The TORCS Racing Board
Username: Password: Remember Me?
Lost Password Register
Author: dummy | Created: 2019-12-09 02:32:52
Subject: Rat Race the champion 2019
Congratulations Tim. What a great performance and the Tigers could have even been in second place, if they'd run the default setup.
Last Edited: 2019-12-09 02:32:52 by dummy
    Author: firechief | Created: 2019-12-09 07:41:00
    Subject: Re: Rat Race the champion 2019
    Let me echo Danny's praise - Tim you've taken things to a whole new level here. And not only that, you've shared one of the tools that helped you achieve such greatness with Mouse this year & that is incredibly generous and community-spirited.

    I guess we'll all be back to contest this again next year? I think I'll retire USR as there are issues with its design that really necessitate a full rewrite, something I just don't have time for. But with your permission Danny I'll keep running Axiom as a fork of Robotec ... was pretty happy with how it performed in the last race, and I'll see where my hacks & tweaks can take it. I must say you've done a great job with it, especially its handling in traffic compared to the previous years' dummydrivers/dandroid.

    Lastly, we'll need to come up with a proposal to Bernhard that we agree upon for the rules next season - lots of good ideas raised recently. I'll create a separate thread where we can discuss them?
    Last Edited: 2019-12-09 07:41:00 by firechief
      Author: dummy | Created: 2019-12-09 08:31:31
      Subject: Re: Rat Race the champion 2019
      Permission is hereby given ;) Thanks for the praise.
      Last Edited: 2019-12-09 08:31:31 by dummy
      Author: timfoden | Created: 2019-12-09 09:33:51
      Subject: Re: Rat Race the champion 2019
      Thanks Andrew,

      > whole new level

      I think this year was just serendipity... my robot seemed to work well with the increased tyre wear and longer pit times. I believe that if the rules had been more similar to the 2017 ones that mouse would just have gained parity with the other fast bots from that season.

      > shared one of the tools

      Only found that CMA thing a few weeks ago, and only used it to optimise Mouse's driving line for the final race (with no manual additions -- might have been able to squeeze a smidgeon more speed out if it with that.) That said it did really well on e-track-2. For the previous races I was still machine optimising the driving line (with my old slow search search algorithm) but could only handle a much smaller number of parameters in a reasonable time, and in addition I was also supplying some manual changes to this driving line. In the 2017 season I was doing this machine optimising using a 16 core server that was left running for days at a time... and it didn't help me reduce that 1...1.5 seconds deficit. :)

      The things I changed in mouse for this year were basically:

      1. Calculate better speed values for the racing line. Errors here either cause: running too slowly (speed could be higher), or running off line in some way (speed is too high), resulting in a slower lap time than is possible.

      2. When controlling the robot, try to achieve the speed specified by the racing line calcs as closely as possible, and also to steer so as to stay on this line as closely as possible.

      3. Generate a better racing line (either using a machine optimisation, or manually, or a combination.)

      I know that my bot is definitely not optimal at any of these 3 tasks, so there's still hope to beat it. :)

      Cheers, Tim.
      Last Edited: 2019-12-09 09:33:51 by timfoden
        Author: phi | Created: 2019-12-10 10:54:17
        Subject: Re: Rat Race the champion 2019
        I don't want to ignore the effect of optimization on the competitive quality of the Mouse, but this is a bit secondary to me. I think what really counts is the extreme adaptive capacity of the car, its flexibility and maneuverability, which makes it an entity capable of overcoming the most diverse circumstances that cause problems. It seems that the Mouse is a "small" piece of machinery that goes into the circuit equipped with some prior knowledge, yes, but what is evident in it is the ability to work on this circuit without waiting for "this or that" to happen there: he goes prepared for whatever could happen, we might say. Perhaps only this adaptability and flexibility will not win a championship by itself, but I believe that optimization alone does not win it, no matter how powerful it is; optimization goes only as far as the flexibility of its integrating code allows it.

        The soul of the Mouse is something that the purest algorithm doesn't capture. It's the way to structure and chain all events after TORCS's appeal: Drive!

        Congratulations, Tim, for the championship, and all the robot work.
        Last Edited: 2019-12-10 10:54:17 by phi
          Author: firechief | Created: 2019-12-10 13:05:43
          Subject: Re: Rat Race the champion 2019
          Optimisation alone won't win if it doesn't give a distinct pace advantage on the competition. But a driver that doesn't make mistakes, avoids collisions and off-track excursions, takes care of the tires and is reasonably good at overtaking backmarkers - just a solid robot, nothing special - and has a second a lap advantage due to optimisation, it will indeed win championships. That's how USR/Hymie won in the past, not by necessarily being a better & more adaptive driver, but simply being quicker on most tracks. It qualified fastest and just drove away from the field.

          This is why some of us have argued for reverse grids, as it rewards the ability of robots to cleanly overtake and fight through traffic, rather than rewarding lap pace alone - and with that type of championship we could well see the more adaptable robots coming out on top.
          Last Edited: 2019-12-10 13:06:21 by firechief
            Author: phi | Created: 2019-12-10 15:16:04
            Subject: Re: Rat Race the champion 2019
            If we have a pure power and sprinter championship the optimization will make all the difference, surely; however, the same algorithm will have different results depending on the robot's ability to receive and manage it. If we have a championship like this year, where there are more variables (not just "power and sprint"), optimization remains important, of course, but it will lose more ground to the car's adaptive ability; and, as I said before, the detail of reversing the grid is quite interesting, should be tested, and, of course, further improves the adaptive quality of cars and its importance over pure power; after that we can go back to the classic grid with the certainty of having better cars.

            Well, I suppose we're talking about the next championship. My question is: which of the "skills" should TORCS care about? the sprint in straight line, or the beauty of getting in first place in twisting lines (the life lines)?
            Last Edited: 2019-12-10 15:16:04 by phi
              Author: firechief | Created: 2019-12-10 15:20:45
              Subject: Re: Rat Race the champion 2019
              For me, I'd say the beauty of "twisting lines" is far far more interesting. We've had sprinting for the last 15 years of the TRB.
              Last Edited: 2019-12-10 15:20:45 by firechief
    Author: timfoden | Created: 2019-12-09 09:14:15
    Subject: Re: Rat Race the champion 2019
    Thanks Danny, although I think you're being a bit generous to the the plain mouse... I didn't tone the Tiggers down that much. :)
    Last Edited: 2019-12-09 09:14:47 by timfoden