1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.

Project CARS Force Feedback Guide

Discussion in 'Project CARS' started by jimortality, Jun 5, 2015.

  1. jimortality

    Premium Member

    FFB Guide

    Input Signals

    The four front tire input signals are the component parts of the whole tire induced torque coming
    through the rack. So if these are all scaled to the same thing (by convention 1.0), this is the
    same as straight rack torque.
    The two rear tire signals are to enable the Seats of Pants concept. Neither of these go through
    rack geometry though, as there is no rear rack and steering wheel. These just go straight to the
    Finally, the G force signal is to enable the Gut physical simulation concept.

    Tire Force

    This is simply an overall multiplier on all of the input tire forces. Note that G forces, the input to Gut, are not
    scaled with this parameter. Note that the other FFB parameter in the Controller section is the
    same as the scaling in the driver. Reducing that does not help saturation, it simply reduces



    This is a multiplier on all of the front tire forces. This was added to allow the following four scale
    to default to 1, and be more intuitively like “weights”.


    Individual scales on the components going through the spindle/rack. To get pure rack forces,
    leave these all at the same value. 1.0 is a convenient value for this, and use SpindleMasterScale
    to dial overall spindle force.


    Individual smoothing on the components going through the spindle/rack. Typically Fx requires
    more smoothing than the others. 0.0 is no smoothing. 1.0 is normalized to “really smooth but
    still some useful signal”. Values above 1.0 are valid.


    SpindleArm is the angle, in degrees, of the attachment of the tie rod to the spindle. Zero degrees
    means the tie rod is attached directly aft of the axis. That particular distance, how far aft, is not
    critical, because that just amounts to scale, which we adjust based on squeezing into the device
    range anyway. The angle though matters a lot in how the forces feel when the steering wheel is
    not straight.
    90 degrees is then with the tie rod directly inboard of the axis, which physically would result in the
    inability to steer. Realistic values I'd guess are between 0 and 45..

    Seat of Pants

    The basic idea of “Seat of Pants” is to present information from what is happening at the rear of
    the car through force feedback. There are two physical forces that are used. The rear side
    loads and the rear vertical loads.


    Overall scaling of Seat of Pants


    Scaling of the rear side load effect.


    Scaling of the rear vertical load effect, which is actually the difference between right and left
    vertical loads.


    Smoothing of the Seat of Pants signal. 0.0 is no smoothing. 1.0 is normalized to “really smooth
    but still some useful signal”. Values above 1.0 are valid.

    Relative Torque Adjust

    The idea here is to present torque to the wheel based
    on the change in torque through time instead of as absolute torque. This means that with
    reasonable parameters, the wheel will never fully saturate. But unlike soft clipping (which can
    also prevent saturation), the high end torques do not get as heavily squeezed.
    There is one side effect to tune out though, and that is the wheel losing center over time. If all
    torque was completely via “Relative Torque Adjust”, centered torque would move around as the
    wheel goes through previously saturating torques. To prevent this, we use the bleed value to
    “bleed” absolute torque back into the mix.


    This is the scaling on the amount of calculated torque change that is applied. 1.0 is the intuitively
    correct value. 0.0 turns this component off.


    This is a time value for bleeding absolute torque back in. 1.0s is a good starting point.


    This is the force to wheel value (so in the 0.0 to 1.0 range) where the non absolute running
    magnitude is clamped. This does not clamp the overall value, and torques can still go above
    this, but it does exert a strong clamping effect. 1.0 is a good starting point for this. Values
    greater than 1.0 can make sense if soft clipping is also used. Values less than 1.0 makes sense
    to give some headroom for spikes to be a little more symmetrical around the clamp.
    Note that with this component on, and with clamp at 1.0 or less, and not too much bleed, there is
    no full saturation. What this means is that what was too much force before now becomes more
    force effects felt near full force. But this too can become too much, as that can start to
    overpower the more subtle unsaturated force range. So you still need to dial overall force (via
    Tire Force and the scales), but that scaling can become an interesting control, not just
    something to avoid saturation with.

    Gut Simulation

    This is a simulation of the G forces on the body of the driver. Basically, G forces move the body
    around via a physical simulation, and the result of that simulation is translated to force feedback.


    Magnitude of the gut simulation in FFB. 1.0 is normalized to “significant but not overpowering”.


    Magnitude of longitudinal effect applied. This is a scaling of the baseline lateral effect. At 0.0, the
    gut effect will be all based on lateral G’s. With non zero GutLongScale, under braking G’s, the
    lateral effect will increase, and under acceleration G’s the lateral effect will decrease.


    This is the mass of the simulated “gut”, which should not be the whole human body. It should be
    some lesser portion, roughly being the effective amount of mass not “locked down” rigid by the
    seat and seatbelts. This is a very fuzzy concept, so the number is really just a very rough
    ballpark number. This is fine, because the simulation is not overly sensitive to this number. It
    matters, but it is not extremely critical.
    The default is 50 kg.

    Arm Simulation

    The arm simulation simulates that the wheel is driven by a non rigid linkage, namely the driver’s
    arms, as well as play and mass in the linkages themselves.. However, this is done purely with
    force feedback. The position of the the controller still directly dictates the location of the
    simulation wheel.
    This simulation also serves as the main global smoothing stage.


    Ratio of incoming signal to pass through the arm simulation. 0.0 if off. 1.0 is application of all
    incoming signal.


    Mass of “arms”, with respect to simulation. This does not necessarily mean the average mass
    of two human arms. This is the effective mass with respect to the degree of freedom that is the


    Spring*like stiffness of the “arms”. Stiffer settings will pass through higher frequency
    information. Softer settings will smooth more.


    This is a multiplier on critical damping of whatever mass and stiffness is set. Therefore, 1.0
    means exactly critically damped.

    Soft Clipping

    This compresses all force within range of the wheel, although the stronger the force, the more it
    is squeezed into the higher force range. In some ways this is like Log Scaling, but Soft Clipping
    guarantees all signal will squeeze into the range, however compressed. On the other hand,
    approaching linear behavior is not implicit with soft clipping, as it can be with log scaling.


    The “half signal” for setting the soft clipper. The value set here is the input signal that will
    become 0.5 as an output signal. Setting this to 0.0 turns the soft clipping off. Setting this to 0.5
    is maybe the closest approximation to linear while on, but is not linear. Setting this to 1.0 will
    match the derivative/slope of the output at zero input (so if you want the lowest forces to feel
    similar, and compress everything else). Therefore, less than 1.0 will amplify some lower force,
    and reduce larger forces. Greater than 1.0 will reduce all forces.


    Straight soft clipping will never reach full 1.0 magnitude, which means for lots of soft clipping
    scenarios, the full force of the wheel is never quite used, possibly to a noticeable level.
    SoftClipUnity sets the expected maximum force that will hit the soft clipper, and rescales such
    that that force outputs at 1.0 (full force of wheel). This means saturation may be reintroduced if
    this is set too low, but it is useful to fine tune output, especially when the soft clipper is used
    more for non*linear response than for anti*saturation. Setting this to 0.0 turns the unity re*scaling


    This is a new component, and is directly in response to some devices going flat in
    response at higher force levels. This is somewhat the opposite non*linear tool as the soft
    clipper, but is shaped differently, to better fit the nature of devices (and be easier to control).
    So what scoop does is reduce lower forces more and high forces less, thereby increasing the
    slope of force where some devices reduce the slope of force. Since devices seem to do this in
    two more or less linear regimes, with a knee in between, this is how this component works (in
    the opposite direction).


    The input force level where the knee is at. If this is 0.0, this component is turned off.
    The input force reduction below the knee. Above the knee, the force slope is increased such that
    at 1.0 input force, the output force is 1.0.

    Tighten Center

    Note that the name of this can be confusing. This has nothing to do with tightening the wheel
    about geometric top center. The “center” for this component means “zero force”, and has
    nothing to do with wheel position.
    The primary purpose of this is to remove wheel deadzones, but it can also be a shaping tool.


    This is the input force below which the output force is increased to remove a deadzone. Put
    more simply, this is the size of the deadzone you are trying to remove.


    This controls how sharply the output force approaches zero force as the input force goes below


    One use of damping can be to counter inherent drag in a device by using negative BaseDrag.
    However, often devices do not have linear inherent drag, so setting BaseDrag such that there is
    little to no device resistance at slow wheel speed will result in accelerating forces at higher wheel
    speeds. This can be fixed by also having some positive BaseDragSqr.
    A technique to set damping to cancel most device drag is to turn off ALL forces, Slow
    Speed Force, and TireForce) and adjust BaseDrag and BaseDragSqr such that the wheel stays
    the same speed or slows down ever so slightly (until it hits a stop) when you give it a good push
    at different rates. It seems better to have a tiny bit of drag left than to have the wheel accelerate
    on its own at any speed.


    This is resistance on the wheel as a function of wheel angular velocity.


    This is resistance on the wheel as a function of wheel angular velocity squared.


    This is smoothing of the angular velocity for drag calculations. Raw position data on some
    devices can be noisy. Note that increasing smoothing can have a secondary apparent effect of
    increasing the effect of drag.
    @Lazarou I expect you to get this perfect now lololololol
    • Haha Haha x 2
    • Beer Beer x 1
  2. I need to drop a few LSD before I can compute that.
    • Haha Haha x 7
    • Agree Agree x 3
    • Winner Winner x 1
  3. jimortality

    Premium Member

    • Like Like x 1
  4. Jake Fangio

    Jake Fangio
    Please don't rain pleeaassee don't rain

    I love it!!:roflmao:
  5. Eric Bergeron

    Eric Bergeron
    Premium Member

    wow that's exactly what I wanted to adjust my driving (DD) but unfortunately there are too many adjustments, I will spent one month to find the right settings.
    thanks :thumbsup:
  6. jimortality

    Premium Member

    Yeah, good luck with that, I think a lot of people have given up, me included lol
  7. Brandon Wright

    Brandon Wright
    I'm just here for the snacks

    I've settled. I use bManic's global settings and then only adjust master spindle and arm angle on each car. It's probably not the best possible feeling, but it's good enough and let's me spend more time on track and less in the menus.
    • Agree Agree x 4
  8. jimortality

    Premium Member

    I've given it a month to be fair so deleted it re-installed and going to start again with Bmanic's global knock down the first 2 to 40 or 50 and leave everything else as standard.
  9. Brandon Wright

    Brandon Wright
    I'm just here for the snacks

    I'd wait until after the update comes out, they are tinkering with the FFB settings a bit so anything we do now may be down the drain after they make changes.
  10. jimortality

    Premium Member

    I think it's the main ffb in controls they are messing with, not the edit car set up.
  11. Brandon Wright

    Brandon Wright
    I'm just here for the snacks

    But those settings could have a knock-on effect to the per car settings. Not saying they will for sure, but I've seen it many times where people spend time getting things dialed in and then an update comes and renders it all useless, so I've put FFB tinkering on hold until the update just to be safe.
    • Agree Agree x 1

    Founder of Soggy Bottom Racers Club

  13. Eric Bergeron

    Eric Bergeron
    Premium Member

    • Like Like x 1
  14. Computer says "You need to upgrade brain to be compatible with this new software" :mad:
    • Haha Haha x 1
  15. Brandon Wright

    Brandon Wright
    I'm just here for the snacks

    I tried that, but unfortunately they said they were waiting on an update that would make the software compatible with my brain............:cautious:
  16. Very rare for a software manufacturer to downgrade a game :laugh:
    • Haha Haha x 1
  17. Brandon Wright

    Brandon Wright
    I'm just here for the snacks

    Apparently you've never played GT6. :O_o: :roflmao:

    Founder of Soggy Bottom Racers Club

    yup they are pretty good on track :)