[x3d-public] LineSensor

Andreas Plesch andreasplesch at gmail.com
Wed Mar 27 20:07:18 PDT 2019 

In preparation for a well presented spec. comment, I developed a
minimal but discerning example scene illustrating the problematic on
edge view for 1D PlaneSensors:

https://gist.github.com/andreasplesch/33016fd0918178f5dfcf4632d477f4e4#file-planesensor1donedge-x3d

For loading straight into browsers:

https://gist.githubusercontent.com/andreasplesch/33016fd0918178f5dfcf4632d477f4e4/raw/7672b81e34ff30855eaad53519e0266d25de876b/PlaneSensor1DOnEdge.x3d

Here in action with x3dom:

https://bl.ocks.org/andreasplesch/33016fd0918178f5dfcf4632d477f4e4

Dragging the red cone in strictly conforming X3D browsers should be
ill defined in the on edge views. Testing shows that in these browsers
the cone jumps around or does not move at all as the spec. currently
requires:
- BS contact
- Instant
- x3dom
- freeWrl
In contrast, other browser ignore the spec. in favour of providing the
expected dragging behaviour even in on edge views:
- Octaga
- x-ite
- view3dscene
- H3Dviewer
These browser must have internally a special line sensor mode but it
is unclear what this mode then does. For example, is
trackPoint_changed returned and where should it be since there is no
intersection with the tracking plane ?

If think I could see that x_ite picks the point on the line closest to
the pointer in screen space and unprojects that point then to local
sensor coordinates.

How does view3dscene behave ? How does it calculate translation and
track point ?

The diverging  browser behaviour reinforces the clear need for spec.
refinement addressing this situation.

Here is my github project page where I put together the spec. comment:

https://github.com/andreasplesch/x3dom/projects/5

With regards to trackPoint_changed, I would favor leaving its value up
to the browser, eg. implementation dependent, in the line sensor case
but perhaps there is a simple way to define the value in a way which
is consistent with regular 2d plane sensor use. If trackPoint_changed
is undefined in the line sensor case, then there would be already 4
browsers which comply with a refined PlaneSensor without any updates.

Any feedback welcome,

-Andreas

On Sun, Mar 24, 2019 at 4:05 PM doug sanden <highaspirations at hotmail.com> wrote:
>
> Sounds great for v4 proposal! Great work Andreas.
> -Doug
>
> ________________________________________
>
> I looked at the freewrl code for PointSensor and agree that it seems to clamp the translation in local sensor space. Not sure how the 3d clamping works with the 2d fields, probably the z is just left untouched. If so, translations clamped in xy but not z would end up off the tracking plane, I believe.
>
> I do not really think that this is useful for PointSensor. If clamping in the xy plane of the local sensor space is needed, a regular PlaneSensor seems like the better solution.
>
> So I am rather satisfied with the proposed way to extend PlaneSensor to also act as line sensor or point sensor. The line sensor role only needs a little bit extra prose, and the point sensor aka screen sensor role one additional field which I think is unavoidable. The slightly modified meaning of min.maxPosition will take an additional two sentences. There are two implementations which demonstrate the same functionality albeit in different ways.
>
> How does that sound for a v4 proposal ? Perhaps there is one already ?
>
> Andreas
>
> ---on the phone---
>
> On Fri, Mar 22, 2019, 7:39 PM GPU Group <gpugroup at gmail.com<mailto:gpugroup at gmail.com> wrote:
> Andreas - tried your x3dom sample - screen mode works great! Great work!
> For freewrl looks like I was clamping to min, max in 3D - ie if your max < min, then it ignors, otherwise it clamps in 3D space and I think that's in Sensor node space. Whether that's useful I don't know.Probably I just copied and pasted from PlaneSensor without testing or thinking too deep.
> -Doug
> https://sourceforge.net/p/freewrl/git/ci/develop/tree/freex3d/src/lib/input/SensInterps.c
>  - line 1500 do_PointSensor
>
>
>
> On Fri, Mar 22, 2019 at 2:46 PM Andreas Plesch <andreasplesch at gmail.com<mailto:andreasplesch at gmail.com>> wrote:
> Hi Doug,I took the plunge and added a planeOrientation='screen' field
> in this exploration:
>
> https://jsfiddle.net/andreasplesch/mzg9dqyc/158/
>
> The idea seems to work as advertised. In addition to dragging the
> colored axis arrows, it is now possible to drag the grey box itself in
> the screen plane whatever the current view may be.
>
> One thing I noticed is that the min/maxPosition field would need to
> have a slightly different meaning since the X and Y axis of the local
> coordinate system are not useful in this case.
>
> Did you have a min/maxPosition field for PointSensor ?
>
> One possible meaning would be to limit translation in the screen plane
> along the horizontal axis and the vertical axis of the screen. I
> thought a bit about that and I think it may be very doable but could
> not quite finish [The problem I am a bit stuck on is how to project
> the intersection point from local coordinates to the new coordinate
> system. Perhaps back to world, then world to camera space using the
> view matrix, then just setting z=0. Presumably this could work for the
> translation directly. Then clamp the translation, and project back:
> inverse view matrix and local transform.]
>
> Another option is to just ignore the field in this case.
>
> Cheers,
>
> -Andreas
>
>
> On Fri, Mar 22, 2019 at 11:25 AM GPU Group <gpugroup at gmail.com<mailto:gpugroup at gmail.com>> wrote:
> >
> > Great ideas and PlaneOrientation "screen" sounds intuitive, I like it. And saves a node definition, and you Andreas having solved the linesensor via planesensor, it makes sense to keep going that way, packing more in the PlaneSensor node. Keep up the great work.
> > -Doug
> >
> > On Fri, Mar 22, 2019 at 9:12 AM Andreas Plesch <andreasplesch at gmail.com<mailto:andreasplesch at gmail.com>> wrote:
> >>
> >> Hi Doug,
> >>
> >> I think a PointSensor is a good idea if there is currently no other
> >> way to drag a point around in a plane parallel to the screen. Since
> >> dragging is still restricted to a plane, perhaps ScreenSensor would be
> >> a better name ? This is a common and useful dragging method,
> >> especially if you have well defined viewpoints, like a perfect map
> >> view.
> >>
> >> To be more specific, the tracking plane is defined by the point on the
> >> sibling geometry first indicated when dragging starts and by the
> >> avatar viewing direction as the plane normal. The avatar viewing
> >> direction is the current viewpoint orientation, or the orientation of
> >> a ray to the center of the screen.
> >>
> >> Autooffset, I  think, works the same by just accumulating offsets for
> >> each drag action.
> >>
> >> And translation_changed and trackpoint_changed, I think, would still
> >> be reported in the local coordinate system of the sensor.
> >>
> >> Would it make sense to have a min. and maxPosition for it ? I think so.
> >>
> >> So I think it would be pretty straightforward to implement based on
> >> the existing PlaneSensor. Perhaps it should become a "screen" mode for
> >> PlaneSensor since it would have pretty much the same fields ?
> >>
> >> Then it may suffice to add only little prose to the PlaneSensor spec.:
> >>
> >> "If the mode field is set to "screen", the tracking plane is not the
> >> XY plane of the local sensor coordinate system. Instead, the tracking
> >> plane is the plane parallel to screen anchored by the initial
> >> intersection point."
> >>
> >> On the other hand, such mode fields are generally not used in any X3D
> >> nodes, probably for good reason. Perhaps planeOrientation would be a
> >> better name, with values "XY" (default), "screen", "XZ", "YZ".
> >>
> >> Although there is no screen in VR, the vertical plane perfectly ahead
> >> in the viewing direction is still an important plane and would be
> >> useful to drag around things in.
> >>
> >> More ideas,
> >>
> >> -Andreas
> >>
> >> On Fri, Mar 22, 2019 at 9:34 AM GPU Group <gpugroup at gmail.com<mailto:gpugroup at gmail.com>> wrote:
> >> >
> >> > Freewrl also does a PointSensor. I think I submitted that to spec comments, or I should have/meant to, for v4 consideration. Internally PointSensor is a planesensor aligned to the screen. So if you EXAMINE around something that's a PointSensor, you can drag it in a plane parallel to the screen from the current avatar viewing angle, and it feels like you can drag it any direction you want by changing the avatar pose.That's unlike a PlaneSensor that's not following/moving with the avatar.
> >> > And confirming what Andreas said -and in freewrl code for drag sensors it mentions Max Limper complaining about x3dom PlaneSensor the same thing about it failing when viewed on-edge.
> >> > -Doug Sanden
> >> >
> >> >
> >> > On Thu, Mar 21, 2019 at 9:46 PM Andreas Plesch <andreasplesch at gmail.com<mailto:andreasplesch at gmail.com>> wrote:
> >> >>
> >> >> I rethought the LineSensor idea and instead opted to add an internal
> >> >> line sensor mode to PlaneSensor. The idea is that if the x or y
> >> >> components of both minPosition and maxPosition are 0, PlaneSensor acts
> >> >> as a LineSensor and is free to use a better intersection plane than
> >> >> the default XY plane which can become completely unusable in the case
> >> >> when it is viewed on edge.
> >> >>
> >> >> In the implementation used in the fiddle below, the intersection plane
> >> >> normal in line sensor mode is computed as the cross product of the
> >> >> line axis (x or y axis) with the cross product of the line axis and
> >> >> the view direction when dragging starts:
> >> >>
> >> >> https://jsfiddle.net/andreasplesch/mzg9dqyc/111/
> >> >>
> >> >> It seems to work well and allows programming of a translation widget
> >> >> in X3D without scripts as shown. On edge view of the PlaneSensor XY
> >> >> plane still works because the XY plane is not actually used.
> >> >>
> >> >> This approach is more spec friendly as it does not require a new node
> >> >> and just allows the defacto use as LineSensor to become robust.
> >> >> However, since it requires use of another plane than the XY plane, it
> >> >> may be necessary to add some language. Perhaps:
> >> >>
> >> >> "This technique provides a way to implement a line sensor that maps
> >> >> dragging motion into a translation in one dimension. In this case, the
> >> >> tracking plane can be any plane that contains the one unconstrained
> >> >> axis (X or Y). A browser is therefore free to choose a tracking plane
> >> >> suitable to robustly determine intersections, in particular when the
> >> >> XY plane is unusable. In this case, the generated trackpoint_changed
> >> >> value becomes implementation dependent."
> >> >>
> >> >> I would prefer this over a LineSensor node, and it was pretty
> >> >> straightforward to add to the x3dom PlaneSensor implementation.
> >> >>
> >> >> I think it would be possible to check for other, equal values than 0
> >> >> and then use the same approach but I am not sure if it would be
> >> >> useful.
> >> >>
> >> >> If there is feedback, especially on the suggested prose, I will glad
> >> >> it to add it to a potential standard comment.
> >> >>
> >> >> For example, one could tighten the spec. by giving a formula for the
> >> >> orientation of a suitable tracking plane in the one dimensional case
> >> >> but that may be overreach.
> >> >>
> >> >> Cheers, -Andreas
> >> >>
> >> >> On Thu, Mar 21, 2019 at 6:06 PM Andreas Plesch <andreasplesch at gmail.com<mailto:andreasplesch at gmail.com>> wrote:
> >> >> >
> >> >> > I now understand the need for LineSensor and may give it a try for
> >> >> > x3dom. I know freewrl has it. It may make sense to replicate it.
> >> >> >
> >> >> > My initial idea would be to use the x axis of the local coordinate
> >> >> > system as the Line orientation. Or maybe the y axis since X3D
> >> >> > geometries tend to align along y (cylinder, extrusion).
> >> >> >
> >> >> > Then one can construct an intersection plane which includes the x axis
> >> >> > and another direction at a high angle with the viewing direction to
> >> >> > get a clean intersection for the inital point and subsequent points
> >> >> > during dragging.
> >> >> >
> >> >> > Is that how freewrl does it ? I think I saw that it may use an
> >> >> > additional field for the line orientation but it may be best to avoid
> >> >> > it.
> >> >> >
> >> >> > Andreas
> >> >> >
> >> >> > --
> >> >> > Andreas Plesch
> >> >> > Waltham, MA 02453
> >> >>
> >> >>
> >> >>
> >> >> --
> >> >> Andreas Plesch
> >> >> Waltham, MA 02453
> >> >>
> >> >> _______________________________________________
> >> >> x3d-public mailing list
> >> >> x3d-public at web3d.org<mailto:x3d-public at web3d.org>
> >> >> http://web3d.org/mailman/listinfo/x3d-public_web3d.org
> >>
> >>
> >>
> >> --
> >> Andreas Plesch
> >> Waltham, MA 02453
>
>
>
> --
> Andreas Plesch
> Waltham, MA 02453



-- 
Andreas Plesch
Waltham, MA 02453

More information about the x3d-public mailing list