Learning a peripersonal space representation for a humanoid robot

The representation of peripersonal space enables humans to integrate sensory data coming from multiple modalities into a coherent framework, which can then be used to control their actions. It has been argued that the body  schema plays a central role in structuring the representation of peripersonal space. In this talk, I will describe a model of the body schema that can be learned online by a humanoid robot looking at its limbs. I also show how it is possible to ground this body schema into the extrapersonal space, thus ensuring coherence between the peripersonal and extrapersonal space. This model can account for many neurobiological and psychophysical observations,  such as adaptation to distorted vision and tool use, fake limbs experiments, body-part centered receptive fields, and multimodal neurons. It is scalable to a high number of degrees of freedom and has been successfully implemented on the Hoap3 and ICub humanoid robots, thus enabling them to learn how to reach visually observed targets in their peripersonal space. I will also discuss this model and its implications in light of some current theories on cognition.

About the Speaker : Dr. Micha Hersch has completed his PhD in the LASA lab (Prof. Aude Billard) at EPFL, Lausanne, working on automatic body schema acquisition and adaptation and motor control in humanoid robots. Currently, he is a postdoc with Prof. Sven Bergmann at the Department of Medical Genetics headed by Prof. Jacqui Beckmann, University of Lausanne (UNIL) and at the Swiss Institute of Bioinformatics (SIB), where he is investigating light response of the model plant Arabidopsis Thaliana.