quick 10 lap race recordings this time.

1: F1-NG cars at Eisenstadt-2013 using Whiplash's AI laps.
BotTest_013a.tsr

2: F1-Prime at TSE_Bruinen
BotTest_013b.tsr

3: Both F1-Prime and F1-NG at TSE_Bruinen. You can follow faster Prime cars and see them struggling a bit when slalomin between slower NG models. But I think it could go much worse so no need to be too negative.
BotTest_013c.tsr

[Whiplash]

Yeah, it's a good idea to combine them. But, speed control has somehow made the faster cars to act like a cowards. Most of the time they slow down and hide behind he slower cars even they can overtake them easily.
Seems like the main problem is that cars can't understand when to overtake and when not. That's why I'm wondering what would happen if you would implement some more aggressive and simpler logic:
- Cars that are behind should always try to overtake.
- Cars should never slow down to avoid other cars, but only use steering. If a car slows down, that should be caused only by ai line.

Maybe you've already tested that? Even if some crashes happen, I feel like they would still look more natural than the current pussy waiting.

[power79]

Yeah, it's a good idea to combine them. But, speed control has somehow made the faster cars to act like a cowards. Most of the time they slow down and hide behind he slower cars even they can overtake them easily.
Seems like the main problem is that cars can't understand when to overtake and when not. That's why I'm wondering what would happen if you would implement some more aggressive and simpler logic:
- Cars that are behind should always try to overtake.
- Cars should never slow down to avoid other cars, but only use steering. If a car slows down, that should be caused only by ai line.

Maybe you've already tested that? Even if some crashes happen, I feel like they would still look more natural than the current pussy waiting.

Yeah, of course I have tried. But I don't think any simple algorithm works. a) They have to check if there's room on the road for overtaking. Not easy, when everyone is moving and bots can "see" the road only certain amount ahead. You have to estimate how long the overtaking will take etc. b) When you move away from optimal driving line, specially if you overtake from outside, you almost always have to slow down, otherwise you drive off road.

But yeah, I try to come up with something better. If any of you have some knowledge or algorithm suggestions, go ahead. I have tried to find stuff on internet but they are all pretty useless.

Here's little more detailed info what I have included into this shit:

1. Estimating road length it takes for overtake based on current speed of cars.
problem: Usually if car can't find a spot to overtake it slows down to avoid collision. This makes the speed difference of the cars smaller --> Even less likely to have enough road to overtake --> Gets stuck behind car

2. Checking if there is enough room on road for overtake ("passing lane").
problem: Cars move around constantly so situation changes and this is not that precise stuff anyway.

3. Checking if car have enough time/space to turn the car to overtake/avoid collision.
problem: Speed/direction of the cars changes constantly and its not linear change, so calculating this is always just rough estimation.

4. Deciding which side is better for overtaking
-Default is from inside. But there should probably be some better way to determine this. Sometimes the side changes during overtake and that can also cause problems. There is also some restrictions for outside overtaking as it is longer route and you usually have to slow down to not drive off road.

5. If car can't overtake, can it wait or does it have to brake.
-Not much of a problem unless you want to keep certain safety distance. Then you have to take that to account.

6. Calculating the spot where bot needs to "aim" when overtaking and deciding when it can get back to normal AI line.
-This could get difficult specially if cars are side by side because then both cars might decide to avoid the other at the same time. Otherwise I think my method works as well as it is possible. Can't make them humans, though.

And thanks for feedback, Whip. I know there are not many here that care enough to check these recordings (and I don't blame them, hehe).

[Whiplash]

Yeah, I guess I got the complexity of the problem you're trying to solve. I believe this could be a very useful stuff for Ande if he decide to continue the game developing one day.
Sure, he could change some things is the game core to make it easier for him self, but still he'll need some basic logic for the bots' behavior. I hope Tijny will also find some nerves to try this thing out too.

[Gustas]

Where to download f1-NG?

[Whiplash]

Where to download f1-NG?

Here.

[Gustas]

Thank you!

[Gustas]

will this project be released soon?

[power79]

will this project be released soon?

No. First I have to get the driving part decent. After that tons of other stuff. And I don't have enough time for all that. But I hope someday...

[Rendy Andrian]

Any new progress? Just checking.

[power79]

Any new progress? Just checking.

No. Motivation is pretty much gone and don't really have any time either. Maybe I get something done at summer time if I get any good ideas.

Right now it just doesn't work as well as I want. I have a terrible feeling(again) that I should just start AI driving algorithms from scratch, but that would really be a huge task. Also there's practically no user interface and adding+testing new tracks/AI lines is too difficult etc.

And today's community interest seems all to be at default car racing which don't work with this, so what's even the point?