The TORCS Racing Board
Username: Password: Remember Me?
Lost Password Register
Author: firechief | Created: 2019-11-18 01:29:38
Subject: Alpine-1 -> Rats dominant (again)
I have no idea how the rest of us are going to catch the Ratmobiles with the current TRB rules. Not only is it bloody fast, it can run at full speed during a race and still 4 stop, which none of the other robots can match - it's like Rats are operating under their own laws of physics! I guess life would be boring if we didn't have any challenges :)

I do have a suggestion for next year, and this should be the topic of a separate thread I guess. Reduce the time for tyre-changes by 5 seconds or so. Even before Tim's team started to assert its superiority, most races favoured the 4 stop over the 5+, so I think the pit stops being a little shorter might help the balance a little. As would allowing changes to front wing angle :)

So we're down to the last race. I wonder if the Ratmobiles will be able to 4 stop on this one at full pace - the nature of e-track-2 might make this difficult.

Oh and Danny, speaking of challenges, I just thought you should know that Axiom is going to make a cameo appearance - running the old Dummydrivers code. It'd be a terrible shame if it was able to beat Robotec to the finishing line wouldn't it ;)
Last Edited: 2019-11-18 04:50:49 by firechief
    Author: dummy | Created: 2019-11-18 08:58:20
    Subject: Re: Alpine-1 -> Rats dominant (again)
    >Rats dominant (again)
    Who would have guessed? I thought Forza would have been an opportunity, but I guess not ;)

    Looking forward to seeing Axiom again. They were really missing this season. And if they beat Robotec there will be no hard feelings, as the Robotec code is much more beautiful (and solid) than DummyDriver's. In preparation for this season I ran the last race of 2017 (e-track-1) with car-3 and Robotec was winning consistently against the 2017 opponents, thus I was confident to have a competitive bot for 2019, but then came the rule changes and especially this untouchable Mouse bot.
    Last Edited: 2019-11-18 08:58:20 by dummy
    Author: timfoden | Created: 2019-11-18 22:45:52
    Subject: Re: Alpine-1 -> Rats dominant (again)
    Coincidentally I have recently run some 500 km practice runs on e-track-2 with (an only slightly optimised) mouse with different pitting frequencies, and the best time attained was for 4 stops. But it beat the best 5 stop time by only 5.51 seconds. :)

    Thus I guess your change to 30 secs would even this out... but it's not really that much of a difference all told.

    I think allowing front and rear wing, brake balance, and diff gear ratio would be my choice for minimal car tinkering for next year.
    Last Edited: 2019-11-18 22:45:52 by timfoden
      Author: firechief | Created: 2019-11-19 00:28:34
      Subject: Re: Alpine-1 -> Rats dominant (again)
      The real test has to be done with optimised setups though, as then you're testing a 5 stop at a very fast pace versus the 4 stop. With your robot this is still difficult, as it seems capable of 4 stopping on many tracks while running at full speed, something none of the other bots can manage.

      Reducing the time needed to change tyres would allow other robots to compete provided they could lap faster than the Ratmobiles, which is enough of a challenge in itself! It'd be easy to just say the rest of us have to improve our tyre wear, but that is extremely difficult to do and depending on the core of the raceline/driving code we're each using, may well be impossible especially given we all have time constraints. Improving the chances of 5 stopping would seem the next best option.

      I agree with the adjustable parameters, totally forgot about brake balance :)

      Last Edited: 2019-11-19 00:29:26 by firechief
      Author: wdbee | Created: 2019-11-19 06:10:54
      Subject: Re: Alpine-1 -> Rats dominant (again)
      Congratulation Tim!

      +1 (allowing front and rear wing, brake balance, and diff gear ratio, as stated at start of the season ;)

      What about a weight handicap depending on the number of points a driver got in the current season?

      What about using two car types (car1 + car2) next season by each team (Getting rid of these unmanaged teams but increasing the number of cars racing)?
      Using the car of the last season and the next in the list would give additional information about the differences and reduce the workload. For new teams it would be a chance to use a last season bot to start into the TORCS TRB.

      And use the 2019 TORCS for 2020 season unchanged (physics), because we all are limited in time.
      Last Edited: 2019-11-19 06:10:54 by wdbee
        Author: firechief | Created: 2019-11-19 06:44:00
        Subject: Re: Alpine-1 -> Rats dominant (again)
        Ooh a weight handicap! That'd be cool, though I've no idea how it could be implemented. It does feel a bit like punishing a team for being better than everyone else...

        If we take that path, an easier alternative is a reverse grid based on the results of the previous race, with only the first race of the season using qualifying, and any new mid-season entrants starting from the front of the grid in their first race. A team that manages to get a clear points majority at the end of the season with that setup will truly deserve their crown. I'd still back Tim's team to win, if the robots are as they are now.

        Your idea of a choice of two cars, with one being the previous year's, is excellent too.



        Last Edited: 2019-11-19 06:44:00 by firechief
          Author: wdbee | Created: 2019-11-19 17:03:17
          Subject: Re: Alpine-1 -> Rats dominant (again)
          A weight handicap is\was used in some real world races while a reverse start grid is not AFAIK.

          Starting behind a klin, what a nightmare!

          As long as we have the current rules to distribute the damage points between opponents, this would be no fun. Closing the door while an overtaker is aside will result in double damage points for the overtaker.

          A weight handicap based on the cumulated points would keep the races more exciting over the whole season. Just using the result of the last race is not the same, balanced bots would result in just be at front/at the end in each second race.

          We would get much more random results caused by damages.

          Last Edited: 2019-11-19 17:04:20 by wdbee
            Author: firechief | Created: 2019-11-20 00:46:15
            Subject: Re: Alpine-1 -> Rats dominant (again)
            Offhand, both Formula 4 and 3 use reverse grids, and I've known of some touring car series that do the same. Even Formula 1 was toying with the idea for next year.

            I've never found Klin to be much of a problem to get past, nor Berniw for that matter - Tim's robots have little difficulty with them either.

            That said, there are some robots that are quite hard to get past due to the introduction of various defensive blocking code in several of them. In fact I recall Daneel being stuck behind a wdbee in the last race on alpine-1 and being rammed into the wall the moment it got alongside.

            There is the rule about trying to avoid collisions with opponents, and sometimes defensive tactics go too far in this regard - if this is the case, we have the rule in place and it can be addressed.

            A robot that can overtake imaginatively and cleanly will prosper with reverse grids, whereas a robot that's not as good at it may well find itself getting stuck and falling behind. I don't see the problem with this - overtaking is an essential part of racing.
            Last Edited: 2019-11-20 00:48:47 by firechief
              Author: wdbee | Created: 2019-11-20 21:31:31
              Subject: Re: Alpine-1 -> Rats dominant (again)
              As long as all bots are working well and follow the rules, you might be right.
              But looking to the very different implementations of the let pass code, I do not think so. If a bot allows let pass only if not in curves and not driving fast and not driving behind another, the let pass code does not help. At Alpine-1 most of the track is switching of the code. I expect similar issues with the inverse grid situation.

              It remembers me a bit to your definition of the minimum of parameters to be allowed for tuning. At the end of the season you came up with the statement that there should be two additional. Looking to the posts you would see, that Tim an I said same before the season ;).



              Last Edited: 2019-11-20 21:31:31 by wdbee
                Author: firechief | Created: 2019-11-21 01:42:23
                Subject: Re: Alpine-1 -> Rats dominant (again)
                The Let Pass code is only for overlapping, so it has little effect on a reverse grid where the faster robots have to fight for position rather than being waved through.

                Defensive blocking code might be an issue though, as I pointed out earlier. If robots get too aggressive with their blocking, then firstly your excellent proposal to adjust how damage is calculated could help, and secondly we could come up with tighter rule definitions on what is acceptable and what isn't - the same for overlapping for that matter.

                I guess it might be easier to persuade Bernhard on the damage idea if someone put the work in to implement the change & test that it works correctly with a wide range of collisions. It'd be much easier for him to roll a patch into the next release than to find the time to code the idea himself.
                Last Edited: 2019-11-21 01:42:23 by firechief
                  Author: phi | Created: 2019-11-21 07:52:07
                  Subject: Re: Alpine-1 -> Rats dominant (again)
                  Why not abolish the code/rule "Let Pass"? No robot is absurd when another car is trying to overlap it -- same for the code "Defensive blocking", I guess. None make an intensive block, and this intensity (or intent) is easily verified in the code: if any of them do so, it is disqualified and will not be able to run. To let another car pass requires some skill, and, like the ability to walk on traffic or anything else, it is just something to be developed; and, on it, one has to distinguish between good intentions and less good intentions. Once again, these intentions are in the code, not what we seem to see when watching the robot in progress. For me, here we do not need "perceptive rules" but "analytical rules".

                  So, as I see it, we don't need to reverse the grid to get the fastest robots into active battle from the start of the race to the end; we can reverse the grid for other reasons, not for that. When passing slower cars, what's the difference between doing it at the beginning (overtaking them) or mid-race (overlapping them; without the current rule, note)?


                  And reversing the grid as incentive to new racers (cf Andrew: 2019-11-21 01:35:00), well, speaking for myself as a new runner, that doesn't encourage me at all. If my car were a top car, then yes, reversing the grid would bring me something more.


                  I think the issue of grid reversal needs to have other contours.
                  Last Edited: 2019-11-21 07:52:07 by phi
                    Author: firechief | Created: 2019-11-21 09:12:01
                    Subject: Re: Alpine-1 -> Rats dominant (again)
                    The reasons for the existence of the Let Pass rule for overlapping and the non-existence of rules prohibiting blocking are simple. They reflect how racing is in the real world.

                    There is indeed blocking code in both Dandroid (particularly last year's version) and USR, and if last race was any example, wdbee is very good at it as well. There is absolutely nothing wrong with this, as a driver should be able to defend its position provided it does so without causing damage to the would-be overtaker.

                    I don't care hugely about the Let Pass rule myself. If it disappeared then it wouldn't be a great loss imho, depending on the track - some tracks can be nightmares to overtake on. However an overlapped car should certainly not try to block an overlapper.

                    If the prospect of getting a start from the front of the grid wouldn't encourage you as a new entrant, I'd have to ask what would? Certainly starting from the back all the time doesn't seem to do much for most new robot authors. And what do you mean by "other contours"? It's one thing to pour cold water on an idea, but please bring alternative suggestions yourself.
                    Last Edited: 2019-11-21 09:12:01 by firechief
                      Author: phi | Created: 2019-11-21 10:51:51
                      Subject: Re: Alpine-1 -> Rats dominant (again)
                      Please don't get me wrong. I'm just saying that there is no big difference between a reverse grid and a classic grid without the "Let Pass" rule conditioner; so my question from the second paragraph above. I'm not putting cold water on the idea, and my suggestion is simply this: why not abolish this rule?

                      By "abolishing this rule" I don't mean that cars can block the fastest, and whether or not they block should be seen analytically in the code, not in what it seems to be through their behavior. If there is a defensive block in normal situations, here (overlaped) should not exist. The question here would be to check if the code is common sense in a consensual sense for all TRB participants.


                      Encourages me to see myself first if I had been quick enough to be there. Moreover, what brings flavor to my presence here is the state of development of my robot. It is one thing to participate only by participating, but another is to participate actively with the intention of reaching further: the latter is my attitude, and what encourages me is to see that nothing here is static and dead.


                      Like I said, a reverse grid would bring me more pleasure if I had a top car. And this is not to say that the idea is not good.

                      Yes, I believe that for some new racers the contour is correct, but not for me: for me its contour is situated in the development of traffic management capabilities in top cars.
                      Last Edited: 2019-11-21 10:51:51 by phi
                        Author: phi | Created: 2019-11-22 19:15:25
                        Subject: Re: Alpine-1 -> Rats dominant (again)
                        ... [Lack of Conclusion] ... I like the idea. I think it's very interesting observe the fight to get at front. The idea deserve at least has to be tested.
                        Last Edited: 2019-11-22 19:15:25 by phi
            Author: dummy | Created: 2019-11-20 02:49:18
            Subject: Re: Alpine-1 -> Rats dominant (again)
            >As long as we have the current rules to distribute the damage points
            I think this can be fixed rather easily. I imagine a rule where you draw a line between the opponents and calculate the damage proportional to the speeds along this line. If you move towards your opponent you get damage, if you move away you get none.

            >balanced bots would result in just be at front/at the end in each second race
            I don't understand this sentence. There would be no second race on each track.
            Last Edited: 2019-11-20 02:50:06 by dummy
              Author: timfoden | Created: 2019-11-20 09:53:40
              Subject: Re: Alpine-1 -> Rats dominant (again)
              As regards your suggestion for assigning damage... I don't understand how this would work, perhaps you could try to clarify it a bit more? Seems to me that each car relative to the other along the line between them would have the same speed, but in opposite directions.

              Edit: Ah, perhaps you mean to take the component of the cars global speed vector along the line between them? I'll have to think about the cases.
              Last Edited: 2019-11-20 09:59:38 by timfoden
                Author: dummy | Created: 2019-11-20 10:11:36
                Subject: Re: Alpine-1 -> Rats dominant (again)
                Yes I thought about the global speed along the vector, but have to admit it's just a first idea and have to think about it a little more as well.
                Last Edited: 2019-11-20 10:11:36 by dummy
              Author: wdbee | Created: 2019-11-20 21:12:38
              Subject: Re: Alpine-1 -> Rats dominant (again)
              If you just use the last race to define the grid and all bots are skilled same, the expected ranking would just be the opposite of the last race ;)
              Last Edited: 2019-11-20 21:12:38 by wdbee
              Author: wdbee | Created: 2019-11-20 21:17:40
              Subject: Re: Alpine-1 -> Rats dominant (again)
              Some time ago I proposed to use the lateral speed of the cars related to the track.
              Keeping the line would not be punished, moving to the side to the opponent would result in damage.

              This way the bots closing the door while an opponent is aside would be punished, at straights and in curves.
              Last Edited: 2019-11-20 21:17:40 by wdbee
            Author: phi | Created: 2019-11-21 07:50:55
            Subject: Re: Alpine-1 -> Rats dominant (again)
            Klin's "nightmare" quality is directly proportional to the opponent's lack of skill. This is clearly demonstrated in the races he has done.
            Last Edited: 2019-11-21 07:50:55 by phi
        Author: dummy | Created: 2019-11-19 09:33:07
        Subject: Re: Alpine-1 -> Rats dominant (again)
        >What about using two car types (car1 + car2) next season
        Good idea to generate more traffic, but I'm not sure about running last years code.

        And I like Andrews, now very old, reverse grid idea and hope next year we're finally seeing it implemented.
        Last Edited: 2019-11-19 09:33:07 by dummy
          Author: timfoden | Created: 2019-11-19 13:34:26
          Subject: Re: Alpine-1 -> Rats dominant (again)
          I agree with both your points here.

          Reverse grid sounds fun. A lot of my overtaking testing over the years has been with Quick Races where I make the grid ordered with the slowest cars at the front, and put the mouse bots right at the back. It can result in some big pileups in the first few corners/laps (and hence led mouse's current way OTT unstuck code.)

          Would reverse grids need changes to torcs or could we repurpose an existing race format to cope (I'm thinking how to share pits between the correct team-mates if a race setup has some essentially random order of bots)?
          Last Edited: 2019-11-19 13:41:18 by timfoden
            Author: firechief | Created: 2019-11-19 14:00:58
            Subject: Re: Alpine-1 -> Rats dominant (again)
            I don't know that it'd need a change of Torcs code. For the first race it'd be a normal Endurance race obviously, and for the remainder it could be a Quick race using a config file that we download from the server here along with the robots, with the starting order all determined & the number of laps set correctly etc.

            Btw I've worked with USR in much the same way over the years, trying to figure out how to get past all manner of opponents as quickly as possible.
            Last Edited: 2019-11-19 14:01:34 by firechief
              Author: timfoden | Created: 2019-11-19 14:10:33
              Subject: Re: Alpine-1 -> Rats dominant (again)
              Without running torcs up I can't remember for sure, but I thought that when pit sharing was on torcs just allocated pit 1 to the first 2 cars in the race config file, and then pit 2 to the next two, and so on. Thus if the race order was set for a reverse start, differing team's bots would be sharing each pit.

              Perhaps I'm wrong about this though! :)
              Last Edited: 2019-11-19 14:10:33 by timfoden
                Author: firechief | Created: 2019-11-19 15:05:37
                Subject: Re: Alpine-1 -> Rats dominant (again)
                Ah yeah you're probably right. Still it shouldn't be hard to change to using the Team name values instead.

                If the way it's done now is simply on grid order our current races could easily get screwed up - all it'd take is a team having cars qualify with different lap times, as indeed mine did on Forza in some people's running of the race.

                Edit: The reverse grid could also be ordered based on team rankings, not individual driver rankings - this would keep the cars neatly lined up together.
                Last Edited: 2019-11-19 15:21:20 by firechief
                  Author: timfoden | Created: 2019-11-19 15:52:57
                  Subject: Re: Alpine-1 -> Rats dominant (again)
                  Nope, I was wrong... looking in the source (raceinit.cpp line 456 or so) it looks like it tries to only assign teammates to each pit.
                  Last Edited: 2019-11-19 15:52:57 by timfoden
    Author: wdbee | Created: 2019-11-20 21:35:16
    Subject: Re: Alpine-1 -> Rats dominant (again)
    The main idea to use a weight handicap is, to make it easier for new racer to keep the gap smaller. Always be lots of laps behind will make them leave the races.
    The inverse grid would not help here.
    Last Edited: 2019-11-20 21:35:16 by wdbee
      Author: firechief | Created: 2019-11-21 01:35:00
      Subject: Re: Alpine-1 -> Rats dominant (again)
      The trouble is that the TRB isn't attracting new racers, and when we do get some it's rare that they're willing to put the required work into making their robots competitive. I doubt handicapping the leaders would change this, but if it did, an inverse grid would surely help. The new racers would see that their cars are starting toward the top of the grid in the next race and would have an incentive to improve the robot over the next few weeks to try and make the most of the opportunity. That wouldn't happen with a weight penalty.
      Last Edited: 2019-11-21 01:35:00 by firechief