The TORCS Racing Board
Username: Password: Remember Me?
Lost Password Register
Author: firechief | Created: 2017-10-30 01:04:25
Subject: Corkscrew Race Review
Here it is folks - https://imgur.com/a/eu0wt

I really enjoyed watching this race, it was quite tense and unpredictable throughout. USR have one hand on the trophy now.
Last Edited: 2017-10-30 01:05:02 by firechief
    Author: dummy | Created: 2017-10-30 07:31:34
    Subject: Re: Corkscrew Race Review
    >USR have one hand on the trophy now
    You're right. Tiger could still win both races and you could be last in both of them :)

    Thanks Andrew, a pleasure to see your review as always. And well done with your new learning algorithm.
    Last Edited: 2017-10-30 07:31:34 by dummy
    Author: wdbee | Created: 2017-10-30 14:18:32
    Subject: Re: Corkscrew Race Review
    > TORCS being quite cruel to a Mouse - it ran out of fuel just before the final corner but had plenty of momentum to get to the pit ... and TORCS eliminates it anyway!
    That's the TORCS bug I reported before the Season, but I was told, this is fixed (in mouse code). As I told you month ago: TORCS does not allow to stop (Speed = 0) before the pit is allocated by the robot, as it was done for years to avoid issues if a car is taken out of the race while the pit is allocated! You have to allocate the pit before stopping! Otherwise TORCS does not give you a chance to allocate the pit, the car is taken out of the race instead (= TORCS bug).

    > The Wolves chose to only fill 2/3rds of the tank ...
    No, this is not a decision. Driving nearly as fast as other cars easily can at this track means, the fuel consume of car 6 is to large to use three pit stops :( (See the Tiger's lap time even driving through the pitlane, which, BTW, should be punished because it is against the rule to not cut the corner at inner side of a curve!)

    > Giskard has caught Tiger 1 now & putting real pressure on ...
    On my version of the race, the USRs (and Axiomes) often drove to fast at the inner side and are damaging the opponent driving close to the outer side of the curve at end of the curve. That's the opposite of closing the door ;).
    BTW, the qualifying lap time was much slower here: 1:09.3078

    It's a pitty that we have again an unprepared (not compiling) bot without a working setup at the race, damaging many others and resulting in a random race result :(

    Wolf-Dieter
    Last Edited: 2017-10-30 14:33:20 by wdbee
      Author: firechief | Created: 2017-10-30 14:36:36
      Subject: Re: Corkscrew Race Review
      > > TORCS being quite cruel to a Mouse
      > That's the TORCS bug I reported before the season. But I was told, this is fixed (in mouse code).

      Yes I remember you reporting it, but this is the first time I've witnessed it first hand.

      > > The Wolves chose to only fill 2/3rds of the tank ...
      > No, this is not a decision.

      Its cool W-D. Whatever the reason, I thought it interesting to see a robot attempting a different fuel strategy, it was a pity in my race that Lobo ran into trouble before the end.

      > Tiger's lap time even driving through the pitlane, which, BTW, should be punished

      I agree, it should be. I can only speculate that the corner-cutting rule doesn't apply if its a pit lane, as the simulation can't differentiate between a car exiting the pits and a car driving normally. Tim was smart to spot the exception & make use of it, but I don't think its possible on any other tracks.

      > On my version of the race, the USRs (and Axiomes) often drove to fast at the inner side and are damaging the opponent driving close to the outer side of the curve at end of the curve. That's the opposite of closing the door ;).

      To be honest I didn't notice that happening. The USRs especially did have a habit of taking the 2nd-last corner too fast when correcting back to the racing line after completing an overtake, which is what caused them to leave the track - but I didn't see them doing that when there were opponents on the outside.

      > BTW, the qualifying lap time was much slower here: 1:09.3078

      Thats odd. It was under 1:09 on everyone else's race. Are you sure you're running the same version of TORCS? No changes or bugfixes to the simu code on your end?

      Last Edited: 2017-10-30 14:40:26 by firechief
        Author: wdbee | Created: 2017-10-30 18:24:20
        Subject: Re: Corkscrew Race Review
        > No changes or bugfixes to the simu code on your end?
        AFAIK no.

        But as we know from other races, the wdbees optimized on my system do not work as well at your systems. There is something different.

        > but I don't think its possible on any other tracks.
        All tracks that have the pit lane (pit entry/pit exit) at the inner side of a curve (e.g. pit entry: G-Track-1, Wheel2 ..., pit exit: G-Track-3, G-Track-2 ...), isn't it?

        > Its cool W-D. Whatever the reason, I thought it interesting to see a robot attempting a different fuel strategy
        This is, why my proposal was to define, that each car has to do more than 3 pit stops (assuming that 3 are needed at least). This way the range of possible strategies would be expanded.




        Last Edited: 2017-10-30 18:24:20 by wdbee
      Author: timfoden | Created: 2017-10-31 09:31:17
      Subject: Re: Corkscrew Race Review
      >> TORCS being quite cruel to a Mouse - it ran out of fuel just before the final corner but had plenty of momentum to get to the pit ... and TORCS eliminates it anyway!

      >That's the TORCS bug I reported before the Season, but I was told, this is fixed (in mouse code).

      Well there may be a TORCS bug, I don't know. But I do know that this wasn't the problem with Mouse, which has it's own bug. If you watch carefully when the car is moving slowly in its pit you will notice that it nearly stops, then floors the accelerator, then slams the brakes and pits. Sometimes when it nearly stops it is going slow enough that if it has no fuel it is pulled from the race, as at this time it isn't requesting to pit.
      Last Edited: 2017-10-31 09:31:17 by timfoden
        Author: wdbee | Created: 2017-10-31 12:32:49
        Subject: Re: Corkscrew Race Review
        Hi Tim,

        > Sometimes when it nearly stops it is going slow enough that if it has no fuel it is pulled from the race, as at this time it isn't requesting to pit.
        The speed limit to be accepted for pitting is set by TORCS:

        (fabs(car->_speed_x) < 1.0) && (fabs(car->_speed_y) < 1.0))

        If you drive slower than this but faster than 0 you can pit without fuel.
        If the car stops (Vxy=0) and fuel is 0, it is taken out of the race!

        TORCS should first allow to request the pit if the car stopped sucessfully even without fuel. If the car does not request the pit standing in the pit without fuel for some time, it should be taken out of the race.

        But have a look at simu.cpp SimUpdate:

        if ( ... || (car->fuel == 0) || ... ) RemoveCar(car, s);

        and engine.cpp SimEngineUpdateTq

        if (car->fuel <= 0.0) {car->fuel = 0.0;}

        To be honest, I do not see why it is still driving without fuel at all.


        Last Edited: 2017-10-31 12:32:49 by wdbee
    Author: timfoden | Created: 2017-10-31 09:27:24
    Subject: Re: Corkscrew Race Review
    Yes it was an interesting race for me too! :)

    I had planned to run the qualifying and a few laps before retiring to bed, and run the rest of the race the next day, but in the end I watched the whole race through. I just couldn't tear myself away!
    Last Edited: 2017-10-31 09:27:24 by timfoden
      Author: firechief | Created: 2017-10-31 12:52:17
      Subject: Re: Corkscrew Race Review
      That's what I really like about this season - most races have had an element of unpredictability & some have been quite gripping. A huge improvement over the processions we often saw in semi-pro mode.
      Last Edited: 2017-10-31 12:52:17 by firechief
    Author: phi | Created: 2017-11-01 02:08:58
    Subject: Re: Corkscrew Race Review
    I am happy to see that the slower teams contribute to the race's interest. They seem to function as a dispersing element of this "ancient" procession, and as opposed forces applied to the fastest ones and that will make them into better bots. All races results except Brondehach are in accordance with the order of the most powerful, so for me there is no reason to speak or to think of supposed random results derived from the action of these teams.

    As for me, I see that JSR has more potential than presented, and FA1 has improved in certain respects. The latter, however, creates huge overhead in my system with huge latency times -- this time even in loading the track blocked my TORCS very close or beyond 1 minute... --, and I do not know what to do: I like this car and its originality, I like to race with it, but I will not buy a new PC just for it, and sometimes I think about putting it out of the race... sorry.

    As for Klin, well, he has a lot to walk. He brought to Corkscrew a new yet poorly tested and tuned avoidance brake system, which makes it more aggressive. He is in fact a «dive bombing» in some extent... I think/agree...
    Last Edited: 2017-11-01 02:08:58 by phi
      Author: firechief | Created: 2017-11-01 03:26:44
      Subject: Re: Corkscrew Race Review
      JSR is a mouse derivative, so if its given proper setups it should be able to run in the midfield - maybe higher if Jacob had the time & talent to devote to it. At the moment though its basically racing with default settings.

      FA1, for me, still causes huge lag spikes when it pits - otherwise its fine. Its traffic avoidance has improved substantially over the season, which is great to see.

      I didn't see any unacceptable behaviour from Klin in the last race. Its not easy to overlap, but it doesn't ram cars off the track & tries to avoid contact - if it holds the leaders up for a while, that's a good thing & makes for a more interesting race. Its better if the leaders have to know how to overtake rather than just getting the fastest laptimes & be able to drive around on the ideal line with everyone hopping out of the way.

      I'm looking forward to the day when you start to optimize Klin for lap speed though - there's a lot of performance in the K1999 raceline code which you're not exploiting yet.
      Last Edited: 2017-11-01 03:26:44 by firechief
      Author: wdbee | Created: 2017-11-01 16:58:12
      Subject: Re: Corkscrew Race Review
      > All races results except Brondehach are ...
      You forgot the different characteristic of the tracks and the cars. Brondehach does not have a long straight, so cars with bad aero parameters do have a chance, while on others they are easily passed at the straight even by bots that are not as powerfull ;)

      I expect that a bot send to the race does compile and not have the same issues three times. And it should have a setup for the race that keeps it on track, may be slow, but it should not stop, block or damage others very often.

      On my race the axioms are blocked and kicked into the pit lane. Andrew saw this with one of my cars. The result is completely different here-> random, the JSR caused a lot of damage. With a litle work the JSR would drive well, but for me it looks like there was no try to make it drive well/compile at all.
      Last Edited: 2017-11-01 16:59:33 by wdbee
        Author: phi | Created: 2017-11-01 19:10:25
        Subject: Re: Corkscrew Race Review
        Sorry Wolf-Dieter, I'm not forgetting anything. I'm just saying that Brondehach has random results: the explanation and understanding of this is something else entirely different, and at this respect perhaps you are wright. I just rely on my calculations, and they indicate that this circuit is the only one to have such results this year, and in fact, he has had those results here for the first time since 2008.

        Race after race I calculate the glicko, since 2005. For a given race X, I have a glicko order [race (X-1)], and I put this order in correlation with the results obtained in race X applying the 'Spearman Rank-Order Correlation' test with a probability of .01 to reject the hypothesis of occurrence of randomness.

        Of course, if you want the VS2013 C# code, I'll be happy to give it to you. It's private, it's in portuguese and it's in development, but I believe this will not be an impediment to an analysis on your part. I believe that in the problem of randomness we can not rely on a subjective approach.
        Last Edited: 2017-11-01 19:10:25 by phi
          Author: wdbee | Created: 2017-11-04 14:51:15
          Subject: Re: Corkscrew Race Review
          I have to say, I do not know what glicko is ;)

          What I mean by random results is:

          We have some new bots, that currently still do not follow the rules.
          If I run a race with all bots send, I get a result.
          Some of the bots do block and damage others a lot.
          Running an additional instance of the race without one of the slow bots, I get a completely different result regarding the top ranking.

          This is why I say, the result is determined by the still incomplete bots, not by the skills of the top teams. But following the rules completely, just these still incomplete bots should not be part of the race.

          Nobody expects, that a new bot works perfect out of the box. But we should see, that the bot is maintained, bugs should be fixed, a setup should be prepared for a race. Just copying a bot without a setup and send it race by race? Hm.

          Fixing the bugs of opponents takes time, I would like to use for my own bot ;)
          Last Edited: 2017-11-04 14:51:15 by wdbee
            Author: phi | Created: 2017-11-04 18:11:07
            Subject: Re: Corkscrew Race Review
            The sanity of the race is indeed compromised, but will it be so flagrant and drastic as to completely compromise the results (read 'bot skills') and make them random? In this way we have two questions, which touch each other, but which do not represent the same thing: (i) sanity and (ii) randomness. I do not defend insanity, nor do I intend to state with what I have said that the fact that there is no randomness, there is no problem: insanity is in fact a problem of the responsibility of each owner, and a clean and healthy race is what we all want...

            And what is 'insane', that is, to what extent does a mistake become insane? Well... I speak for myself: Klin is insane (i) by his blindness in the pit lane, (ii) by his acceleration in the exit of the corners -- he will hit the car that passes him out --, (iii) by his trajectories and accelerations in the entrance and exit of the pits, (iv) by his current characteristic of "dive bombing", (v) etc.... What is this insanity really? With experience and development it tends to diminish, but will it ever disappear?

            From my point of view, to say that a result has nothing to do with the ability of the bots is somewhat exaggerated, but I understand what you mean.
            Last Edited: 2017-11-04 18:11:07 by phi
              Author: wdbee | Created: 2017-11-05 11:05:47
              Subject: Re: Corkscrew Race Review
              > to say that a result has nothing to do with the ability of the bots ...
              If I exclude one of the slow bots and run the race I get a result that shows a completely different ranking of the top ten, compared to the original race or to another race, where I excluded another slow bot.

              The fast bots and their setups are same in all this races.
              For me this shows, that the resulting ranking is not determined by the skills of the fast robots but by "when/where/how often" a robot is hampered and damaged by others. And "when/where/how often" is random somehow.

              > Klin is insane (i) by his blindness in the pit lane, ...
              > (iii) by his trajectories and accelerations in the entrance and exit of the pits ...
              This is same for most other robots as well (remember the images Danny posted some weeks ago).

              > With experience and development it tends to diminish, but will it ever disappear?
              Yes, a maintained robot will work better and better from race to race and this is fine, and you are completely right, it will never be perfect.

              I started working on my bots 10 years ago, but they still do not work as well as I would like ;).
              Last Edited: 2017-11-05 11:05:47 by wdbee
                Author: phi | Created: 2017-11-05 19:29:51
                Subject: Re: Corkscrew Race Review
                The difference between the two results (with the slow bot and without the slow bot) must be significant so that we can conclude that this bot is harmful, and you have the statistical tools to verify that meaning.


                If a WDBee compete with another similar car it is "always" straight forward; but if you add a Klin (or other), the fast cars will need to reveal a little more skill, just because the slow car is an integral part of the race and an additional and dynamic obstacle: in both races what counts is the skill of the cars; the slow car acts on both cars in the same way.

                You agree that there will always be the presence of errors, accidents, with or without slow cars. Then there will always be a random factor due to them and other things, and the amplitude of this random factor can be statistically ascertained. If a result is significantly in agreement with the expected one due to the power/skill of the cars, in reality what this means is that it is 'unlikely' to be random. In my calculations there is always the probability that a result is random, but (i) it is so low (<= .01) that I decided that this factor is negligible, or (ii) it is so high (> .01) that I give the result as random.

                Yes, «"when/where/how often" is random somehow», but this happens with or without slow cars. What can I say? everything depends on the measure of that random.
                Last Edited: 2017-11-05 19:29:51 by phi
                  Author: firechief | Created: 2017-11-05 23:48:50
                  Subject: Re: Corkscrew Race Review
                  I wouldn't necessarily say that the slower bot is harmful simply because its presence changes the results. To me, its only harmful if it behaves badly - by frequently blocking and/or ramming into the leaders when they try to overtake, by losing control & spinning off the track, then damaging cars when they merge back in, or as we saw in the last race with JSR, by failing to pit & criss-crossing the track again & again with no regard for other cars.

                  I don't see Klin doing any of the above. Early in the season it had a bug which causes it to occasionally ram an overlapper, but I haven't seen that for a long time now so I presume you found a fix. Sure, Klin doesn't necessarily make it easy for overlappers to get past. It doesn't see them coming 100m back and pull off to the side & drive at half speed until they're past. But I don't have a problem with this at all - Klin has its own race to run, and when an overlapper is right behind it & faster than it, Klin does let it through.

                  What Klin (and other robots like it) will do though is create a requirement that faster robots must know how to overtake. If all the slower cars just jumped out of the way, faster cars wouldn't need much overtaking skill at all and the race would be a boring procession based solely on which robots had the quicker lap time.

                  That's why in the past I've raised the question of getting rid of the "blue flags" rule altogether, however W-D rightly raised the issue of some robots behaving poorly in wheel-to-wheel racing - berniw_2004 in particular was quite dangerous to try and overtake as it could often fail to notice cars beside it. Nevertheless, if this issue was resolved and all robots were good at avoiding collisions, removing blue flags would I believe make for a very interesting championship where the ability to safely and quickly navigate past slower cars would be far more important than simply achieving the fastest laptime.
                  Last Edited: 2017-11-05 23:48:50 by firechief
                    Author: phi | Created: 2017-11-06 09:09:18
                    Subject: Re: Corkscrew Race Review
                    I agree with you guys in the sense that the current state of the races brings with it a surplus problem that ends up compromising seriousness, satisfaction, and all the efforts of development. There are limits to acceptable insanity, and races have not in fact been clean and harmonious. After all, if I wanted to compete in real life I would have to take a special license to do so; here this license is the query, the good sense of the initiated, and the opinions and directions of the most experienced.
                    Last Edited: 2017-11-06 09:09:18 by phi
                  Author: wdbee | Created: 2017-11-09 17:21:26
                  Subject: Re: Corkscrew Race Review
                  > Yes, «"when/where/how often" is random somehow», but this happens with or without slow cars.
                  This is not the point here. A slow car driving well does not cause such issues. You have to have the skills to overlap, but can overlap whithout being hit every time ;)

                  It is the unmaintained, bad driving bot that blocks and damages others often.

                  And if we look at the races we see the following again and again:
                  Two or more fast drivers are behind a blocking bot. The fast drivers can drive side by side without damaging the others. At some time, the blocking bot opens one side and the lucky one at this side can pass. So it is not the skills of the faster driver but the "random" event, which side will be opened first by the blocking bot.

                  Last Edited: 2017-11-09 17:21:26 by wdbee
                    Author: phi | Created: 2017-11-09 22:22:55
                    Subject: Re: Corkscrew Race Review
                    You see the punctuality of the random event, but we must ask ourselves if this event has a drastic effect on final result. This year only Brondehach has a random result, and even you interpret that result as derived from track characteristics (not because slow drivers and their errors?). The results have been largely coincident with the order of strength of the drivers, in such a way that we can not consider the results as random; we can rely on a safe margin that the USR is up front because of its skill and not because of its luck. But, who knows? perhaps all my calculations are wrong; still in this case we must ask and try to respond in a effective way.

                    Again, I'm not defending unsupported bots, nor races with little or no quality. These bots or such races are for me another type of problem.

                    By the way: I'm in a low state regarding Klin development, but I want to keep it running: does this make him a bot without support? I ask this because for me Klin have a lot of work to do for a race with quality, and, if turns to an unsupported bot, I prefer put him out of championship. I know we can not please everyone, but I entered the league with the self-condition that Klin was either tolerable to all or none, and then any opinion of any of you is important to me as long as it is sincere and meaningful. How much time should we devote to a bot to be considered as maintained? How long should we wait for an initiate to put his bot to run quality races... tolerable races? My common sense tells me that Klin is prepared to make tolerable races; that's why I entered the championship, but... I do not know, I'm just an initiate who wants to be here and improve.
                    Last Edited: 2017-11-09 22:22:55 by phi
                      Author: firechief | Created: 2017-11-09 23:43:37
                      Subject: Re: Corkscrew Race Review
                      I think there's a key difference between an unmaintained bot that behaves well, and an unmaintained bot that behaves badly. The former is completely acceptable, the latter is not.

                      If you stopped working on Klin, it'd be in the first category as it drives safely & predictably. At the moment, JSR would be in the second due to driving with a default setup that doesn't work at all tracks. Fabian Racing would also be in the second category if development stopped (slamming the door on sharp corners & freezing the game when pitting), though its certainly much improved now than it was in the early races & is quite close to being well-behaved.

                      It should be noted that even mature robots still have races where they cause trouble for other cars, and this is because its so difficult to make these things perfect & our robots are still under development. Because we're only human and make mistakes, sometimes we inadvertently introduce new bugs - it happens, but we fix them as soon as possible.

                      I do hope you're able to continue work on Klin, as there's a lot of potential in your robot that's still to be unlocked. I could definitely see it contesting for the championship one day.
                      Last Edited: 2017-11-09 23:47:05 by firechief
                        Author: phi | Created: 2017-11-10 02:06:52
                        Subject: Re: Corkscrew Race Review
                        I do not stop, but my head is in other places these days. On the day I stop, I go on vacation and retreat.

                        As for the development of Klin, I need to slow down and formulate/systematize/structure new problems, and I hope next year drop the difference to the fastest by ~5s (I think she's in the 10s or more) and, of course, allways seeking to give him more reliability and quality. Klin is not being neglected, and once he is, I am the first one to kick him out of the TRB.

                        «contesting for the championship»...? I do not know, perhaps in the hands of someone more knowing, like you, but in reality I believe that no machine will surpass the double man/machine; but this is another subject. I just know that I will do my best: to reach a second-class car, but decent and confident; I liked that you said «Klin has its own race to run», because that's what I fundamentally feel about Klin; of course: he (I) also has the need to compete...

                        Thanks Andrew.
                        Last Edited: 2017-11-10 02:06:52 by phi
        Author: phi | Created: 2017-11-01 21:55:54
        Subject: Re: Corkscrew Race Review
        About JSR compilation, I just do not understand how it can be possible that everything is fine in Jacob's system (http://www.berniw.org/trb/forum/showthread.php?topicid=5333)... To be honest, I do not understand much about this intervention from Jacob. Maybe it's my problem...
        Last Edited: 2017-11-01 21:55:54 by phi