A Cacsd-based Alternative for Autonomously Developing Control Systems

The paper proposes a CACSD-based alternative for autonomously developing controllers for complex controlled systems. The proposed alternative deals with a neurallike controller with a special architecture. The parameters and the structure of the controller are estimated by a specific developing algorithm, during an off-line training stage. The developing algorithm is implemented like a multi-agent system, so that the agents cooperate one another. We also discuss in this paper how the proposed alternative can be extended to goal-oriented movement problems, such as robotics.