Re: [Flightgear-devel] autopilot vs joystick

[leee <le...@sp...>]

On Friday 06 Nov 2009, Curtis Olson wrote:
> On Fri, Nov 6, 2009 at 9:32 AM, James Turner wrote:
> > On 6 Nov 2009, at 14:47, dave perry wrote:
> > > It seems to me it should be possible in controls.nas to
> > > choose between the two values for aileron and two values for
> > > elevator based on an agreed upon autopilot properties.  To do
> > > this, one needs to be able to access the joystick axis value.
> >
> > This is one of the reasons I had to make the autobrakes be C++
> > code, and make some other unpleasant changes - we're missing a
> > layer of abstraction between 'input system output properties'
> > and 'FDM input properties'. I found there's many joystick
> > config files that expect full ability to mess with the
> > /controls properties - which seems reasonable.
> >
> > What's not reasonable is that the FDMs read the same properties
> > directly, which makes life hard for auto[pilot|brake|etc]
> > systems. For YASim, it's bearable, since the 'input' control
> > properties are defined in the aircraft XML files (so I redefine
> > the YAsim brake inputs to be /
> > autopilot/autobrake/left-brake-output instead of
> > /controls/left-brake, for example), but for JSBSim I needed to
> > change the glue C++ code to stop it reading FGControls
> > directly, if the autobrakes are active. (Yuck!)
> >
> > I'm not sure what the solution is - I think it's to create a
> > tree of 'fdm-input' properties analogous to /controls, which is
> > not touched by any joystick scripts / external inputs / etc,
> > and which by default is simply a copy of the /controls values,
> > but which the autopilot (and similar systems) can manipulate.
> >
> > That's exactly what my C++ autobrake currently does, btw, and
> > this was necessary to be able to correctly implement the
> > interaction between autobrake input and pilot over-ride inputs;
> > namely that any pilot control input automatically disarms the
> > autobrake system.
>
> Hi James,
>
> I think your analysis is pretty much dead on.  We are missing a
> layer of abstraction.  Ideally, the raw control inputs would be
> fed into a separate branch of the property tree, and then the
> actual control surface positions would be sent to the FDM.  This
> would allow us to create fly by wire systems that translate
> inputs to outputs in arbitrary ways.  For simple aircraft with
> direct mechanical controls (or for aircraft where all the flight
> controls and control surface modeling is handled at the FDM
> layer), we could have a simple FBW module that just does a 1-to-1
> copy of the values.
>
> In the case of JSBSim aircraft, perhaps there's a way to
> implement autobrakes entirely within JSBsim and work around the
> problem in this particular case?
>
> Regards,
>
> Curt.

As I've mentioned, we can already do this using autopilot filters.  

The only real problem that I hit was that YASim absolutely requires 
the /controls/flight/elevator-trim input axis to be bound to the 
<hstab> Flap0 YASim control element, otherwise the solver will fail 
because it assumes no elevator control.  I think this was probably 
just a dumbness check, to ensure that some elevator control was 
possible, but in any case, it's possible to work around it.

I've attached a diagram (exported in PDF format) that I made to help 
me keep track of an FBW type autopilot config that I was working 
on, to give an idea of how you can route JS inputs using filters.

LeeE