this paper, our aim is to show that such behavior, including switching between goals, can be simulated by simple artificial Neural Networks (NN) where no complex computation is performed. We will present a real development and simulations about a Khepera^TM robot (fig. 1) and a simulated system named Prometheus. Figure 1: The Khepera^TM robot developed at the LAMI [Mon93]. We use a novel neural architecture named PerAc (Perception-Action) which is a systematic way to decompose the control of an autonomous robot in perception and action flows [Gau94b]. We show that action simplifies the interpretation of perception: each action is a choice and conditions entirely the future of the robot. That way, acting in the world is necessary to the categorization and hence the interpretation of perceived signal, i.e., to the emergence of an elementary "cognition". The greatest advantage of this type of approach is that it makes cognition sequential, thereby avoiding the possible large duplications and relaxation mechanisms needed by massively parallel systems. We also focus on the interest to perform an autonomous on line learning of the relevant places to the robot in its environment. Furthermore, we compel ourselves not to touch modify the internal structures of the artificial robot "brain" by hand, while it operates. Thus we have to: - design a self modifiable connection diagram. - pay attention to the self adaptability of each simple block to data variations. - allow the robot to use the signals correlations which are really relevant to it. - introduce a limbic system to control the robot's learning, motivation and behaviors. We emphasize the interest of a constructivist approach [Mat87], [Ste91] as implemented for instance by the subsumption architecture [Bro86]. A special stre...

In this paper, we propose a novel neural architecture named PerAc which is a systemetic way to decompose the control of an autonomous robot in perception and action flows. We first present an application of the PerAc architecture to the simulation of a vision system with a moving eye. Then we propose a second application where the robot learns to return from any starting place to a previously discovered and learned position without any a priori symbolic representation.

neurocomputation involves discrete signals communicated along fixed transmission lines between discrete computational elements. This concept is shown to be inadequate to account for invariance in recognition, as well as for the holistic global aspects of perception identified by Gestalt theory. A Harmonic Resonance theory is presented as an alternative paradigm of neurocomputation, that exhibits both the property of invariance, and the emergent Gestalt properties of perception, not as special mechanisms contrived to achieve those properties, but as natural properties of the resonance itself.