This thesis presents methods, techniques and architectures to build intelligent control systems. The Kalman filter potentialities are extended to several domains, such as robust adaptive control, data fusion and fuzzy learning. The main contributions are: 1) The introduction of stochastic active observers. Model reference adaptive control is achieved through Kalman filtering techniques. 2) Applying stochastic active observers, a six degrees of freedom compliant motion controller is designed to perform the robotic peg-in-hole task. 3) A data fusion paradigm is proposed to fuse vision and pose data. It consists of two independent modules for optimal fusion and filtering. The data fusion algorithm applies stochastic evolutionary concepts linked with Kalman techniques. The data fusion architecture has artificial neural networks that map geometric information into desired compliant motion signals. Off-line learning is based on human demonstration. 4) A different perspective of fuzzy control is introduced. A method of fuzzy cloning and rule learning is proposed. The cloned rules are obtained directly from state space control. Kalman gain properties are the basis for rule learning.