ECOLE NORMALE SUPERIEURE DE CACHAN

1
Needs for sensor fusion
Some examples Non destructive control (large
variety of geometrical, mechanical, thermal,
determinations, through photonics measurements,
in the form of surface fields) Investigation of
human brain activity (local field potentials,
MEG, functional NMR, fluorescence, laser
microscopy,) Robotics and C3I (including
defense applications, spatial observation of
earth Global Monitoring Environmental Security-,
navigation,collective behaviors) Man-machine
Interface (including neural control of prothesis
or environment) One general trend the
overwhelming importance of internal models (the
only way to cope with heterogeneous
sensors)  multiple clones of Virtual Reality
within the process  A new tool for
investigating sensor fusion the study of
illusions
2
Needs for sensor fusion
Architectures and artefacts Sensor fusion inside
the brain how do we make use of redundancies ?
hereafter head angular velocity, eyes
movements, vection and direction of gravity
3
Needs for sensor fusion
A new challenge for sensor fusion Virtual
Reality for cheap Recipe . Accept man as he/she
is as the fusion (post-) processor . Mediate the
outside world (environment of a production task,
of chirurgical actions) via Virtual Reality .
pre-process VR so as to make the best profit out
of the shortcuts or shortcomings of brains
internal representations The issue at
stake Simpler (and more economical) hardware
interfaces (e.g. isometric instead of isotonic
haptic device, in conjunstion with
vision) Prerequisite In depth study of
perceptual illusions (3D, haptic,) See for
instance recent work by LRP-INRIA-CEMIF-Matra-Aéro
spatiale Pseudo-haptic feedback can isometric
input devices simulate force feedback?
4
Needs for sensor fusion
SpaceballTM emulating a real spring embedded in a
piston
5
Needs for sensor fusion
The  unity of perception  issue (in engineering
or physiology) new solutions for an old problem
(Edelman). How do we maintain a unique internal
representation of external events in spite of
large computational and transmission delays
(sensor fusion at large)? 1st level (data base
management) maintenance of data (a priori or a
posteriori quality management, time of arrival,)
by a specialized middleware (a precursor of
which was SPICE, by Maarten Boasson, HSA) 2nd
level stability of the overall computation, for
instance by exchanging  wave variables 
(theoretical justification thanks to the
contraction analysis of Slotine Lohmiller)
see The intermediate cerebellum may function as a
wave variable processor, Massaquoi Slotine,
Neuroscience Letters 1996