Dynamic significant feature extraction for embedded intelligent agent implementations

“Autonomy” and “adaptability” are key features of intelligent systems with environment awareness. Many applications of intelligent agents require the processing of information coming in from many available sensors to produce adequate output responses in changing scenarios. For such applications, the concept of autonomy should apply not only to the ability of the agent to produce correct outputs without human guidance, but also to its potential ubiquity and portability. However, processing complex computational intelligence algorithms in small, low-power embedded systems, very often with tight delay constraints, is a challenging engineering problem. In this paper a computationally efficient neuro-fuzzy information processing paradigm is tested in an ambient intelligent scenario to evaluate its appropriateness for future embedded SoC (System on Chip) implementations. The system has been endowed with an information preprocessing module based on Principal Component Analysis (PCA) that produces reduced input space dimensionalities with little loss of modeling power. An eventual on-chip PCA module could be applied to dynamically update the reduced meaningful space of information from the outside world. Moreover, the applicability of the PCA module to obtain a fault-tolerant agent in the presence of sensor failures has also been investigated with satisfactory results.