Team flowers

Overall Objectives
Scientific Foundations
Application Domains
New Results
Contracts and Grants with Industry

Section: Software


Participants : Pierre Rouanet [correspondant] , Pierre-Yves Oudeyer.

RoboDrive is a software which allows mobility during the interaction with a robot. By looking at the screen, the user can see what the robot is looking at. He can also drive the robot by sketching on the touch-screen using the stylus. Then, specific gestures are used to draw the attention of the robot towards new objects, and the software allows the user to associate a name to these objects. Moreover, the software allows the user to ask the robot to search and reach an object given its name/the word associated to it. Finally, the softwares provides a menu to trigger pre-programmed robot's behavior.

RoboDrive is based on an iPhone in order to use its multiple-interaction and multi-touch abilities and with the mid-term aim of making the software available to a larger audience. It is also providing interesting functionalities, such as driving the robot by litterally using the iPhone as a steering wheel (with the accelerometer). The interface has been designed based on iPhone API, making it much simpler and easy to use.

Figure 1. RoboDrive is an iPhone application that facilitates intuitive and robust human-robot teaching interactions

Users can also launch specific behavior such as “say hello”, “point”, “sit”, etc...

As mentioned above, this software allows users to associate names with new visual objects and so the interface has been designed to allow users and especially non-expert users to really provide the robot with good learning examples. Thus, when the user wants to teach a name for a new object, he first needs to encircle the object directly on the screen which provides a rough, but still very useful, segmentation of the image.

The software was also linked to a visual recognition and machine learning framework to allow a robust and fast recognition of any object. This framework is based on the bags of visual approach where lots of visual descriptors such as SIFT of SURF descriptor are extracted from an image or here only a portion of an image. These descriptors are then clustered into words and add to a vocabulary. Then we can use statistical methods based on the frequency of these words to recognized objects.


Logo Inria