Question about Flight Models?

[28sweep]

What does it mean when sombody says that a Flight Model is based on "Physics?"  I have heard people on these boards say that AH does not have a Flight Model based on "Physics" but rather has a model based on "Tabels."  What does this mean?  What about the full-motion simulators the military uses?  I would think they would be very very accurate...how do those flight models work?

[ccvi]

Instead of doing calculations in real time, data is looked up in tables. The tables themself don't really make a difference to real time calculations.

It all depends on the quality of data the tables are filled with.

[hitech]

It's a meaningless statment. Physics and tables are not exclusive items. Milatary sims tend to be much more table based then games do. The choosing between an equation or look up table when coding has more to do with performance than it does with the outcome of the model. Either way you get the same numbers out. For instance you can run equations one time to populate a table. Then during the real time run you just use the table for speed of lookup.

[Booky]

When I say something like this I am usually refering to the fact that they are made by data. This will never be the same as RL, thats why it is called a simulator  Because of this, sometimes planes can do things in a simulator that they could never do in RL because of physics. Yet other times in a sim the very same plane may be limited by the data.

It is not any programers fault unless they enter bad data. It is just the fact that the way models are designed, by military and civilian, they will never be 100% accurate.

IMHO AH has some of the best that I have seen. Ofcourse I haven't flown military WWII wims, just military modern jets, so my view could be way off as well:rolleyes:

I guess untill someone who flew the aircraft gets in here and says, its right on, or it is way off, then I think I will take HT models as the way it was. All except that damn la7 anyway

Booky

[AtmkRstr]

I think I know what they mean.

A FM based on tables uses known information from historic data in order to make the planes fly like they should.

In a Physics based FM, the planes are modeled using accurate blueprints modeling every shape of the wing and fuselage, and then the FM interprets what the plane is supposed to fly like.

When WWIIOL was made (Physics based), they modeled planes and put them in the Physics Engine and they had a tendancy to fly like they're supposed to.  In a physics model, some tweeking based on tables would probably be necesary.

A Physics model has numerous advantages because where data is not available, the FM can extrapolate/interpolate where as a table based FM depends entirely on data (or speculated data).  A Physics FM would be much harder to code IMO.

[Turbot]

Originally posted by AtmkRstr
When WWIIOL was made (Physics based), they modeled planes and put them in the Physics Engine and they had a tendancy to fly like they're supposed to.  

Hehe I still get a chuckle out of the 110 helicopter

[LoneStarBuckeye]

An accurate flight model comprises a set of linked, non-linear differential equations.  Such equations can be solved by a variety of numerical means (e.g., Runga-Kutta method) on a computer, but the computations can be very time intensive.  One can make it easier for the computer by using simpler models (i.e., less complex with linearized and easier-to-solve equations), but the fidelity of the simulation may suffer.  I assume that so-called "physics" models use a differential equation model and solve the equations in real-time to continuously compute the state variables incorporated in the model (e.g., linear and angular position, velocity, and acceleration in three dimensions).  

I assume that what makes a "table-based" model fundamentally different from a "physics" model is that the table-based model does not involve solving differential equations in real time (i.e., as the simulation is running).  I can imagine that a table-based model might include entries that would specify that, for example, at a particular position (e.g., altitude and orientation), velocity (3d), and acceleration (3d), a set of inputs (e.g., thrust, control surface deflections) results in a pre-defined change in position (3d), velocity (3d), and accerlation (3d).  Obviously, the more entries the table has, the more accurately it can simulate a "physics" model.

I can see a couple of advantages to a table-based model.  First, and perhaps most importantly, the CPU overhead needed to implement it is far below that necessary to chug through a complex physics model.  Second, it may be difficult to construct accurate differential state equations to model a particular plane--these are generally complex and, as stated in an earlier post, depend critically on all of the plane's physical characteristics.  A table-based model allows the designer to more readily incorporate heuristic information (e.g., climb rate and speed charts) about a plane's performance into its flight model.  

In the end, I don't think that one approach is "better" than the other in every situation.  If you have "perfect" equations for your physics model and ample CPU overhead with which to implement it, then that is probably the way to go.  My guess, however, is that HTC had neither when designing AH.  

- JNOV

[funkedup]

What HT said.  Usage of physics and LUT's (look up tables) are not mutually exclusive.  LUT's are quite common in engineering.

[funkedup]

Also, despite what some have said here, you can most definitely solve differential equations in real time using LUT's.  The state variable derivatives are equal to nonlinear functions of the state variables and control inputs.  You can use LUT's in the calculation of the derivative functions and then integrate normally.  

And don't forget that many of functions are curve fits of discrete data points.  In that case there may not be a significant difference in accuracy between using the curve fit equation (which is just an approximation of the actual data points) or using a table based on that curve fit.

I'd wager that for a program like AH, the error introduced by the neccesity for numerical integration exceeds any error introduced by using an LUT instead of a calculation.  In fact, using LUT's can make each integration cycle process faster, allowing a smaller time step size, and therefore increasing accuracy AND graphics performance.

[GRUNHERZ]

Lets not forget blade element theory....

[LoneStarBuckeye]

Funked:

I didn't mean to imply that using a table would eliminate the need to "integrate."   Ineed, I was referring to what I think you describe when I mentioned changing the state variables by a pre-defined amount.  (I wasn't precise and was trying to leave out the math.)

I didn't read HT's post carefully before posting (that's a cardinal sin -- sorry).  He mentions "running the equations to populate the tables."  I wonder if that's really what they did?  If so, perhaps they do have the equations and it's really all about performance.  Maybe WWII era planes are sufficiently similar that a set of parameterized equations is sufficient to model the lot.

The only sort of table-based simulation I've ever created invovled linearizing a very non-linear system about a number of operating points in the state space, effectively dividing the state space into a set of linear systems.  I didn't think about it at the time, but I suppose that could be considered a table-based simulation.  (We didn't linearize because of performance concerns; we were concerned with creating control systems and, in that context, linear systems are much easier to deal with than non-linear systems.)  In any case, I doubt that this is the sort of approximation that AH uses.

For a general time-invariant, non-linear system of the form

dx/dt = f(x,u)
y = g(x,u)

what does "populating the tables" involve?  I would guess that it involves defining f and g at a sufficient number of discrete points-- basically discretizing the functions so that they are indexable at discrete points in the state and input spaces.  With such tables, numerically integrating would be reduced to look-ups and simple arithmetic.  It would be quite fast, I imagine.  Do you know if that's how AH does it?  If it is, it would be interesting to know how many "index points" and "dimensions" the tables have.

- JNOV

[Ossie]

*reads thread*




*head explodes*

[jbroey3]

X-plane  

If you want the most "accurate" flight model for your computer at home. Get it.

http://www.x-plane.com

[streakeagle]

Unfortunately, no calculated physics model that can run on a present day pc will ever work very accurately. Even NASA's computers fail to model compressibility and transonic flight regime accurately enough to eliminate the need for wind tunnel and full scale flight testing.

X-Plane seems to work ok for Cessnas and Pipers (I say that having flown both in real life). But try taking up the F-4 Phantom and telling me that X-Plane is the most realistic sim. Jane's Fighters Anthology was based on tables that were based on the standard height-Mach energy-maneuverability graphs. Some of the player made libraries had very accurate tables that produced results far superior to X-Plane.

I like some aspects of X-Plane, but by no means would I consider it the most realistc flight sim.

[jbroey3]

X-plane can model supersonic flight, as well.

Current version is 6.40 take a look.