Starting with RARS
  •  I don't know anything about programming
  •  Adding robot drivers
  •  Status of RARS
  •  RARS source code

  • Understanding RARS code
  • Track formats and track design
  • Car flashing
  • Alpha, vc and distances
  • Understanding s.nearby

  • Writing your own robot
  • Debugging your robot code
  • Optimization of the robot

  • RARS Wishlist

    Starting with RARS

    1. I don't know anything about programming. This was MY first question in RARS: Unfortunately I have never learned anything about programming. I would like to start learning it on basis of programming a driver, but don't know where. What files should I have in my working directory and what exactly should be my command line on these compilers to make my robot work as one of the drivers in rars?
    One generally makes a makefile... They are included in current distributions In working directory you should have all files except track files which must be in track subdirectory. Copy files from robots directory and your system directory (unix, dos, or vc) to main working directory. You might need to edit library files path in file called Makefile. If you are on UNIX, find out where your X libraries (libX11.a or are. Mine are in /usr/X11R6/lib so I have added -L/usr/X11R6/lib to Makefile. In DOS makefile lines containing CC, TLIB, TLINK, LIBPATH, INCLUDEPATH should be edited according to your compiler and library files location. Then you would just type make.

    2. Adding robot drivers
    I downloaded some more drivers from WWW site. How can I compile them into Rars? How can I add my own driver?
    For changing drivers you should edit file drivers.cpp. You should replace one of the existing robot's name (and/or colors) in robot's list AND in drivers array. Robots name can be found in robots file on line con_vec ROBOT_NAME(situation &s) It might be different from what robots shows on screen. Also, change robots filename in makefile. That's all. Nevertheless, if you are ADDING robots to drivers[] array, you should increase constant MAXCARS in the beginning of file car.h

    I have programmed a car nearly from scratch which seems to be pretty efficient, it nearly finishes all tracks at third position which seems to be quite good. Could it be possible to download the latest versions of other players in order to compare its car against others?
    Most of the robots are available from my robots page. Some additional robots are also in FTP site.

    3. Status of RARS
    Could you tell me what is the current status of RARS?
    Status is active :-), with one current season on F1 tracks and several occasions of other competitions. Check out announcements about current racing series.

    I have heard of a 3 dimensional version of Rars, does it really exist ?
    If you mean 3D-graphics of RARS then check out Marc Guery's page. It looks nice, but it is not universally ported to all platforms (needs OpenGL).
    3D model for track description was planned as a part of RARS 2 (see next question).

    What is the status of RARS 2 (OORARS)?
    Here is a short comment from Kim Laurio:

    >There has not been any substantial additions since the progress
    >report was released to the public at 7 January 1997.
    >This page should give you
    >a pretty good picture of the current status. No coding was done
    >(except for some minor tests to evaluate the CPU-usage of some
    >I'd suggest that you'd at least take a quick glance on that text to see
    >where we were and what we thought of. It is not part of any 'overall'
    >continuity/design plan of RARS, it's just a bunch of ideas that you are
    >free to use or not. For a good introduction to tire physics,
    >I'd recommend that you read the short text by David Sparks;
    >that should get you up and running in no time.
    >If you have any questions, I'll do my best to answer them.
    >Although I certainly don't have time to do any work with coding,
    >feel free to send me any ideas that you want feedback on.
    >Here is the main page for OORARS.
    >Good luck!  Kim Laurio
    4. RARS source code
    To which platforms have Rars been ported ?
    Current version is for DOS (Borland), Linux, Windows(MS Visual C++).
    The "UNIX" version uses X Windows for display, and has been known to run on Ultrix, SunOS, OSF/1 (DEC Alpha), Linux, and probably a few more that I can't think of right now.
    Older or different versions exist for OS/2, DJGPP, AIX, Symantec, Amiga, Macintosh, Watcom, Java.
    For those old versions it is probably easier to start porting from latest DOS or UNIX version rather than add changes to old port.

    AIX: tkn@VNET.IBM.COM (Thomas K. Netousek):
    I have done a "port" of the linux version to AIX.
    It is a matter of some #include's #ifdef'd AIX

    Amiga: (Marcel Offermans):
    Amiga version Release 3, based on the RARS 0.60 sources.
    Special requirements:

  • AmigaOS Release 3.0 or better.
  • The SAS/C 6.55 compiler to recompile the sources.
  • An FPU is highly recommended as it will speed up the simulation considerably.
    This port can be obtained from the normal FTP site, or from AmiNet:

    DJGPP: (Dave `Gizmo' Gymer):
    I've ported RARS 0.61 to DJGPP 1.12m4 (GNU C/C++ for DOS) using libGRX. The graphics are still lacking polish (there's no flood fill, for example) and the OS timer function has been ignored. This version is about twice as fast as the Borland one running in "fast" mode, but requires a 386 or better. This port is no longer actively maintained.
    My current port is of 0.62 to DJGPPv2 and GRX2, both of which are currently in beta. The version works correctly under (and requires) a DPMI host such as DQPMI, Windows 3.1, Windows/NT, or OS/2. It is very slightly faster than the DJGPPv1 port but is currently rather unstable under some DPMI hosts (most notable Windows/NT).

    Macintosh: (Hoon Im):
    I've ported RARS to the Mac; it is not complete yet but it runs. I used the 0.5 version of RARS. When I get it to draw all the strings cleanly, then I'll post the sources.
    The Mac version does not accept command line options but allows the user to enter simulation options through a dialog before the simulation is run. Otherwise, everything else is the same (for now).

    OS/2: tkn@VNET.IBM.COM (Thomas K. Netousek):
    There's an OS/2 port in the works from one of my colleagues... (John Howard):
    I am working on a port to GNAT Ada 95 for OS/2. It is not really RARS anymore since I am implementing for 3d vectors and experimenting. The goal is a simulator using DLL-based inputs, and an OpenGL viewer for the animation frames that are output from the
    simulator. (Nicole Greiber):
    [...] there is yet another OS/2 port in progress. It is done by (Ralph Scott) and by me. The port is still in beta, but already (partly) functioning. [...]

    Ralph Scott is actively maintaining last OS/2 version. This is at least version 0.67?

    Symantec C++: (GPFault):
    I am currently restarting my Symantec C++ port. There's nothing to it except for the changes to the graphics (to 320x240x256 I expect, due to the availability of XLIB).

    Watcom: (Mike Inman):
    There is a Watcom C++ port on the FTP site.
    For version 0.45?

    Windows/NT: (Muir Lee Harding):
    There is an alpha Windows port (using win32s or Windows/NT or '95) which uses DLL drivers on the FTP site. Version 0.64?

    Where could i find the latest versions of Rars ?
    Latest for DOS and UNIX systems is version 0.72b.
    For VC++ on Windows 0.71c.
    For OS2 - 0.67,
    For Amiga 0.65

    I've downloaded Java version of rars. It looks great, but:  how can I make my own robot in Java?
    Is source code of Java version available?
    Ralph can probably answer that question.

    Understanding RARS code

    5. Track formats and track design
    I'm in the process of setting up new track, but the start  and finish of the track don't line up (by just a little). How should I make the ends meet?
    I don't think it is possible to adjust just the last curve.
    The way I do it is to have my car driver code print out the
    start and end coordinates in its setup routine.
    I then use that to determine how far I'm off, find a straight
    that is approximately equal to the direction I want and just
    adjust that straight. ---ralph
    I use a program called TRACKEDT (I think!). It tells me by
    how much the ends don't meet, both in angle and x,y. I then
    fiddle with the track to make the ends meet.
    Usually I'm content with a deviation of <1 ft in either direction :-)
    I'm quite sure that I found this utility on the RARS ftp site
    somewhere (sorry, that's as precise as I can remember it).
    A GUI-ish tool to assist in track-building would be much
    appreciated - anyone? ;-) - Torben
    6. Car flashing
    If my car is flashing red does it mean collision or damage?
    Sometimes you might be colliding, but not taking damage.
    The cars only flash when they take damage.

    7. Alpha, vc and distances.
    What is alpha? How is it measured?
    Alpha is angle between your car's velocity vector ( which direction your car is moving) and your cars pointing vector (which direction your car's body is). It is measured in radians. It is positive when your cars body is pointing left from direction of motion and negative if your car's body is rigth from your direction of motion. Alpha is set according to your request (result.alpha) and it can be thougth as how much you turn your wheels (or steering wheel).  It is not entirely realistic, because real car does not turn its body instantly when you turn the steering wheel. That is the reason why alpha is sometimes compared to "angle of attack" from aeronautics.

    What is vc? How is it measured?
    Vc is "velocity commanded" - the desired speed you set with your robot code. It is measured in feets per second as most other speed variables inside the RARS code. You can set vc higher than your s.v (current speed), which gives you acceleration or lower than s.v, which gives you braking.
    Of course you do not get your desired vc instantly - real situation will depend on traction and physics. For example if you ask for vc=0, you just block your wheels and slide straigth until you stop. If you set vc=1000, then RARS program calculates how much power you have and sets the real vc according to it (you get maximum acceleration that your car's engine can afford).
    IMHO, vc should be replaced by other variable (between 0 and 1)- that measures fraction of throttle opening. But this needs major rewrite of rars physical engine (fraction of  throttle_opening is not equal to fraction of requested power). 

    Also, I was wondering, if I am 15 feet from the left wall, does this mean my left edge is 15 feet away from the wall, or is my center 15 feet from the wall?
    s.to_lft and s.to_rgt (as most other distances) is measured from the center of the car.
    Remember,  s.to_lft + s.to_rgt = track.width.

    8. Understanding s.nearby
    How to use variables in s.nearby for passing function?

    >I've been working on my passing algorithm and have just realized I'm
    >not sure about the data I get from s.nearby[x].rel_x. For
    >s.nearby[x].rel_y, I believe the data given is the distance to the
    >front of the car. But for
    >s.nearby[x].rel_x is the data given back the distance to the left edge
    >of the car, the right edge of the car, or the center of the car? Does
    >it matter if the car is to you right or left (e.g. s.nearby[x].rel_x
    >is the distance to the nearer edge of the car)? I'm going to
    >investigate this when I have some free time but if anyone already
    >knows the answer I would appreciate it.
    Both rel_x and rel_y are related to the center points of the cars.
    So the check (s.nearby[x].rel_y - CARLEN < 0) works (somewhat,
    depending on rotation of cars wrt velocity) for checking
    if you're bumping into a car from behind.
    What coordinate system are rel_x and rel_y based on?
    They're not based on y pointing down the track and x to the side walls.
    They're not based on my car's pointing vector (aligned with alpha).
    They're based on the velocity of my car.
    So it's not easy to determine an absolute position of the other car.
    Since version 0.71, you can use car's public variables for finding  their absolute X and Y position.

    For that use:
    extern Car* pcar[];
    othercar_x = pcar[other_car_ID]->X;
    othercar_y = pcar[other_car_ID]->Y;

    Car ID is based on drivers[] array (pcar[0] is first car from drivers array and so on, see drivers.cpp). It is also the same order that cars appear on scoreboard.
    See car.h: Class Car{public:} for other available variables.

    >Also in this same area, is s.to_lft the distance from the center of my
    >car to the left edge of the track or from the left edge of my car to
    >the left edge of the track?
    Center of car to edge of track. (s.to_lft + s.to_rgt = width)
    However offroad (and damage) isn't set until your center goes off the track.
    (s.to_lft < 0.0)
    >Let's say the car width is 9 feet and s.nearby[x].rel_x gives back a
    >value of 3.13. Am I in contact with that car? or is there 3.13 feet
    >bewteen the right edge of my car and the left edge of his car?
    Yes you are in contact, but the bumping algorithm isn't invoked if you're
    moving apart.
    >If I'm trying to go between two cars do I need there to be 9 feet
    >between there respective s.nearby[x].rel_x or 18 feet (4.5 feet for
    >each car - if s.nearby[x].rel_x is the center of the car and another 9
    >feet for my car - a tight fit :))
    You need 18 feet. You can get this once you know that rel_x, rel_y are
    based on the centers of the cars. Currently, for my driver, I'm trying to
    calculate the other car's to_rgt and to_lft to see if I should go inside or
    outside of that car. With the rel_x and rel_y being based on velocity vector,
    rather than a more fixed coordinate system, I'm having difficulties.
    >Just to clarify, does this mean that s.nearby will hold ALL cars in front
    >of my car, not just the nearest 5?
    Currently just 5. This is defined by  constant NEARBY_CARS in car.h
    >And is there a maximum distance for the cars in front before they're dropped
    >off the s.nearby list?
    No, it works just like the previous version, the closest
    three cars on your current segment or the next segment. "Closeness" is now
    measured by position on the track, not distance from your car.
    >> s.nearby[].alpha - The other car's current alpha.
    > >Relative to what?
    This is the same alpha as your car, the angle between the path of motion
    and the angle of the body. You need it to determine the angle
    of the other car's body wrt your car. This is related to the problem you were
    relating about not knowing how much of the track the other car was blocking.
    >> and what is "int braking "?
    Set to 1 if the car.s is reducing its speed by more than 5 fps, 0 otherwise.
    For other available variables for nearby cars see car.h: struct rel_state

    Writing your own robot

    Is it allowed to block other cars or ram them?
    Generally not, but it depends on race director. In earlier BORS series I allowed any team tactics. It was possible to drive  side by side, so that no other car can pass, have kamikaze robots that search for certain competitors or have bodyguards that avoid ramming from behind.  Currently any such "possibly dangerous" tactics is not allowed, because  robots are not yet strong enough to avoid such situations.

    9. Debugging your robot code:

    > I am getting following error message. What's up?
    > asin: Domain Error
    > floating point error: Domain
    > Abnormal program termination ...
    > Our robot does use asin.
    A domain error indicated that the input to the function is invalid.
    For asin(x), x must satisfy -1 <= x <= 1. Either your car or
    someone elses car is calling asin() with a bad value - possibly
    a result of a not checking boundary conditions (i.e. what
    happens at the extremes).
    My robot has a bug that reboots the system. I cannot find where is it.
    >:)The rebooting problem is a killer. My usual method for this sort of
    >:)thing is to narrow it down via printf's. If they go by too quick,
    >:)then I try sounds. Different pitches at different points in the
    >:)program. This should at least get you to within 5-10 lines of the
    >:)problem spot, if you can't take it from there, try getting back to
    >:)the list with your new found knowledge.
    What I HAVE done, in the past, is to use fprintf's to dump my
    debug data to a file and look at the result after the carnage flies.
    The only thing you have to do for this method is:
    setvbuf (file, 0, _IONBF, 0); This makes the stream unbuffered so
    that each fprintf goes right to the file. Otherwise, the line you
    are looking for may be in the buffer at crash time and you wouldn't
    know it. To the best of my knowledge (and Borland's help file),
    the line I gave is standard from stdio.h. - Christopher Lund
    Torber writes:
    I've had very good experiences with this little function
    (transcribed from memory, actual syntax may vary):
    void DEBUG(char *dumpstr) {
     FILE *fp = fopen("dumpfile","ra");
     if (fp) {
     fprintf(fp, "%s\n", dumpstr);
    // Use:  DEBUG("Here we are"); A couple of points here...
    It writes all your debugging messages into a file (here: "dumpfile"),
    closing it after every write so that no messages are lost.
    That's the thing I like about the setvbuf method I wrote of yesterday,
    namely, you don't have to open and close the file for each write.
    In any event, if you setup your function as: void DEBUG
    (char *dumpstr,...) and then use vprintf in <stdarg.h>
    (again, a function I believe is standard across platforms),
    your debug function can be used to dump variables to the file as well,
    easing the debugging further. - Christopher Lund
    10. Optimization of the car
    Which machine-learning methods have been used for finding optimal line around the track? 
    Just a very simple and slow trial and error procedure:
    A given racing line is modified slightly, then some laps are run.
    If the car becomes faster, the modification is accepted.
    This is repeated thousands of times until the car is very fast.
    The racing lines are stored in strings that look like gene sequences,
    but the algorithm is no GA according to the definition of GA that
    someone posted to the list last year. Henning
    Other people have used Neural Networks, Genetic algorithms,
    trial and error learning.

    What is the simplest working robot?
    In 1996 we had "midget" competition for cars less than 450 characters. Some of the midget competitors are still available. One example of short and efficient robot is here:

    #include "car.h" con_vec hk(situation&s)
    { my_name_is("hk");
    con_vec r={(, s.dead_ahead?.9*}; stuck(s.backward,s.v,,s.to_lft,s.to_rgt,&r.alpha,&;
    return r;
    Actually I used 3 copies of the same robot on that race day.
    The only differences were the names and some whitespace
    added so they could run in all three size classes. You can
    use it as a demo robot. It demonstrates how easy it is to win
    a race with a very simple robot. On adelaid2 the triple won
    place 1, 2 and 3 and on brazil place 2, 3 and 4! ;-) Henning

    1. Zoom or more realistic 3D view from inside the car.
    2. 4-wheel driving car, with suspension and gearbox.
    2. Timelimit for each robot to make it's calculations.
    3. Various improvements in reality of track and car setup.
    4. Realistic weather conditions