Mathias Fr�hlich wrote:

> Sent: 20 October 2004 07:41
> To: FlightGear developers discussions
> Subject: Re: [Flightgear-devel] status of aircraft carrier
> 
> On Dienstag 19 Oktober 2004 21:23, Vivian Meazza wrote:
> > > Hmm,
> > > I am not satisfied with anything which is only working on a per frame
> > > basis.
> > > Just because, if so, we will have different bevour of our physical
> models
> > > dependent of the frammerate.
> >
> > I think I put this bit badly. The geodetic output would be dependent on
> the
> > speed of an iteration loop. If we are only modeling an arrester wire
> > _surface_ it should suffice. I can generate some proof-of-principle code
> > for this quite readily, I think. I'll see what I can do.
> Ok, let's start now :)
> I would like to have those positions of the arrester wires not in
> lat/lon/alt
> but rather than in earth centered coordinates (cartesian coordinates: x
> towards lat/lon=0, z towards northpole). Just because we already have all
> scenery values stored in this format. We have a scenery reference point
> and
> relative to that, we have rotated vertices.
> With those values we can cheap compute (double valued) positions of these
> vertices. And then transform them, just by translation and rotation, to
> some
> coordinate frame required for modelling the hook, gear ...

We calculated the output in geodetic terms (lat/long/alt) for submodels so
that they could be displayed, and it's no problem to output in x,y,z
aircraft terms. It didn't seem to be expensive computationally. Further,
lat/long/alt was the easiest to get at for the aircraft location. I think
that I'll need some help here with the necessary conversion factors. I'll
look into it further. 

> 
> If you use geodetic lat/lon, you need to transform the values to geodetic
> lat/lon/alt values which is expensive as such since this requires the use
> of
> an iteration method. And then transform those values back to that same
> flat
> coordinate space suitable for comuting ground reactions.
> For that backtransform, I can't think of anything not using those earth
> centered corrdinates.

Do we need ground reactions as an intrinsic part of this code? Although I
can see that it might represent an opportunity to improve the current
situation. 

> So all together: if lat/lon/alt is omitted in those computations, we can
> get
> the same effect with less computations.

Yes, agreed.
 
> What I have here in my private tree, is code to have a small, per FDM
> instance, cache of ground tiles around the aircraft. With this one it is
> easy
> to taxi on slopes. Also if such arrester wires are stored there, it would
> be
> easy to compute the interactions with them.
> Intersections with those few triangles can then be done often per
> timestep,
> since there are only few triangles in that cache. The cache itself is at
> the
> moment filled on demand, whenever the FDM cannot find a triangle below the
> contact point, the scenegraph is queried for a new triangle.
> That works for our scenery, but for that simple plane modelling the
> arrester
> wires, this does not work, since we find a triangle from the scenery, we
> do
> never query the triangles modelling the wires.
> 
> What I plan to do here, is to build up a cache of all triangles in the
> area of
> an aircraft. This is easy to do with the hierarchical boundingbox
> intersection routines from ssg*.
> Having that cache, then implement wires as geometry nodes in the scene
> graph,
> which will then show up in the cache if we are near them. Intersection
> with
> geometry nodes in that cache can be implemented using ssg routines.
> Coupling the results into a FDM is straight forward.

Good

> This approach with that cache would also allow to implement this stuff
> together with FDM's in child processes or even different machines, like
> Curt
> was talking about some time ago.
> 
> > My main conceptual difficulty is the compass alignment of the arrester
> wire
> > surface. If we test the hook position in geodetic terms, i.e. lat a <
> hook
> > pos < lat b etc. how do we account for the odd triangles at the corners?
> See above ...
> Use cartesian coordinates for that and then use ssg* for that.
> My head includes the solutions here :)
> Seriously, I have much of a concept here. So if we can work together, we
> might
> be faster :)

Good, let's go for it.

> > Good point. The probability of catching a wire given that the hook
> > intersects the wire surface is ... ? Some function of hook design,
> number
> > of wires, aircraft velocities, ship velocities ... some pure chance ...
> ?
> Yep, this kind of stuff. But not relevant at this stage of development...

Just teasing :-) - some arbitrary number related to number of wires in the
surface and aircraft type will do in due course: Buccaneers almost always
caught a wire, Gannets suffered from hook skip and often had to do a bolter.

> With that cache it would also be easy to model individual wires. If we
> have
> them deterministically modelled, multiply with some catch propability.
> 

That would be nice, but a surface would probably do to start with. The
probability of catching a wire is more to do with the aircraft than with the
wire if we model individual wires, if we use a surface then we would need to
add a probability related to the number of wires in the surface, but this
would be pretty arbitrary.  

I'll get on with seeing if I can provide hook position - we would be well
under way with that. I think this would be a worthwhile activity since we
seem to have most of the bits almost in place.

Regards,

Vivian





_______________________________________________
Flightgear-devel mailing list
[EMAIL PROTECTED]
http://mail.flightgear.org/mailman/listinfo/flightgear-devel
2f585eeea02e2c79d7b1d8c4963bae2d

Reply via email to