Resumen de Design of an emotion-based controller for dynamical systems
This thesis studies the effect of emulating human emotions in control systems. The introduction of a new control strategy applies well known facts from neuroscience and psychology about decision-making processes in humans, and combines them with computer science and artificial intelligence tools into control system design. The approach to the concept of emotion specified in this thesis requires a dynamical system to be controlled, a second system is employed as a reference model, and a mechanism for predicting the future dynamics for both systems. The comparison between these predictions triggers an emotion in the controller, and each emotion plus its intensity serves to define the appropriate input to the plant. Thus, the controller avoids negative emotions such as anger or fear, while seeking for positive emotions such as calm or satisfaction. This scheme causes the plant to approach the dynamic of the reference model. Given the importance of prediction in the emotion-based controller, this thesis proposed three algorithms to anticipate the value of a function. The first algorithm approximates the rate of change of any function given samples, the second reconstructs a periodic signal, and the third approximates the period of a periodic target function. In several tests, including linear and nonlinear systems, the proposed controller outperforms a classical controller for all considered cases. One of the main contributions of this research pertains to the reduction of the negative effects of nonlinearities in a plant by defining a linear reference model and leading the system to approach it. The application of emulated emotions in control systems opens a wide range of new solutions for dynamical systems, which may include aspects such as fault tolerance, motion planning, and control of complex systems.
