- 1 VVV2012 EFAA Project
- 1.1 Participants
- 1.2 Short Term Projects
- 1.3 Tutorials
- 1.4 Directions to investigate:
VVV2012 EFAA Project
EFAA is a european project (http://efaa.upf.edu/) using the iCub, the Reactable and the iKart in order to develop human robot interaction. This page will sumarize our work during the timelapse of the summer school.
Integration objectives include: iCub mobility with the iKart, and re-calibration with the ReacTable after iKart mobility :)
- Maxime Petit (INSERM - Lyon - France)
- Stéphane Lallée (INSERM - Lyon - France)
- Grégoire Pointeau (INSERM - Lyon - France)
- Your name here ?
Short Term Projects
- Please attempt to the navigation tutorial of Marco
- Integrate that into the EFAA architecture somehow
- 1) Find the transformation between Map coordinates & iCub initial reference frame (front of the table)
- 2) Add an object to the OPC that is far away from the table
- 3) Change the call to PMP to:
- 1) Check if the target is in range
- 2) If not express this target into map coordinates
- 3) Navigate to the target
- 4) Send the command to PMP
- 4) Come back to the table and drop the object on the table
Skin sensing --> Emotions
- Preliminary demo working: iCub is "off", skin information is fed into OPC by awareTouch (Hector), read by IQR (Vicky) wich compute emotions of the robot to both wake up the robot, send them to the OPC and update the facial expression (Stephane). Congratulations !
- Next steps:
- classify the type of touch (poke, grab, caress)
- turn the iCub in a caressoholic !
Statut : RAD retrieve the emotion "state" which will be at 0 at first (sleep mode) and become 1 after a touch => allow to continue interaction
Calibration / Motor test with the reactable
- Calibration procedure has been debugged / improved.
- Next Steps:
- Tutorial on calibration (monday)
- Automatic recalibration after a motion
Differents modules existing :
- interactionManager (need POSTGRSQL): a SQL database with all the interaction of the robot. This module is a first version and will be upgraded. (Grégoire + Maxime)
- opcEars : module doing some comparisons of the OPC at different momenst. This module provides some temporality and returns the consequence of an action. (Grégoire)
- inputs : "snapshot name" : create a snapshot of the OPC stored at "name"
- inputs : "diff name1 name2" : return a bottle with the difference between the 2 OPC states name1 and name2.
- add some temporality to the OPC [INSERM]
- input : "snapshot + name" to create a snapshot of the OPC at a given time
- input : "difference + name1 + name2" : Output : a Bottle with the differences betwwen the 2 states of the OPC
- Recognized gestures are commited to the OPC as relations (Ilaria)
- A gesture-speech based game is achieved using RAD (Maxime)
- Next steps:
- Gestures could influence the robot emotions ? (Vicky?)
Statut : RAD spoken interface catching OPC relation in order to catch gesture from either Ilaria's or Kyuhwa's module
Tactile Shape Recognition
- Objects could be categorized by their physical shape, sensed through hand encoders
- Inputs: instance of each object
- Output: object classified
- The picture below shows the connections of the shapeRecognition module with the Supervisor and OPC modules
- Signals from the Supervisor module
- collect <left_arm/right_arm> <label>
- classify <left_arm/right_arm>
- ShapeRecognition module sends OPC module the relation of the object classified, i.e. "iCub detects a small object"
- Basic information is shared among modules through the OPC client (iCub already has some believes & emotions populated)
- We can go on and try to come up with a "drive based proactive demo"
Statut : RAD check if some behavior are allowed (sleep, work, play, socialize) and send the human go to hell if he asked something the icub don't want to do. Go to sleep automatically if it is the only thing to do
The modules, what they do, how to use them.
Current calibration iCub / Reactable (and reference frame management in general)
How to align iCub / reactable referentials and more generally how to align 2 different referentials using Ugo's calibration library.
efaaHelpers library and OPC format specification
How to access the OPC through it client and benefit from various classes to represent entities known by the robot.
Directions to investigate:
detect face [recog face]
- Output: x, y, z of the face [name of human]
decisionalModule – DAC (UPF/INSERM)
compute appropriate decision at multiple levels (DAC layers)
- Input: read the OPC (Working Memory) content
- Output: control robot action (triggers attentionSelector, pmpActionModule, emotionsInterface, speech)
- Change the actual decisional system (finite state machine) to a more biologically plausible system based on needs.
Human action prediction
- Input: Kinect (human) + ReacTable (object position)
- Output: Action currently executed + confidence, parameters…