The TORCS Racing Board
Username: Password: Remember Me?
Lost Password Register
Ban precalculated setups and racing line optimization info?

Yes, ban it (1, 25.00%)

No, leave as is (3, 75.00%)

Total votes: 4.
Poll open for racer.
Start: 2017-01-15 17:23:55.
End: 2017-01-28 17:23:55.
Author: trbadmin | Created: 2017-01-15 17:35:37
Subject: Banning precalculated setups and racing line optimization info?
Would you like to forbid shipping precalculated setups and racing line optimization info? The drivers would then just qualify 50 laps each, setting themselves up in those few shots using tCarPitSetup during a pit stop. Beware, in that mode the first driver might qualify significantly worse than the second.

Kind regards

Bernhard
Last Edited: 2017-01-15 17:35:37 by trbadmin
    Author: firechief | Created: 2017-01-16 05:57:39
    Subject: Re: Banning precalculated setups and racing line optimization info?
    It should be noted that neither of my robots (USR and Mouse-based) have any learning code whatsoever, neither do I know if I'll have enough time to implement those features - but nevertheless in the interests of making the competition interesting I'd be happy to see this rule come into effect as it'd concentrate on making the robots smarter rather than focus on the author's setup skills.

    Bear in mind though that you'll have to precisely define what is meant by "precalculated setups" and "racing line optimization info".

    Would it be ok to tell my cars how fast in general to attempt to drive on a track? Whether or not they should look for bumps? How much margin they should leave on the inside & outside of corners, and so on? It would be very difficult to forbid that, as all robots use instructions of that kind and have to use at least some kind of default values. Is it only instructions for specific areas of the track that are to be banned? If so, that would make sense as its the kind of thing a learning algorithm should be capable of resolving.
    Last Edited: 2017-01-16 05:58:31 by firechief
      Author: berniw | Created: 2017-01-16 21:41:34
      Subject: Re: Banning precalculated setups and racing line optimization info?
      My definition would go in this direction, regarding racelines and setup:
      Forbidden:
      - It is forbidden to deliver a car setup in the package (files and constants in code), all robots start with the TORCS default setup as base
      - It is forbidden to ship track specific metadata of any kind. Track specific metadata is data which assigns information to a location in space directly (e.g. for x,y coordinates, or per track segment or distFromStart, etc.) or indirectly (e.g. encoded as [time, velocity] pairs, because s=v*t)
      - It is forbidden to run a simulation in the bot (a simulation is in this context a calculation which calculates a result for time t based on a previous step or steps (t-1, t-2, ...).

      Allowed (during runtime):
      - It is allowed to do static analysis of the track geometry
      - It is allowed to calculate static properties based on track, car and recorded driving data
      - It is allowed to record data during the ride to analyse it for setup and driving optimization
      - It is allowed to alter the setup with the tCarPitSetup struct
      - It is allowed to store setups for later use (for the race and for the 2nd driver), including driving optimization data
      - It is allowed that both drivers use the final setup and drive optimization data for the race

      Something like that, of course I would put some work into it to refine it.

      Bernhard
      Last Edited: 2017-01-16 21:43:15 by berniw
        Author: firechief | Created: 2017-01-17 00:18:53
        Subject: Re: Banning precalculated setups and racing line optimization info?
        Ah ok, so no setup of the car either? That too must be "learned" by the robots, at the same time as they're learning how fast to drive the track, and in only 50 laps? This will make things exceedingly difficult, and bumpy tracks will be a disaster.

        Also most robots currently use what could be described as a "simulation" for avoiding collisions with other cars - I guess you're not referring to this in the new rule?

        Edit: If this rule change is voted in, does that mean we'll see new development on the berniw robot?? Also it does seem the kind of challenge that might interest some past competitors such as Christos, maybe Tim as well ... I can but hope!
        Last Edited: 2017-01-17 07:39:54 by firechief
          Author: berniw | Created: 2017-01-17 21:50:12
          Subject: Re: Banning precalculated setups and racing line optimization info?
          I have no idea, currently we are just discussing it, if people do not like the idea I will not introducing it:-) But what I envisioned back then when I initially implemented the tCarPitSetup stuff was racing like a human in a low budget racing team: Coming to the track, make the car ready, give it a spin and optimize driving style and setup in a small amount of laps (~100). And because setting up cars is considered a time hog I thought this could be a welcome alternative challenge.

          Bernhard
          Last Edited: 2017-01-17 21:50:12 by berniw
            Author: firechief | Created: 2017-01-17 23:04:18
            Subject: Re: Banning precalculated setups and racing line optimization info?
            Totally understood, and what I like about the idea is that it puts more emphasis on coding than manual fiddling with setups. However when real-life teams come to the track to race, even low-budget outfits have at least some idea of how to set the car up to start with - they don't send it out there in some default state & try to figure it out from scratch every weekend. Hence my suggestion to allow the shipping of minimal track-specific setups, which would take some of the burden off the code to get a decent car in only 50 laps while also making it far easier for the authors to create the setup, as there'd only be a few parameters to tweak.
            Last Edited: 2017-01-17 23:04:18 by firechief
            Author: berniw | Created: 2017-01-17 23:56:15
            Subject: Re: Banning precalculated setups and racing line optimization info?
            Apart from the TRB it would be an useful behaviour for shipped bots, because they would adopt without manual tuning.
            Last Edited: 2017-01-17 23:56:27 by berniw
              Author: firechief | Created: 2017-01-18 01:10:53
              Subject: Re: Banning precalculated setups and racing line optimization info?
              This is very true.

              Well, I'd like to read input from W-D & John and see if Danny has more thoughts on the matter. This will only work if enough of us have time to develop a decent learning algorithm, and/or we can attract more contestants.

              If I can make one suggestion though, if the robot code is good enough to produce a first-guess setup just from analyzing the track without driving on it, it should be allowed to write a setup XML before starting the race (as wdbee_robotics does for tracks it doesn't know about) and use that, rather than having to plod around in 1st gear for a lap to get to the pits in order to get the setup changed - that'd be a waste of a lap, and there's only 50 ;)
              Last Edited: 2017-01-18 01:15:37 by firechief
            Author: jspenn | Created: 2017-01-18 11:33:12
            Subject: Re: Banning precalculated setups and racing line optimization info?
            100's a small number of laps? I have just had my thoughts rearranged. I can see the problem with Valk. How many laps does it get?
            Last Edited: 2017-01-18 11:33:12 by jspenn
              Author: firechief | Created: 2017-01-18 12:16:24
              Subject: Re: Banning precalculated setups and racing line optimization info?
              Valk didn't even run in the championship for most of last year, but when it did it was using default setups with no learning at all. Hence it was indeed quite slow.

              Bernhard's idea for this season is to remove the need for robot authors to come up with the ideal wing angle, gear ratio, diff, brake, wheel and suspension (lots of variables in that last one) settings and instead we would concentrate on writing code which could figure out the optimal setup from both an analysis of the track and also iterative feedback driving the car over the 50 (100?) practice laps.

              Not only setup, the robots will also need to figure out for themselves how fast to take each corner and to slow down if needed to avoid getting airborne over bumps. I believe Wolf-Dieter's robots are probably the most advanced in this regard, as his wdbee_robotics already comes up with its own setups plus his code has a learning option for driving the track - Dandroid has the latter but not the former. USR/Axiom have no learning code at all, either for setup or driving, so I'd have a lot of work to do, as would Bernhard if he chose to upgrade berniw_2004.

              If this rule change goes ahead, this year will be the war of the learning algorithms, as the better the learning, the faster the robot's laptimes will be & the more chance it'll have of winning. There's also other aspects, such as ability to avoid collisions with other cars, overtaking safely and efficiently etc, but the 4 main codebases we have now (wdbee, dandroid, usr and axiom) are already quite mature in that regard, which has largely turned races into a procession.

              And as it'll be possible to run the learning for opponent robots, you'll be able to see before the race starts whether your car is likely to be competitive. Although that would, of course, be dependent on whether opponent authors are improving their learning code between races. Does everyone have time & energy for this? That's why I'd like to hear W-D's, Danny's and John's thoughts. If its all too much work this year and we risk losing people, there would be other simpler ways to mix the competition up as discussed in earlier threads.

              Edit: From my own point of view, it does not help that I've been struggling to find consistent employment over the last year - on the one hand this has given me more (unwanted) free time than usual, but I can't spend too much of it on Torcs or I'll lose the roof over my head :\
              Last Edited: 2017-01-18 13:10:22 by firechief
                Author: dummy | Created: 2017-01-18 17:31:47
                Subject: Re: Banning precalculated setups and racing line optimization info?
                The same here. I don't see myself spending a lot of time with development.

                EDIT: Maybe in 2018 we can write code with C++11? This would be interesting to me. Learn some cleaner coding styles.
                Last Edited: 2017-01-18 17:37:30 by dummy
        Author: firechief | Created: 2017-01-17 08:20:17
        Subject: Re: Banning precalculated setups and racing line optimization info?
        After giving it more thought, I don't think banning setups will work. If the robot learning algorithms are going to devise a setup, it'd be quite simple to make the car pit after the first lap and install the ideal configuration as enshrined in the code for that race - and that would be a difficult, if not impossible, thing to police.

        Better, I think, to allow the robot authors to apply their minds & experience to setup if they wish, but to prohibit the shipping of any "private" robot-specific sections in the XML, ban any other track-specific datafiles, as well as the ban on track-specific metadata as you described, whether in XML or in source code.

        This way, the qualifying laps can focus purely on how to drive the car as fast and safely as possible around the track, rather than doing that PLUS setting up the car as well.

        Bernhard, you could of course tighten the setup rules if you like. Allow, for example, only suspension spring pressure, brake front/rear bias, brake pressure and wing angles to be adjusted, with everything else locked to use the default. That way setups would be quite simple & not the time-consuming rocket science they've tended to become in recent years.
        Last Edited: 2017-01-17 08:21:06 by firechief
          Author: jspenn | Created: 2017-01-17 09:43:53
          Subject: Re: Banning precalculated setups and racing line optimization info?
          "Allow, for example, only suspension spring pressure, brake front/rear bias, brake pressure and wing angles to be adjusted, with everything else locked to use the default."
          Thanks, time-consuming rocket science isn't that attractive to a newbie.
          Last Edited: 2017-01-17 09:43:53 by jspenn
        Author: firechief | Created: 2017-01-17 08:26:10
        Subject: Re: Banning precalculated setups and racing line optimization info?
        Sorry, just thought of one more thing relating to the track-specific metadata.

        Both my robots occasionally have considerable trouble getting into the pits on some tracks, hence I added distFromStart-based XML elements telling them where to start braking and leaving the racing line.

        If this were banned, I'd probably spend much of my limited spare time between now & the start of the season figuring out how to get the cars to pit on their own reliably. And that would not bode well for the coming season as I'd have nothing whatsoever in place to make the cars competitive, they'd probably even be slower than berniw_2004 :)
        Last Edited: 2017-01-17 08:26:10 by firechief
    Author: dummy | Created: 2017-01-16 08:16:48
    Subject: Re: Banning precalculated setups and racing line optimization info?
    Andrew's questions are the same that I'm concerned about. At the moment I tend to answer no. I think it's better to have not too many rules, which force us to invest more time than we would like.

    If only one driver pulls out because of this, it would be bad for all of us.

    EDIT: After thinking about it. I would vote yes for a rule that says you have to drive the whole season with one setup only (can be different for each teammate).
    Last Edited: 2017-01-16 08:25:42 by dummy
      Author: jspenn | Created: 2017-01-16 11:37:31
      Subject: Re: Banning precalculated setups and racing line optimization info?
      "I would vote yes for a rule that says you have to drive the whole season with one setup only"
      Oh, Thanks! I was getting scared about setups.
      Last Edited: 2017-01-16 11:37:31 by jspenn
      Author: firechief | Created: 2017-01-17 00:14:21
      Subject: Re: Banning precalculated setups and racing line optimization info?
      Hmm. I'd definitely not vote for a one-setup rule. Reason being that if you got it wrong, you'd be condemned to mediocrity for an entire season with no chance to compete.
      Last Edited: 2017-01-17 00:14:21 by firechief
    Author: wdbee | Created: 2017-01-19 09:03:28
    Subject: Re: Banning precalculated setups and racing line optimization info?
    I did think about such things for a longer time and worked hard on learning code. But using such a small number of laps for learning will not work. Even if manually setup a car, you need much more laps to make it drive well. To get a good setup the wdbee_robotics is running some thousands of races (not laps) starting with the default setup (not the TORCS setup! This would increase the needed number of races!).

    It will not work at all driving on known tracks, even if selected randomly. Only if always new tracks will be generated, it could have a chance. Nobody will be able to decide, wether a constant in the code is allowed of not (Except for G = 9,81).

    Nobody will be able to decide, wether a change of the robot code is an allowed improvement of the learning or a forbidden hidden track setup.
    If using a neuronal network, nobody could look into it at all! Rules like this cannot work.

    So you just could drive all ten tracks on one day. (Same if using same setup for all tracks, why wait a year for the result???)

    The TORCS tracks have bugs, that have to be fixed by robot code and/or setup parameters (like entry and exit to/from pit, start and end of track do not match together in all cases, ...).

    The TORCS setups for cars for different car types are not same regarding usability for robots, some work nearly, some not at all.

    Using a pit stop for optimizing the own setup is wasting time. Being out of track while learning can result in being not able to come back to the pit at all. You are lost.
    Instead the robot should be able to just restart the simulation and use a modified setup. The limitation could be the time allowed for this test drives.

    This is, why I say, this is not an option for the TRB for this season and not for the next three years.

    Cheers

    Wolf-Dieter
    Last Edited: 2017-01-19 09:07:36 by wdbee
      Author: firechief | Created: 2017-01-19 09:35:54
      Subject: Re: Banning precalculated setups and racing line optimization info?
      > Nobody will be able to decide, wether a change of the robot code is an allowed improvement of the learning or a forbidden hidden track setup.

      Very good point - I tried to say the same thing using a lot more words ;)

      > So you just could drive all ten tracks on one day. (Same if using same setup for all tracks, why wait a year for the result???)

      And another excellent point. If the setup can't be shipped, the only point to spreading the competition out over the year would be to allow improvements to the learning algorithms - which then brings the first point above into play. If the learning parameters, the constants the robot uses to start with, change from one track to the next then is the learning actually improving or is it a subtle way to include a hidden track setup? This would be impossible to police, and we shouldn't have a competition where there's a lot of suspicion over whether competitors are sticking to the rules.

      > The TORCS setups for cars for different car types are not same regarding usability for robots, some work nearly, some not at all.

      This I'd see as less of a problem, as it'd be up to the authors to look at the cars in their TORCS setup and choose the fastest. It'd actually be easier than last year, where a change in the physics brought the car2 to the front of the field when given an optimal setup - but coming up with the optimal setup & thus identifying it as the quickest car was certainly not straightforward.

      > Using a pit stop for optimizing the own setup is wasting time. Being out of track while learning can result in being not able to come back to the pit at all. You are lost.

      This concerns me as well. Part of robot learning is pushing the boundaries, which results in cars leaving the track when they push too hard and on some tracks that can be fatal, as cars get stuck in a "box of death" area or on a gravel trap where the wheels spin & the car can't move. If that happened, the car would be doomed to enter the race with very little learning.

      > This is, why I say, this is not an option for the TRB for this season and not for the next three years.

      W-D would you prefer to see TRB races remain the same as they've always been or some kind of reverse grid used to mix up the field and put more emphasis on overtaking? Or is there another option we haven't discussed yet?
      Last Edited: 2017-01-19 09:37:57 by firechief
        Author: wdbee | Created: 2017-01-19 16:55:18
        Subject: Re: Banning precalculated setups and racing line optimization info?
        As long as we have robots like the berniw in the championship, running a race in reverse order will not work. The rules say, blocking is not allowed, but the robot blocks, even while overlapping, at least for 5 seconds. Well there is the other rule saying what berniw does is good behaviour.

        Means, if all robots show that good behaviour, we will get no race but a massacre.

        Bernhard likes to say, it is same for all, but that is not true. Just try to overtake such a car in the chicane at wheel-2. It is not possible without getting a lot of damage. But pass it at side of the main straight is easy, it likes to brake hard there being middle of the road. I never understood why it is braking having no opponent in front?

        It is random, where you have to overtake it, but you cannot wait, you have to overtake. Means the result of the race will be random, you will not win because your robot is the best. So why work on the own robot???

        Another thing is, that the fuel consume is not equal for all cars. Some can start with lower fuel but use only 3 pit stops on many tracks, while others have to always refuel as much as possible and to start with a completely filled up tank, otherwise it needs an additional pit stop.

        The argument, dont use 100% of the engines power, is possible, if the car still is the fastest. But if you would need 105% power to be competitive, this is not a real option.

        So my proposal was to add some rules.

        First was, to define a number of pitstops a car has to use, that is more than 3. This makes it possible for all to work out a better strategy and will result in getting rid of the processions.

        Next was to define a maximum number of allowed pitstops. Cars that do not drive well and get a lot of damages, will be taken out of race. E.g. this means you will have to rethink Dirt-2.

        The additional rule changes I proposed was to use the lateral speed of cars regarding the track while collision to define who gets damaged. A car holding the line in curves and at straights should not be punished if crashing together with another car closing the door at start of the curve or driving into the outer opponent at end of the curve because of speeding.

        Same if a car is steering into an opponent at straight while it is passing holding the line.

        This is simple to program and much better than the current way to punish both, the one in front less than the one at rear. This is OK if both hold the line and the one at rear does not brake as needed.

        Later, if all the robots work well using such rules, racing in reverse order will work much better. But still I would prefer to add weight depending on the amount of championship points a driver did get up to the current race.

        These new rules result in providing you advantages making your robot drive better. For me this would be a reason to rework my code. As a side effect, robots not driving well do not punish others randomly as much as in the current races.

        This keeps the door open for new racers. We all know, you need at least one season to learn to make a robot drive well and respect the rules and conventions. If new robots do not hamper the more experienced teams, the barrier is lowered to enter.






        Last Edited: 2017-01-19 16:55:18 by wdbee
          Author: berniw | Created: 2017-01-28 13:47:52
          Subject: Re: Banning precalculated setups and racing line optimization info?
          Hi all

          >Bernhard likes to say, it is same for all, but that is not true.
          Erm... ok:-) I highly disagree on that, because averaged over multiple races you will have "average luck", so it is for all the same. I was for a long time active in the top league of a real thing, where you have been dependent on weather conditions and judges. There it was the same: sometimes you had bad luck, sometimes good luck, regarding one specific run. But over many runs it does even out.

          So I agree that you might lose some points in one race because of it, but in another one you will profit from it (if somebody else has bad luck). If you are always the one with bad luck, well then, then it must be you, simple as that.

          >So why work on the own robot???
          Because it is fun, and there is still a lot to do?

          >Another thing is, that the fuel consume is not equal for all cars
          Yes, this is intentionally. Nobody is forced to use a specific car, btw. Anyway, based on our discussions (mail) it is possible the car6 gets a little improvement.

          >First was, to define a number of pitstops a car has to use, that is more than 3.
          I dont share this argument. If you want more pit stops I could increase the fuelFactor (I introduced this in some former version), see changelog:
          "Added fuel consumption and damage factor configuration to race manager, so it is now possible to set up racing with no damage/no fuel consumption (0.0) or up to 5 times damage/fuel consumption for more pit action, see practice.xml for the options"

          What about that, should we try 3 times fuel burn?

          >Next was to define a maximum number of allowed pitstops
          Does not make sense, it punishes cars with bad luck.

          >The additional rule changes I proposed was to use the lateral speed of cars
          I played a bit with it and the thought, it just does not work for me (I did manually test drive the idea). Easy to abuse, and wrong driver punished when dive bombing in a hair pin.

          So although I highly disagree with most of the stuff I appreciate that you care.

          Kind regards

          Bernhard
          Last Edited: 2017-01-28 15:06:12 by berniw
            Author: wdbee | Created: 2017-02-04 11:48:46
            Subject: Re: Banning precalculated setups and racing line optimization info?
            > If you want more pit stops
            The topic is not to have more pit stops but to make pit stops be independent from fuel consumption. At the moment you have to use a minimum of pit stops to win, no room for strategy. But if we get the tyres in addition, this may be changed.

            > What about that, should we try 3 times fuel burn?
            This would be the opposite of what is needed! Cars with low fuel consumption would get more Advantage in addition. In fact the time used to refuel is too long compared with F1 pit stops in less then 4 seconds.

            > Does not make sense, it punishes cars with bad luck.
            Sorry but it makes a lot of sense, if there are different teams racing for the same owner. Why not use one team to block/damage others as much as possible still regarding the rules? It is easy, just make it drive like the driving well by definiton team and go to pit if the other own team has to over take ;)

            > (I did manually test drive the idea)
            I did test it with the robots of 2016 and it worked as expected (I fear, what you did is not allowed as the rules say, you are not allowed to intentionally damage others).

            > Easy to abuse, and wrong driver punished
            Not so easy as the way it is handled at the moment. It is not perfect but much better than the simple current way to punish both and the one behind more than the one in front independent from how the steering is. Not punishing the driver holding the line is not punishing the wrong driver.

            At the moment we have the rules saying this and the damage system doing another thing.

            Last Edited: 2017-02-04 11:48:46 by wdbee
              Author: firechief | Created: 2017-02-04 13:34:04
              Subject: Re: Banning precalculated setups and racing line optimization info?
              > In fact the time used to refuel is too long compared with F1 pit stops in less then 4 seconds.

              They don't refuel in F1, hence the faster pit stops. Not taking anything away from what you're saying though - I agree that thirstier cars (or those with smaller tanks) shouldn't be penalized by having to make an extra stop as they have to on some tracks, as that 20-30 second delay puts them at an unfair disadvantage.
              Last Edited: 2017-02-04 13:34:04 by firechief
      Author: berniw | Created: 2017-01-20 08:13:51
      Subject: Re: Banning precalculated setups and racing line optimization info?
      >Nobody will be able to decide, wether a constant in the code is allowed of not (Except for G = 9,81).

      I disagree on that. If the constant is linked with a x,y pair in any way it would be illegal according my "ad hoc" definition above (track metadata). If the constant would be fed back directly into a car setup it would be obvious as well, the car setup must be a function of the collected data, nothing else (so if the neural network would not take into account the input data this would be pretty obvious).

      To pit entry/exit: Those are trivial to find, you have just to iterate along the track segments, then you "see" the space for possible pit entry paths.

      Laps: All would have the same amount of laps to learn, so the better learner will get a better result, who cares if this is still 2.5 seconds away for a possible optimum?

      Anyway, I see that you guys are afraid of it, so I will not introduce it in this form, but I have another possibility in mind, maybe for 2018 then;-)

      Bernhard
      Last Edited: 2017-01-20 08:13:51 by berniw
        Author: firechief | Created: 2017-01-20 09:20:39
        Subject: Re: Banning precalculated setups and racing line optimization info?
        I'm more intrigued by it than afraid of it, however I do think there's too many grey areas at the moment that'd have to be thought through. For example:-

        > the car setup must be a function of the collected data, nothing else

        My first thought was that my approach to the setup would be a static analysis of the track before the race begins, looking at things like the friction, the ratio of corners to straights, the amount of Z changes and whether any bump-takeoff-prevention is necessary. From that, I could have an algorithm that determines a fairly good setup in terms of wing angles, gear ratios, brake bias and suspension, based off my past experience in setting up these cars & knowing what kind of parameters work for each type of track.

        I wouldn't therefore have track-specific information embedded in the code, but I _would_ have minimum and maximum settings for each value based on my past experience. Would that be allowed? Or would I be required to send the car out in its default configuration & drive in circles to try to determine setup that way? Which would be very difficult as a setup doesn't really show its value until the car is being pushed to its limits, but that won't be the case as the driver is still learning how to drive the track.

        So working out as much as possible before the tyres even start turning - that seems to me to be the optimal approach, and indeed it'd be one that your "real life" example of a team visiting a racetrack would take. They wouldn't just go out there blind with a default setup, they'd research as much as possible & walk the track in order to have a good idea of the setup from the outset.

        Hence this is something you'd need to clarify - and I don't think this is the only aspect that will rear its head when/if enforced learning is introduced to the TRB.

        Last Edited: 2017-01-20 11:22:43 by firechief
        Author: wdbee | Created: 2017-01-20 12:58:21
        Subject: Re: Banning precalculated setups and racing line optimization info?
        > (so if the neural network would not take into account the input data this would be pretty obvious)
        But if it is linked to all the data you can take out of the TORCS track description, how will you decide that it is really using the information?
        If there is a basic function working in it, it will show results on all tracks but provide an optimal setup for the current race. As Andrew stated, we should come to black and white (rules) instead of gray (zones).

        > then you "see" the space for possible pit entry paths.
        But using these raw values without modifications does not work to savely enter an leave the pitlane at higher speed and results in damaging opponents often.

        > All would have the same amount of laps to learn
        This is OK, but the amount of laps is much to small. Have a compare to the number of parameters needed to define a competitive setup. Using the pit stop way for learning, you cannot use large steps, only small changes can keep the car on track. But this increases the amount of needed laps.
        The number of steps you need for a parameter depends on the function it has. Let's look at the wing angles. Using 1 degree steps it takes 12 at front and 17 or 18 at rear. But you have to test it together, so 12 * 18 laps are needed.

        As I found this takes to long I implemented two different parameters, scale download and scale balance. Now I can say, increase the downforce but keep the balance or vice versa. This resulted in faster learning.

        But this has to be combined with the ratio of the transmission. Less downforce needs longer gears. And this is only a very basic part needed for a car setup. Otherwise you would have to drive at Aalborg the same setup as at Forza (Like using a fixed setup for all tracks).

        Trying different ways I decided not to implement all the learning AI in the robot but in the racemanager. The advantage: it works for all robots even the berniw!

        You know about the other championship Daniele (Polimi) is working for. I had a look at it and found, that a basic robot using deterministic computations only was driving better than most of the leraning opponents. I did send him this basic robot and indeed it worked better than most others in one of their sessions. The sensors defined at this time did not allow to drive fast, the range and the direction of the radar sensors did not fit the needs. As result the cars did not race, just drive, or let's say explore the track. This made me not enter this championships again.

        > I will not introduce it in this form
        As discussed, this form will not result in exciting races.

        > but I have another possibility in mind, maybe for 2018 then;-)
        OK, let's discuss it now, not at the end of the next season. The time to implement new features to the robots is much too small otherwise.

        What about to setup test races as soon as possible?
        Last Edited: 2017-01-20 12:58:21 by wdbee
    Author: jisham | Created: 2017-01-24 19:51:16
    Subject: Re: Banning precalculated setups and racing line optimization info?
    I'm not really adding anything to the state of the art here already, other than being a beta tester. And I probably won't have the time to even do that next year, so don't factor me into your decisions.

    If there is energy for a new idea, and it advances the community/competition here, then by all means go with it and don't let me hold you back.

    The point of this competition was always to advance the state of the art in the Torcs robotic opponents. The fact I could find enjoyment as a participant in the races and the championship was only a happy side effect.
    Last Edited: 2017-01-24 19:51:16 by jisham