This paper presents an original link between neural networks theory an
d mechanical modeling networks. The problem is to find the parameters
characterizing mechanical structures in order to reproduce given mecha
nical behaviors. Replacing ''neural'' units with mechanically based un
its and applying classical learning algorithms dedicated to supervised
dynamic networks to these mechanical networks allows us to find the p
arameters for a physical model. Some new variants of real-time recurre
nt learning (RTRL) are also introduced, based on mechanical principles
. The notion of interaction during learning is discussed at length and
the results of tests are presented, Instead of the classical {machine
learning system, environment} pair, we propose to study the {machine
learning system, human operator, environment} triplet. Experiments hav
e been carried out in simulated scenarios and some original experiment
s with a force-feedback device are also described. (C) 1998 Published
by Elsevier Science B.V. All rights reserved.