[robotics-worldwide] Open Post-Doc position in Robot Audition, LAAS-CNRS, Toulouse, France

Patrick Danès Patrick.Danes at laas.fr
Mon Jan 18 06:51:40 PST 2010

Post-Doctoral position at LAAS-CNRS, Robotics Action Perception Group
18 months, starting immediately
Monthly gross salary: 2500EUR
Contact : Patrick Danes
	  Associate Professor, LAAS-CNRS & Universite Paul Sabatier
	  LAAS-CNRS; 7, avenue du Colonel Roche
	  31077 Toulouse Cedex 4 - France
	  E-mail: patrick[DOT]danes[AT]laas[DOT]fr


Robot Audition is a fairly  new research area.  The auditory modality,
which   is  near-omnidirectional   and  insensitive   to  illumination
conditions, constitutes an ideal complement to vision.  In the topical
Human-Robot Interaction context,  audition can enable the localization
of a speaker located outside the robot viewfield, his/her recognition,
the extraction of  his/her utterance out of the  environment, prior to
the interpretation of his/her message.

Two approaches to robot audition  can be envisioned.  On the one hand,
array processing methods exploit the  redundancy of the data sensed by
an array  of microphones so as  to improve the  properties of auditory
functions  under design:  resolution of  source  localization, spatial
filtering  efficiency,  robustness to  noise,...   In  this vein,  the
LAAS-CNRS has developed an integrated  auditory sensor on the basis of
a linear  array of 8 microphones  and a FPGA processing  unit.  On the
other hand, binaural approaches consist in fitting two ears on a head,
and  mimic  the human  system  to  various  extents.  Yet,  exploiting
bioinspired cues may require a  very accurate model of propagation, as
well as a careful  identification---generally in an anechoic room---of
the scattering  of the  wavefield induced by  the head.   Such systems
often  lack  adaptability  when  facing realistic,  a  priori  unknown
environments.    On  the  whole,   biomimetic  approaches   to  source
localization and extraction have so far led to mitigated results.

The proposed post-doctoral study  aims at designing and implementing a
bioinspired sensor  for the HRP2  humanoid robot hosted  by LAAS-CNRS.
This  sensor  will  comply  with  the  geometrical  and  computational
foundations involved in humans.  So, the HRP2 head will be fitted with
two pinnae  which reproduce the  spatial filtering performed  by human
ears.  Similarly, the physico-electrical transduction will reflect the
human operation,  e.g.  through the tonotopic frequency  coding in the
cochlea, the global response of  the hair cells, etc.  The sensor will
have  to  be  embeddable   and  reconfigurable,  in  order  to  enable
experiments  on  the  robot  as  well as  iterations  and  adjustments
necessary to an "optimal"  design.  On this sensory basis, bioinspired
acoustic  cues (interaural amplitude  and phase  differences, monaural
spectral cues,  etc.) will  be generated and  will serve as  basis for
localization.   An  evaluation  will  be  conducted on  how  a  strong
biomimetism affects  the accuracy  and robustness of  localization and
extraction strategies  in the changing  and unpredictable environments
typical of robotics.

These  scientific  objectives entail  software  development, but  also
hardware (design of the  pinnae, selection of the transducers, signals
digitization,  hardcoding  on  FPGA   of  functions  which  have  been
successfully  tested  offline, etc.).   The  work  will  draw on  past
achievements  at  LAAS,  with  the  support  of  the  engineering  and
technical staff of the laboratory.

This  post-doctoral  study takes  place  within  an internal  research
project at LAAS-CNRS, and  within the BINAAHR project (BINaural Active
Audition for Humanoid Robots) recently  accepted in the framework of a
joint  international  program  between  the French  National  Research
Agency  (ANR)  and the  Japan  Science  and  Technology Agency  (JST).

More information about the robotics-worldwide mailing list