Bio-inspired situated computing.
 
Bio-inspired situated computing.
Reinforcement Learning
 
 
 

Information, Multimodalité & Signal

Bio-inspired situated computing.
 
 

The organization of the cerebral cortex in mammals appears as a quite homogeneous neural circuitry, able to specialized on specific modal and multi-modal processing when specific skills are required at the global behavioral level. We aim at deriving computational paradigms for situated robotics from such considerations.

 
 

The cerebral cortex is a neural sheet, tiled of elementary neural circuits, the microcolumns. Such a tissue is quite homogeneous from a physiological point of view, while functional investigations have revealed clear specification over the cortical surface, according to modality, planning, etc. This suggests that self-organization of the whole cortical surface allows to tune each very part of it for achieving every single component of the global behavioral control. Keeping the whole surface highly adaptive while learning stable regularities from the world, maintaining coherence of all parts in such a distributed architecture without any supervision, are computational challenges for designers. On the basis of what the cortex actually does, and as far as possible on how it does it, our goal is to build a generic computational "substrate" that could exhibit self-organizing properties for the learning of autonomous robot controllers. This leads to the design of neuromimetic fine grain algorithms, applied on robotic platforms.

In a more philosophical point of view, the study of the brain in general, and more precisely the frontal part of the cortex that is engaged in planning and language, suggests that one could address artificial intelligence by the understanding of situated behavior. How can symbolic and abstract intelligence emerge from a "device" dedicated to situated and pragmatic interactions with the world ?

 
  Tools for cortical computing