Wow, you guys are GOOD!
Sorry for my absence over the last week or two; I've been doing some
work with GM. Got to put a G-Cube in a prototype vehicle and drive
around some cones. Can't say much more about that or they'll dispatch
the black helicopters and sharpshooters...
John's explanation of Usage is right on. The friction "tube" is an
important part of all of this. This helps us to plot your car's forward
potential at various speeds. However, the question that Brian asked
about why we need to handle forward acceleration as a seperate case
bears a little more discussion. Here's the deal.
We want to allow you to score well when you are driving with your foot
to the floor, as long as you are in the right gear, etc. So what we are
trying to capture is when you have your foot on the floor in places
where you are primarily trying to accelerate. This is done by looking
at the angle of the vector from the origin. If the vector is with 45
degrees of straight up, then we assume you are trying to accelerate
primarily. So in those cases we give more credit for acceleration.
This also helps to eliminate some of the lumpiness in the acceleration
"memory map" caused by shifting.
Now, several of you have hit on what I had decided was a weak point in
the Usage system...transition rates. The version of the software that
you all have uses ONLY the change in the x values, that is, the lateral
transition rate. I think the first person to point out this problem to
me was Jeff Cashmore, who's HS Celica was nearly impossible to get high
Aggressiveness scores. (Aggressiveness is scored in part by high
transition rates.) Jeff noticed, as several of you have, that no matter
how hard he tried, a transition from gas to brake was not scored well on
Usage, and it also didn't do well on Aggressiveness.
The solution is one that I have in place now, but I'm not yet ready to
release it as I'm putting other new stuff in this version too. But
what's different is that in this new version we take transition rate in
all directions, rather than just lateral or just longitudinal. This is
done simply by plotting the distance between two successive points in a
friction circle diagram. The distance times the sample rate is the
transition rate per second. This now allows lateral, longitudinal and
even off-kilter transitions like going from left-brake to gas-right
quickly.
Also, we previously never had a way to eliminate peaks in the plotted
transition rate data. In lateral and longitudinal g's we filter out he
peaks by using sustained peaks rather than absolute peaks. This keeps
us from trying to compare our runs to a single point of 1.40g's when we
hit that bump in the sweeper, when our real sustainable lateral g's are
more like 1.10g. The new version does something similar for transition
peaks by using the sustain period to calculate unsustainable transition
rates. In short, this means that if you have a single incidence of a
transition rate of 5.0g/sec, but everywhere else you get only 3.0g/sec,
the 5.0g/sec peak won't hurt your score when you are using sustained
peaks.
This is all part of an entirely new rating system, which I hated to do
because we all get used to the previous ratings, but which I felt was
needed. The new ratings were done using about 30 test runs of drivers
of varying abilities, all recorded with G-Cubes rather than G-Analysts.
I'm happy to report that I'm pretty satisfied with the new ratings, but
I would like to gather a bit more data before I say I'm completely
satisfied. The new rating system seems to track extremely true with
fast times, and as an added bonus, I've observed no ill effects on
g.Analyst data.
Look for this new rating system to hit in the next couple of weeks,
together with a variety of new functions which I'll cover in more detail
later.
THANKS to all of you who answered these questions so accurately while I
was out! And I'll have the new ratings available for release very
shortly.
--Byron
Engstrom wrote:
> Well, I'm not Byron, and I don't claim to be able to dazzle, but I will try
> to explain how usage is computed based on my conversations with
> Byron and my reading of the help files.
>
> Before explaining usage, let me take a detour and explain how Geez
> works with regard to longitudinal acceleration. This will become relevant
> in a minute.
>
> Geez doesn't use a friction "circle" but instead uses a friction "tube".
> The friction tube is generated by taking friction circles and stacking
> them on top of each other. Geez has a friction circle for each speed
> quantum. The reason is a car could effect a much higher longitudinal
> acceleration vector at one speed (say 25 MPH when the engine is
> in the meat of its powerband) than at another speed (say 35 MPH,
> right after the shift to 2nd gear when the torque at the rear wheels
> is less than at 25 MPH because the gearing ratio change and the
> fact that the engine is at an RPM which puts it lower in the powerband).
> When a use the term "friction circle" below, I mean the friction circle
> that corresponds to the speed you are currently traveling.
>
> Now, usage at any point in the run is calculated as the highest of three
> numbers.
> 1) Total G percentage - this number is the distance from the
> origin (or center of the friction circle) of the total Gs that
> the car is currently pulling divided by the distance from
> the origin of the edge of the friction circle that touches
> a line that runs from the origin through the total G point
> that the car is currently pulling.
>
> 2) Maximum acceleration G percentage - this number is the
> longitudinal acceleration that the car is currently pulling
> divided by the maximum longitudinal acceleration that the
> car can pull. If the friction circle for a given speed was
> always a circle, then this percentage would always be less
> than or equal to #1 above. However, the friction "circle" for
> any given speed is actually a friction ellipse. This means
> that although you might only be pulling 90% of total Gs,
> you are accelerating at 95% of the maximum longitudinal
> acceleration. Since Geez doesn't know whether you want
> to pull max longitudinal Gs or max total Gs at this point on
> the course, it gives you the benefit of the doubt.
>
> 3) Transition rate percentage - this number is calculated as
> the current transition rate (Gs per second) divided by the
> highest transition rate that you've pulled in your car.
>Basically,
> Geez doesn't penalize you for moving through the middle of
> the friction circle quickly.
>
> I'm still a little unclear if there is a max longitudinal transition rate and
> a max lateral transition rate that are used. Maybe Byron can enlighten
> us about this.
>
> John Engstrom
|