Coupling Dynamical and Connectionist Models: Representation of Spatial Attention via Learned Deictic Gestures in Human-Robot Interaction

AbstractA proper representation of space and a joint attention mechanism are indispensable for an effective deictic communication with embodied agents. Taking inspiration from developmental psychology may help us to tackle computational challenges for robots. Although some developmental joint attention models for robots have already been proposed, to the best of our knowledge, there is no such model that can stand for the effects of pointing gestures on covert attention in infants. Thus we have designed and implemented a developmental robotics model for joint spatial attention combining connectionist and dynamical approaches. The hybrid architecture was structured over two existing computational models: a connectionist model of gesture comprehension and a Dynamic Field (DF) model of spatial attention in infants. These models were extended with various perceptual modules and dynamical neural fields, and implemented on the state-of-art iCub humanoid robot. In this paper, the computational architecture is introduced with some preliminary results that show the model’s capability of representing deixis and perceived objects, and their effects on attention over space and time.

Return to previous page