The development of Autonomus Agents capable to survive and develop tasks in a complex, dynamic, unforeseeable and even hostile environment, as is the real world, is a very challenging research area. Life beings, as mammalians, birds, insects or even worms, present a bigger survival capacity in the real world when compared with any Autonomous Agent ever developed.

This work proposes that biological inspiration can be the source of mechanisms and solutions that, once understood and implemented, allow the development of Autonomoous Agents with high level of autonomy and utility.

Computational Intelligence, in this work, is understood as adaptive behavior that allows the system survival and operation in a given environment. Looking for a biological inspiration, its considered that the Evolutionary Algorithms give the computational model that simulates, even in a simplistic fashion, the natural procedures of reproduction, mutation and selection that allow the evolution of individual structures searching the most capable solution for a given environment. What concerns about modeling aspects of animal behaviors, we consider and use the Connectionist approach as an apropriate tool to model and simulate behavioral aspects when implementing Autonomous Agents.

This work presents the idea that a series of behavior classes, observed in animals, can be implemented through different Artificial Neural Networks architecture's. We also emphasize that these different architectures can be obtained using Evolutionary Algorithms. We try to confirm the hypothesis that pure reflexive behaviors could be implemented through simple static neural networks architecture's like feedforward architectures. More complex behaviors, as reactive behaviors, that persist and develop, even after finished the sensorial exciting stimulus, require more complex neural networks architecture's, like a recurrent one, with dynamic neurons to insert dynamic and memory in the system.

Firstly, motivational facts and some basic concepts are presented. Just afterwards the concepts of Autonomous Agents and the main implementations strategies are presented as well as a formal conceptuation, using the General Systems Theory is proposed. After that, the concepts of behavior, learning, evolution and nervous system, are described and analysed, both from the biological as from the equivalent computational point-of-view. Finally, some original contributions based in the previously presented subjects and that allow a new strategy in Autonomous Agents implementation are described. Some implementations of simple examples that illustrate the ideas are presented. A summary of what had been discussed is presented and open points for future works are listed.

Autonomous Agents, Neural Networks, Evolutionary Algorithms, Artificial Life, Robotics.