The TORCS Racing Board
Username: Password: Remember Me?
Lost Password Register
Author: firechief | Created: 2016-11-22 00:12:44
Subject: Dirt-2 Review
Firstly, I need water. So much dust from that track choking up my throat that its hard to breathe let alone type this post. My cameras did, however, manage to capture enough material for my customary photo-review of the race...

Just browsing through the results, of which there were only 3, its really odd how completely different things were on wdbee's computer. I get the feeling that maybe we didn't have the same setup for wdbee_2016, as there's no way it was good enough on my machine to take 1st and 2nd. Though this track certainly does throw up odd results.

Anyway it looks like, barring a complete catastrophe in the final race, USR has sown up the championship. The glorious e-track-2 track will therefore hopefully see 2 new teams debut in preparation for next season, should be a lot of fun. If there's any aspiring robot authors out there, why not get your bot involved for the last race of the year!
Last Edited: 2016-11-22 00:13:01 by firechief
    Author: wdbee | Created: 2016-11-22 13:14:18
    Subject: Re: Dirt-2 Review
    Well, compare the USR qualifying time on my system, it was much slower here. So getting the pole was possible. This made the USRs crash in the first lap as the Hymies on your system, and the big crash was starting just behind LosLupos, damaging all the jumpers heavily. In race, all faster teams (all jumpers) did not make it to the end of the race.

    BTW: You talk about "Why drive around an opponent, if I can jump over it!"
    In fact you cannot jump over an opponent, because the simulation does handle it as crash (It is 2D only). So the well driving team in front gets a lot of damage too. You only can jump along the side of it.

    Setting up a car this way is intentional, so according to the rules, the result on my system was OK for me and I did not try to find the car or TORCS code that is causing the infinite loops here and at the jedis system.

    As it worked without all the jumpers after a small number of laps, it should be one of them. The berniws did work well here.


    Last Edited: 2016-11-22 13:14:18 by wdbee
      Author: firechief | Created: 2016-11-22 13:53:34
      Subject: Re: Dirt-2 Review
      Yes, that speed variance is quite confusing. I wish I knew the reason for it. Also in my race, most of the jumpers made it through to the end, with non-jumping teams just as likely to have a car caught in the middle of a pile-up as the jumpers.

      Certainly the result on your end is fine, nobody would assume you're doing something to cause loops! This is, after all, just a fun competition where we each chase perfection for our own robots in our own ways.

      Personally I'd be happy not to see dirt-2 again for a long while, even though I know Bernhard likes it ;)
      Last Edited: 2016-11-22 13:53:34 by firechief
    Author: jisham | Created: 2016-11-22 13:55:59
    Subject: Re: Dirt-2 Review
    Nice review :-)

    Sorry I wasn't able to submit a result. Not for lack of trying.

    And I will re-emphasize the encouragement of new robot authors. Running this championship is taking too much of my time from work, and I feel I am not adding much of anything useful by running old bot code, other than variety. I will likely go back into retirement after this season and open up a grid spot for someone with more energy and a more creative bot design to add more life to the championship.

    Last Edited: 2016-11-22 13:55:59 by jisham
      Author: firechief | Created: 2016-11-23 02:27:00
      Subject: Re: Dirt-2 Review
      I can understand your reasons for wanting to retire again - the time consumption was what made me stop a few years ago. You'll certainly be missed if you don't return!

      If you have second thoughts, might I suggest using one of the more current robots instead of ancient simplix?

      I'd love to see more competitors, and I'm sure there'd be programmers out there who'd take part if they knew about it. That's the thing - Torcs is an old game, and TRB is hardly advertised anywhere. We'd need a way to promote the TRB if we want more people to join in.
      Last Edited: 2016-11-23 02:27:00 by firechief
        Author: jisham | Created: 2016-11-23 13:32:05
        Subject: Re: Dirt-2 Review
        Well, let's say I'm pulling a Jenson Button and not necessarily retiring, and you never know when I might show up as a reserve driver :-)

        Anyone offering a more current robot? I'm still pretty attached to car2, unless a bavaria2 becomes available :-)

        Last Edited: 2016-11-23 13:32:05 by jisham
          Author: firechief | Created: 2016-11-23 15:02:20
          Subject: Re: Dirt-2 Review
          As you've probably noticed from the leaderboard, car2 is currently the fastest - no need to switch ;)

          All of the robots are published under the GPL, so you're free to use any of them. There's a "gentlemen's agreement" written into the competition rules that we don't borrow code from other robots while the comp is still running, but anyone can do this during the off-season.

          You could if you wished take an Axiom/Sieger (both updated versions of Mouse), a DanDroid or a (God help you as this robot is hellishly complicated) USR and field your variant of it next season. I'm even doing that for the final race, with W-D's permission as the season is still running, introducing a new bot based on the current wdbee_robotics.
          Last Edited: 2016-11-23 15:02:20 by firechief
            Author: jisham | Created: 2016-11-23 16:15:12
            Subject: Re: Dirt-2 Review
            I just didn't want to needlessly overload any car/code combinations. I guess it only eliminates one from the pool of possible choices... albeit a good one.

            I'll try to take a look at various code over the off-season and see if anything in particular catches my fancy. Who knows?
            Last Edited: 2016-11-23 16:15:12 by jisham
              Author: firechief | Created: 2016-11-24 06:27:37
              Subject: Re: Dirt-2 Review
              Well, if I may make some suggestions, the enhanced "Mice" (Axiom/Sieger, which have identical code) are easy to configure & the code structure is quite clean although Tim did use obscure abbreviations for his variable names which makes understanding various sections of it quite difficult.

              Alternately, there's the new versions of Wolf-Dieter's robots, including the one I'll introduce next race where I spent much of last weekend and several lunchtimes this week making my customary fiddles & tweaks, which I'm really looking forward to seeing in action. The wdbee code is very clean & neat indeed.

              I haven't looked at DanDroid for quite a while, but I do believe Danny's added some kind of raceline learning/optimization which helps him come up with the ideal configuration for each track. My robots are all setup manually, which is what I'm comfortable with but which can be time-consuming.
              Last Edited: 2016-11-24 06:28:24 by firechief
    Author: dummy | Created: 2016-11-22 17:00:51
    Subject: Re: Dirt-2 Review
    Thanks for the review Andrew. It was certainly not easy to select the best moments from a 6 hours marathon.

    Even though we had again very few results, I feel that USR absolutely deserves the win. The tactics with many pitstops was the right one here, the
    Droids were too optimistic with 6000 damage points pitstops.

    I'm also one who doesn't really like this kind of track, but it had one advantage. I could test and fix some collision bugs.
    And it should be possible to find the hanging bugs, if all of us would run the races in debug mode.

    John, if you have time to run the race a few more times, that would be great. But first I want to introduce bugs in my bot and give you instructions how the debugging should work.
    Even as I got one hanging race in testing, it seems to be too stable now on my setup.

    And I hope you can stay with us next year. We are just too few racers at the moment.


    EDIT: I tested how I got the backtrace of gdb the last time.

    Start torcs with -d in my case:
    $ ~/torcs/bin/torcs -d

    When Torcs is hanging open a new terminal and do:
    $ ps aux | grep "torcs-bin"

    This resulted with:
    daniel 3202 3.5 2.1 98796 66304 pts/1 S+ 17:57 0:00 gdb /home/daniel/torcs/lib/torcs/torcs-bin
    daniel 3204 88.2 1.9 414536 61172 pts/1 RLl 17:57 0:18 /home/daniel/torcs/lib/torcs/torcs-bin -l /home/daniel/.torcs -L /home/daniel/torcs/lib/torcs -D /home/daniel/torcs/share/games/torcs
    daniel 3215 0.0 0.0 12728 2144 pts/2 S+ 17:57 0:00 grep torcs-bin

    Then kill the process, in this case it's 3204 with 88.2 % CPU usage:
    $ kill 3204

    Then the backtrace appears in the torcs terminal. Good luck!
    Last Edited: 2016-11-22 19:09:49 by dummy
      Author: wdbee | Created: 2016-11-22 23:24:03
      Subject: Re: Dirt-2 Review
      > And it should be possible to find the hanging bugs,
      > if all of us would run the races in debug mode.
      Well, the issue is, that with windows not all robots can be used in debug mode (The old mixed new/free or alloc/delete bug).


      Last Edited: 2016-11-22 23:24:03 by wdbee
      Author: jisham | Created: 2016-11-23 13:29:04
      Subject: Re: Dirt-2 Review
      There's the bit I missed. I tried ^C, ^Z to no avail, and even sent a sighup which promptly killed it (might have been before --enable-debug).

      I will try again with a vanilla kill.
      Last Edited: 2016-11-23 13:29:04 by jisham
    Author: dummy | Created: 2016-11-23 10:00:58
    Subject: Re: Dirt-2 Review
    >A brief musical interlude where the USR team show off their elegant dance moves

    ..and a car flying over the musicians to avoid a blood bath.

    You are a great artist, may I say.
    Last Edited: 2016-11-23 10:00:58 by dummy