Robotics involves complex processing and requires modular controllers.
For the connectionist approach, the adaptation of each module within
the global system remains a major problem to be solved. This paper pro
poses the idea that biological learning can take advantage of the stru
ctures of the modules and the nature of modular decomposition. Therefo
re, we address this problem starting with the architecture of the syst
em. We illustrate this approach using a robotic application: the visua
l servoing of the arm's end-effector. The on-line adaptation of a simp
le controller permits excellent results. To process several variables,
and to limit the size of the memory required, this controller is deco
mposed into modules, in the image of sensorial or motor processing cen
ters. The learning of the modules is realized on-line, a bidirectional
architecture permits the adaptation of each module using a simple alg
orithm. The results obtained with various modular arrangements, both d
uring intensive computer simulations and on our robotic platform, conf
irm the practical interest of this approach. (C) 1998 Elsevier Science
Ltd. All rights reserved.