Bio-inspired autonomous dynamical systems for interactive and cognitive environments

Interaction and cognition are two strictly related functionalities of biological systems that are not necessary considered as joint aspects when the development is considered of artificial autonomous dynamical systems. In particular, many evidences from neurosciences [1] tell us that cognitive reasoning on world states and events in biological entities can be described in terms of a joint (distributed) dynamical representation of the interaction of the (represented) internal states of an organism with the (represented) states of the objects that move in the contextual scenario. Such a joint representation of dispositional and situation based knowledge [2] is a key aspect within current neuroscience models to explain higher level performances of natural cognitive systems. As reverse engineering of brain working modalities [3] [4] is showing that new computational approaches can be inspired by studies of human brain functioning not only at neuron level but also in terms of higher level functionalities it becomes of interest to investigate what could be the impact of including new basic object representations within autonomous dynamical systems aiming at providing services within cognitive environments. In this talk, a brief overview of Cognitive Systems, namely, using terminology introduced by Simon Haykin [5], Cognitive Dynamical Systems will be provided. The concept of cognitive cycle and the models are presented so far proposed to represent under a single category systems like Cognitive Radios, Cognitive Radar, Cognitive Surveillance, etc. Then, the problem of the impact of selecting appropriate representations of joint dispositional-situational knowledge on the architecture and inference processes of Cognitive dynamical systems will be addressed. In particular, the possibility and the techniques suitable for of using new Probabilistic Graphical Models (PGM [6]), namely a new type of Coupled Dynamical Bayesian Networks (ICASSP 2012 [7]) for describing interactions- between systems “dispositions” and contextual “situations” will be introduced that resonates with Damasio´s theories. It will be shown that such models are capable to describe both observed interactions and dynamical interactions involving system´s interactions with external users/operators. This capability will be shown to be a key issue in allowing a system to re-use interaction knowledge learned by observing external interacting entities as a basic knowledge to perform the same type of interactions when required to do, provided that the considered problem allows a mapping from observation space into an internal system action space. Examples coming from running projects will be provided that concern with Cognitive multisensory surveillance: in particular related crowd behavior monitoring and control will be considered where an operator based control of a cognitive environment will be shown to be learned by an autonomous dynamical system on the basis of the same knowledge representation that the system uses for describing interactions between observed entities. Other examples describing other application fields, like intention based driver assistance in intelligent vehicles and security in cognitive radio networks will be discussed.