UI Improvements

From Openmoko

Jump to: navigation, search



 Obviously the tools are in the wild to build interfaces that could rival
 (or better IMO) anything Apple comes up with. We just need to organize
 this stuff. This would need hardware that can support dynamic
 interfaces. I can help here, too.

In fact, this place shall be dedicated to human-machine interactions improvements discussion;

Human-machine interaction can be separated into several aspects:

  • the physical contact/input device: the mono-touch touchscreen
  • the graphics:
    • accelerated rendering can add more consistency to zooming user interfaces, which seem to be quite an interesting concept for embedded scrensize-limited devices
    • adding physics "look and feel" to (ex: menu's) behaviours can add coherence too
  • the input methods
    • the virtual keyboard
    • handgestures

Finding inspiration ...

Seen Live


Zooming user interfaces


Science fiction


n-D navigation: the polyhedra inspiration

When we want to navigate files, mp3s in an mp3 player, etc... Every control that the application needs is a button. What about looking at the polyhedrons? We could find one for each usage, with as many surrounding subzones that may be used as controls. Ex: you need 5 buttons, let's take a pentagon with 5 surrounding zones all around. That way, it's always optimized...

http://en.wikipedia.org/wiki/Polyhedra http://en.wikipedia.org/wiki/List_of_uniform_polyhedra

Our weapons

We can't improve the human-machine interface without knowing the strengths / weaknesses of our hardware; some of the weaknesses might turn out as exploitable features, some strengths as limiting constraints.

The touchscreen


 What exactly does the touchscreen see when  you touch the screen with 2 fingers
 at the same time, when you move them, when you move only one of the 2, etc. I'm 
 also interested in knowing how precise the touchscreen is (ex: refresh rate, 
 possible pressure indication, ...)?


  • The output is the center of the bounding box of the touched area
  • The touch point skips instantly on double touch
  • Pressure has almost no effect on a single touch, but not so on a double touch. The relative pressures will cause a significant skewing effect towards the harder touch. You can easily move the pointer along the line between your two fingers by changing the relative pressure.


  • we can detect double touch as jumps, and that's all
  • no pressure
  • This could be an interesting input method for games - e.g. holding the Neo in landscape view, letting each thumb rest on a specific input area; probably needs to be checked for usability with a real device


 What does one see when sliding two fingers in parallel up(L,R)->down(L,R)?


  • In theory you see a slide along the center line between your two fingers. In practice, you can't keep the pressure equal, so you will see some kind of zig-zag line somewhere between the two pressure points in the direction of your slide.


 What does one see when narrowing two fingers in slide (=zoom effect on iphone)?


  • In theory you see the pointer stay at the center of the zoom movement. In practice, you can't keep the pressure equal for both fingers, so the pointer will move towards one of the two pressure points.

Graphics and computational capabilities

It would be good to report what performance the current hardware allows:

  • There was no pure X11 benchmarking done (AFAIK) (how many fps at full VGA scrolling, ex: 1024*480 image scrolling?)
  • what about the lcd reactivity? What if we don't see anything but blur while moving items fast?

Please report here your impressions.

Areas of improvement

  • OpenGL for fluid zooming interfaces (2D: the infinite sphere model, 1D: the infinite wheel of fortune/ribbon model, exposé)
  • HandGestures
  • Physics-model based improvements: inertia and friction
  • multi touch screen for natural handgestures
  • improved virtual keyboard
  • switching to another GUI toolkit (EFLs)

Physics-inspired animation a.k.a. "Digital Physics"

If we want to add eye candy & usability to the UI (such as smooth realistic list scrolling, as seen in apple's iphone demo on contacts lists for instance), we'll need a physics engine, so that moves & animations aren't all linear.

The following aticle explains the Digital Physics term from the iPhone example.

The most used technique for calculating trajectories and systems of related geometrical objects seems to be verlet integration implementation; it is an alternative to Euler's integration method, using fast approximation.

We may have no need for such a mathematical method at first, but perhaps there are other use cases. For instance, it may be useful to gesture recognition (i'm not aware if existing gesture recognition engines measure speed, acceleration...).

Open Dynamics Engine

ODE is an open source, high performance library for simulating rigid body dynamics. It is fully featured, stable, mature and platform independent with an easy to use C/C++ API. It has advanced joint types and integrated collision detection with friction. ODE is useful for simulating vehicles, objects in virtual reality environments and virtual creatures. It is currently used in many computer games, 3D authoring tools and simulation tools.



The akamaru library is the code behind kiba dock's fun and dynamic behaviour. It's dependencies are light (needs just GLib). It takes elasticity, friction, gravity into account.

If you want to take a quick look at the code: svn co http://svn.kiba-dock.org/akamaru/ akamaru

The only (AFAIK) application using this library is kiba-dock, a *fun* app launcher, but we may find another use to it in the future.

As suggested on the mailing list, it is mostly overkill for the uses we intend to have, but this library may be optimized already, the API can spare some time for too. Furthermore, Qui peut le plus, peut le moins.

Verlet integration implementation from e17

There's an undergoing verlet integration implementation into the e17 project (by rephorm) see http://rephorm.com/news/tag/physics , so we may see some UI physics integration into e17 someday.

Robert Pernner's easy equations


See the demo: implements non linear behaviour (actionscript), but may give inspiration

Extending the touchscreen capabilities and input methods

  • Multitouchscreen emulation

If we got it right, when touching the screen on a second place, the cursor oscillates between the two points depending on relative pressure distribution. Using averaging algorithms, we may have the opportunity to detect peculiar behaviours.

We need raw data (x,y,t) from the real hardware for the following behaviours:

  • slide two fingers in parallel - vertical up/down (scroll)
  • turn the two fingers around (rotate)
  • slide two fingers towards each other (zoom-)
  • slide two fingers apart (zoom+)

When touching the screen with two fingers at the same time, we necessarily see the two points, or are able to extrapolate the position of the second one. This solution can add feature, but will probably be little erratic...

  • Touchscreen kernel module hacking

We may correct the "half distance" phenomenon on double touch: if double touch is detected, then assimilate the cursor as twice further than the first touch. It would allow finer control, but higher instability.

The double touch detection may be implemented in the driver itself, as well as stabilization.

  • Other detectable behaviours

The warping can be used in the 4 diagonals, plus the up/down/right/left cross:

 ----------------  ----------------  ----------------  ----------------
 - 1            -  - 1          2 -  - 1            -  -      2       -
 -              -  -              -  -              -  -              -
 -              -  -              -  -              -  -              -
 -              -  -              -  -              -  -              -
 -              -  -              -  -              -  -              -
 -              -  -              -  -              -  -              -
 -             2-  -              -  - 2            -  -      1       -
 ----------------  ----------------  ----------------  ----------------

It's not double touch, but two sequential presses with a short time in between (~0.5 s)

  • Clever hacks

It's been said that having no multitouch screen allows less freedom for innovation. Maybe we could get something out of our touchscreen drivers.

Why ? Think of apple's scroll up/down feature on macbooks touchpads (which aren't multi touch, it's a clever driver hack, iscroll2):

 To scroll, just place two fingers on your trackpad instead of one. Both fingers
 need to be placed next to each other horizontally (not vertically, the trackpad 
 cannot detect that). Some people get better results with their finger spaced a
 little bit apart, while others prefer having the fingers right next to each other.
 iScroll2 provides two scrolling modes: Linear and circular scrolling.
 For linear scrolling, move the two fingers up/down or left/right in a straight 
 line, respectively, to scroll in that direction.
 Circular scrolling works in a way similar to the iPod's scroll wheel: Move the two
 fingers in a circle to scroll up or down, depending   on whether you move in a 
 clockwise or counterclockwise direction.

Maybe we can port/adapt/get inspiration from this macintosh driver.

Improved virtual keyboard

One nice idea for virtual input is finger-splash

Yet, optimization does not only apply to the plain one-letter-at-a-time input. We need some sort of T9 (dictionary-based input help). When typing a word, the first letter determines the next possible ones. Therefore, we may let disappear the less-probable following letters. Ex: type an L, there's no way an X follows...


  • ZIP's huffmann compression applied to SMSs/mails for detecting the most used characters/words/sentences.
  • html tag-clouds, one-letter tag clouds ; font size proportional to the probability of being used

The most critical point is the initial disposition of the letters, before any letter is typed. We may also want to use horizontal two-parts keyboard (with the neo in hands like a psp..)

The hexinput concept is interesting. What about hiding the less probable letters and increasing the remaining ones during typing?

A plain old dialpad like any other phone would be a nice fallback.

Towards OpenGL compositing

There are lots of possible GUI frameworks with various software architectures that could be used for Openmoko.

GTA01 hardware uses GTK+/matchbox without hardware acceleration, and it's not enough: this is a first that a mobile Linux device has such a high DPI resolution. OpenGL ES compositing seems to have a bright future on embedded devices, because compositing seems to give natural zooming interfaces reality (at last!).

Considering recent changes in destkop applications, opengl has a definite future. For instance, the expose (be it apple's or beryl's) is a very interesting and usable feature. Using compositing allows the physics metaphore: the human brain doesn't like "gaps"/jumps (for instance while scrolling a text), it needs continuity, which can be provided by opengl. When you look at apple's iphone prototype, it's not just eye candy, it's maybe the most natural/human way of navigating, because it's sufficiently realistic for the brain to forget the non-physical nature of what's inside.

So, opengl hardware will be needed in a more or less distant hardware, for 100% fluid operation. Benchmarking will be needed to compare the different alternatives that are cited further.

The Enlightenment Foundation Libraries

EFL's Evas is a powerful and power sparing canvas drawing library. It can be OpenGL accelerated. Python/Ruby bindings are available in the "proto" e17 cvs folder.

Moved here

Clutter Toolkit

Clutter, an OpenedHand project, is an open source software library for creating fast, visually rich graphical user interfaces. The most obvious example of potential usage is in media center type applications.


Clutter uses OpenGL (and optionally OpenGL ES) for rendering but with an API which hides the underlying GL complexity from the developer. The Clutter API is intended to be easy to use, efficient and flexible.

It does integrate gstreamer (for easy media playback, even camera or mic inputs), allows pango text rendering, cairo graphics rendering. Provided bindings are python, C# and Ruby.

GTK off screen rendering is supposed to be on it's way; once it is here, there will be a possibility of using GTK apps directly within OpenGL apps as textures, which would lead to the possibility of creating a full OpenGL "application manager" (as well as media consuming app) with ZUI features.


  • Core languages: C, OpenGL
  • Bindings: Python, C#, Ruby
  • Backends: GLX / SDL / EGL
  • Media support: Gstreamer integration, Cairo graphics rendering, Pango fonts rendering
  • Embedding: GTK embedding

Some experimental code in the clutter svn show that embedded devices are one of it's targets:


An early demonstration of Graff, which is a lighweight high-performance graphics rendering library. http://www.mdk.org.pl/articles/2007/04/23/chapter-1-in-which-we-meet-graff

Be sure to check out this demo (scrolling list with inertia scrolling) http://files.mdk.am/demos/graff-demo-3.avi

Of course it will remind you of Apple iPhone's UI. But this one runs in software mode on Nokia N770&800 already. The most notable part of Graff seems to be the inertia and physics integration in general.

Pigment API

Fluendo's (the Gstreamer guys) Pigment is a Python library designed to easily build user interfaces with embedded multimedia. Its design allows to use it on several platforms, thanks to a plugin system allowing to choose the underlying graphical API. Pigment is the rendering engine of Elisa, the Fluendo Media Center project.


  • Core languages: C OpenGL
  • Bindings: Python
  • Backends: DirectFB OpenGL
  • Media playback integration: using Gstreamer


Benchmarking will be needed. We have therefore to define a std testing application that would allow to compare alternatives.

Some Clutter VS Pigment information: http://www.taimila.com/?q=node/14

Improvement ideas

Please add here any idea that seems of relevance.

Kinetic scrolling

EDIT: demo now on the neo

Apparently the topics discussed here have found an echo in moko devs :)


enjoy the video demo; everything seems to be rendered software fine. More details about the implementation to come.


Graff's inertia scrolling list example: http://files.mdk.am/demos/graff-demo-3.avi

Because an example is worth 1000s words...


  • Sliding up/down = Single click + maintained for a minimal distance

Effect: scroll in an inverted/negated fashion (slide down = scroll up, slide up = scroll down)

When finger is released (i.e. touchscreen doesn't detect any press):

 if (last_speed_seen > 0 ) then keep this speed and acceleration, with friction
 else stop scrolling

Scrolling here is seen as unidimensional, but can apply to bidimensional situations (ex: zoomed image) too

  • Action = quick double tap
  • Details/select = short single tap
  • Right click = long tap
  • Sliding left/right: switch sorting method

Parts to modify

Having a scroll that isn't a 1:1 map to the user's action isn't hard. It's just an extra calculation in the scroll code.

<---- Where is the scroll code? :)

The best would be to add the feature for both finger and stylus scrolling.


  • make the entire list a "scrolling zone", i.e. an overlay transparent scrolling widget?
  • define controls
  • add the inertia feature

1D Scrolling: inertia friction integration into openmoko's finger wheel

The same, but for the wheel. It can be very short to do: you don't have 1:1 anymore, but, for example, 1/4 wheel turn = 1 item. It's demultiplicated, but has inertia.

Left-handed UI Support


A discussion on the community list identified a desire to have the ability to switch the Openmoko UI into "left-handed" mode.

The main problem is scrollbars, when they're on the right, dragging  
the scrollbar left handed results in your hand covering the screen so  
you can't see what you are doing. So having the option of scrollbars  
on the left would be useful.
I don't think the whole screen should be mirrored! There are some elements
that should remain..like the main top bar with the status icons and such.
Scrollbars are the main thing I can think of right now.

3D Launcher

Instead of a traditional filemanager type launcher or a start menu, we could use 3D capabilities and have a cylindrical launcher. Slide your finger left and right to rotate, up and down to move up and down the cylinder. There are arrows top and bottom to indicate you can move in that direction. For when you enter sub folders there will be a button top left/right allowing you to go back up a level.

Mock up of the launcher, the final version would have proper icons and the text would bend around the cylinder.


Handgesture recognition proposals

Using simple, localized warp as modifier key

As discussed on community list:

 If you hold down one finger and tap the other one, the mouse pops over and back
 again. If you keep your second finger touching, the cursor follows it. When you
 release it, cursor goes back to first finger position. This could be a way to
 set a bounding box or turn on the mode. So the second finger can do something
 like rotating around the first, or increase or lower the distance to the first.
  • the so-called "first touch" can be done on the mokowheel zone itself: put your left thumb on the black area; if you touch the screen with another finger, there is a warp; the warp is detectable and allows to enter "fake multi-touchscreen mode"
  • afterwards,
    * slide your right-hand finger down, it scrolls up
    * slide your right-hand finger up, it scrolls down
    * slide it left, next page/item
    * slide it right, previous page/item
    * do a circle: rotation
    * narrow towards the black circle: zoom -
    * go away: zoom +
  • if you had kept your first finger on the black quarter circle, you can continue issuing other gestures

The advantages of using simple origin-driven cursor warping as double touch detection criteria is that:

  • you don't have to use the wheel as button, which would slow things down and generate errors (false button presses)
  • simpler detection algorithms that can pass by the fluctuation due to pressure relative distribution
  • the space taken by the wheel itself is "useless" (i.e. displays no information); using it as modifier allows to keep the screen clean for reading
  • the origin of this zone lets use maximum surface of the screen, allowing more fine controlling

Fake multitouch rotate right.png

Fake multitouch scroll previous.png

Fake multitouch scroll down.png

Fake multitouch zoom-.png

  • Pros:
    • who said we need multi-touch hardware?
    • this may be the easiest way (in terms of design/implementation complexity, reliability)
  • Cons:
    • no matter what we'll invent, we'll need two hands for on-the-move controlling
    • what about left-handed people?

What is to modify ?

We need to emulate key presses. We need to work at a layer where we can get raw cursor coordinates. <---- X server layer?

There is a fake keyboard module (for dev purposes) in the main kernel tree, which could be used to simulate keyboard presses (hence keeping keyboard-enabled apps unmodified).

Full multi-touch emulation

Doable, but tricky...

Preparing the multi touch

One day we might get multitouch devices. Let's get ready.


The Multi-Pointer X Server is a modification of the X server to support multiple mice and keyboards in X. It provides users with one cursor per device and one keyboard focus per keyboard. Each cursor can operate independently. MPX is the first multicursor windowing system and allows two-handed interaction with legacy applications, but also the creation of innovative applications and user interfaces.

The multipoint X server project

MacSlow's Lowfat getting multitouched


If you want details, you can contact fursund

Open questions

  • will the neo/openmoko graphics system be powerful enough for such uses? I apple uses an OpenGL ES acceleration on this device (as well as on recent iPods), which is on the way with GTA02.
  • how does the touchscreen behave? We need a detailed touchscreen wiki information page, with visual traces. How hardware-specific is it?
  • is multi touch really that awesome? I guess not.
Personal tools