Bio-inspired robotic control schemes using biologically plausible neural_structures

Resumo

The cerebellar circuitry belonging to the Central Nervous System (CNS) consists of a set of neurons and synapses that present rich dynamical properties. In the field of traditional artificial neural network (ANN), most approaches are based on very simplistic connectivity rules between very simplified neuron models which produce an output in each propagation cycle (the time domain is not introduced in the simulation). Nevertheless, if we want to study computational neuroscience and consider biological nervous systems, we need a higher degree of detail. For instance the cerebellum does not consist only of very complex neurons in a sophisticated network but also this cerebellum network deals with non-continuous signals called spikes. Hence, it is clear that in order to understand the foundations of cerebellar processing (from a computational neuroscience perspective); it is mandatory to work with a realistic cerebellar spiking neural network. In the case of the Cerebellum, its functionality has been studied for decades and it is well accepted that it plays a fundamental role in human motor control loops by means of regulating movement and also cognitive processes. The cerebellum is able to dynamically regulate its activity (it can present a highly non-linear behavior) and it is also able to tune its synaptic connections by distributed and heterogeneous forms of synaptic plasticity. Along this thesis, we have focused on studying the cerebellar functionality and how it is related with its structure (network topology), neuron models and synaptic adaptation mechanisms. To that aim, we have developed a biologically inspired cerebellar like network (based on neurophysiological findings) embedded into a robotic system in order to evaluate circuit functioning under closed-loop conditions. According to the embodiment concept, we have developed a complete framework that allows researchers to contrast different experimental cerebellar hypotheses. This work was partly supported by the Spanish Subprogram FPU 2007 (MICINN), and the EU projects SENSOPAC (IST-028056), and REALNET (IST-270434).