Aude Billard developed in her MSC Thesis a neural network control architecture called DRAMA which we are applying to robots which are communicating and interacting in a social context, given a teacher-learner setup. We are using a simple imitation strategy (bi-directional following) to implement "social bonding". A hilly environment is used as a "meaningful" habitat for the robots.
Papers about this work
This is a technical report (University of Edinburgh) about a project which we had done in December 1996 at the VUB AI Lab in Brussels.
A paper about
this project had been accepted for presentation at ECAL97.
Here
is the abstract.
The paper Grounding communication in situated, social robots was presented at
TIMR,
Manchester, Towards Intelligent Mobile Robots TIMR UK 97. The paper is published in the Proceedings (Technical Report Series of the Department of Computer Science, Manchester University).
In the following we show some photos which we took during the experiments
in Brussels.
The four agents involved in this project: Aude (left) and Kerstin (right), plus two fischertechnik robots.
The teacher (left) and the learner (right) robot in the initial position. The robots are not identical, they have different shapes, plus sensori-motor characteristics. We assume that the teacher robot "knows" how to interpret the world, i.e. it is emitting 2 different signals (bitstring, by radio link communication) for moving on a plane and moving on a hill.
The learner robot. It has to learn the teacher's interpretations of "words" on the basis of its own sensory inputs. Learning means here creating associations.
Approaching the hill. The robots have a bi-directional following strategy, i.e. both teacher and learner try to maintain the following relation.
Climbing up the hill. Every run is different from the other, due to the structure of the ground, sensory disturbances etc.
On top of the hill. After a few of such runs the learner has learnt that certain "words" are associated with a certain context, defines by sensory inputs.
Technical details, experiments, results, as well as problems and disadvantages of this approach to robot communication are discussed in the papers which are mentioned above.
