It is time to locate Connectionist representation theory in the new wave of robotics research. The utility of representations developed in Artificial Neural Networks during learning has been demonstrated in Cognitive Science research since the 1980s. The research reported here puts learned representations to work in a decentered control task, the disembodied arm problem, in which a mobile robot operates an arm fixed to a table to pick up objects. There is no physical linkage between the arm and the robot and so the robot's point of view must be decentered. This is done by developing a modular Artificial Neural Net system in three stages: (i) a Classifier net is trained with laser scan data; (ii) an Arm net is trained for picking up objects; (iii) an Inter net is trained to communicate and coordinate the sensing and acting. The completed system is shown to create new nonsymbolic transformationally invariant representations in order to perform the effective generalisation of decentered v...

The learning of sensory-motor functions have motivated important research works that emphasize a major demand: the combination of multiple neural networks to implement complex functions. A review of a number of works presents some implementations in robotics, describing the purpose of the modular architecture, its structure, and the learning technique that was applied. The second part of the chapter presents an original approach to this problem of network training, proposed by our group. Based on a bi-directional architecture, multiple networks can be trained online with simple local learning rules, while the robotic systems interact with their environment.