Latest Telepresence and Visual Collaboration News:
Could Sweden's Social Robot Furhat be used for Telepresence?
Furhat is built so that we can bring a virtual animated face into the physical space. Many factors have been carefully taken into account to establish perfect fit with the projection model to the mask, and good visibility of the projected image.
There are two ways to acquire your own back-projected robot head similar to Furhat, the first one is to try a DIY. For that, we've outlined some of the important steps needed to built it below. However, due to a large interest in Furhat from the community and our interest to disseminate the technology behind Furhat due to its exceptional versatility and affordability, Sociobotics�(www.sociobotics.com) is now in the process of building the head (and similar heads) for interested parties. If you would like to acquire one, please contact�Sociobotics�for inquities.
We below outline the important steps that we've gone through to build the robot head. We hope this can be helpful to those who want to build their own head.
1- Use an animated face model. There are a number of animated face models that are available (academic or commercial). To build Furhat, we used our in-house developed 3D animated face models. The one that we used in Furhat is called Sven, and shown to the left of the image above. These models can be automatically lip-synchronized with a speech signal (or a phonetic label file) and can produce gestures and eye-movements.
2- Get a 3D mask. We believe that this is one of the more important steps that make Furhat look natural. It is possible to use an off-the-shelf translucent mask to project the animated model on, but we wanted to get a perfect fit of the projected model (match the size and location of the eyes, eyebrows, lips, nose...), to do this, we designed a 3D mask using the same projected model (Sven in this case, 2nd image to the left in the image above). We then printed the mask using a 3D printer. The printed translucent plastic mask is in the size of an average adult, and has a very high resolution and detail thanks to the advances in 3D-printing.
3- Uniform light diffusion. When light is projected on the printed mask, the mask will have different illuminations at different parts (this effect can be severe in the middle of the mask, that is where the lens of the projector is directly projecting). To solve this, a rear-projection paint is sprayed over the mask.�This gives a very strong effect and diffuses the light equally over the whole area that is cured with the spray.
4- Get a micro projector. Micro projectors are getting better every day (smaller and brighter). One option is to use a laser pico projector. This is light-weight, small, and quiet, and most importantly keeps the image in focus no matter how far away the projection surface is. The main limitation is that the laser pico-projectors in the market today are not bright enough to make Furhat visible under normal room lighting (15-20 lumens), so until laser projectors get better, we use a DLP micro projector with 100 ANSI Lumens.
5- Rig the mask with the projector for back projection. When the mask was tested and proved ready for back projection it was rigged with the micro-projector that was placed on top of the mask, the projector then projects light onto a mirror that reflects back the face onto the mask. This approach allows for more distance between the projector and the mask, which in turn, allows for the projected image to be in focus and to fit the entire mask (current projectors have limited projection area). An alternative approach is to use a fish-eye lens to directly project onto the face.�
6- Grab a neck that can support the head. In Furhat, we are currently using pan-tilt unit that offers his 2DOF (e.g. head shakes and head nods)�which allows Furhat to direct the head in any direction.
7- Give it a voice. If you want it to speak, get a Text to Speech Synthesis (TTS) system to support it.�FurHat gets his great synthetic voice from�CereProc
8- Program the brain.�Bring the head to life using some A.I. We belive that the power of Furhat comes from its behavior.�In Furhat, we have built frameworks for multiparty multimodal spoken dialogue systems, that can also generate gaze movements, head pose, expressions and gestures in addition to speech. Depending on the scenario, one can think of different "sensing devices" that can support the system with, such as proximity sensors, cameras, microphones... Furhat has no embedded sensors in it, and depending on the application, we use different sensory devices that are detached from the head itself.
If you want to refer to this work, please use this citation:
Al Moubayed, S., Beskow, J., Skantze, G., & Granstr�m, B.�(in press). Furhat: A Back-projected Human-like Robot Head for Multiparty Human-Machine Interaction. To be published in Esposito, A., Esposito, A., Vinciarelli, A., Hoffmann, R., & C. M�ller, V. (Eds.),�Cognitive Behavioural Systems. Lecture Notes in Computer Science. Springer�[Draft]
Add New Comment
Telepresence Options welcomes your comments! You may comment using your name and email (which will not be displayed), or you may connect with your Twitter, Facebook, Google+, or DISQUS account.
See what happens when YouTube and TPO come together at the Telepresence Options YouTube Channel.