Hareket seçimine ilişkin beyin esinlenmeli hesaplamalı modeller ve robotlar üstünde gerçekleme

Computational models of neural circuits enhances our comprehension of brain functions. In addition to the simulation of the models which helps to anticipate the cognitive processes, embodiment of these models is essential. Such embodiment would provide necessary setting to explain neural functioning ongoing in real environmens under oncoming sensory information. Also, these studies boost the work on intelligent systems by providing new approaches and techniques for the implementation of intelligent methods. Even though studies pursued in neuroscience can be considered as being in inception period, the embodiment of models done since now, reached the pre-results faster than the animal experiments. So, computational neuroscience is promising to lead further understanding of cognitive processes and design of related experiments. In this thesis, the main aim is to show the embodiment of computational models is possible for different scales of computational models that are biologically meaningful. Still another aim is also show that the implemented models are meaningful to get inference about the behavioural processes of brain circuits. For the embodiment part of the thesis, the Darwin-OP humanoid robot platform is utilized mainly, while the Bioloid robot environment is also considered to get some of the results. To realize the aims mentioned above, a temporal sequence task related to action selection is utilized. In this task, we investigated the associations between the sensory stimuli and desired actions, and also the mechanism by which reassociations result in development of new associations over the built up ones. Since the action selection is basically linked to the basal ganglia, thalamus and cortex (BTC) circuit in the brain, the BTC structures of brain are modeled in different scales to realize the considered task. The proposed models are the mass model approach of nonlinear dynamical system modeling and point neuron based models. In order to ensure the second aim, the mass model approach is deeply investigated to obtain some of the biological results with this model. Afterwards, the cortex part of the model is redesigned using point neurons to realize a more realistically plausible model. In addition to realization of BTC circuit, learning process is considered to make associations in order to select the right action in long term encountering. So, the temporal difference learning (TDL) is utilized to ensure the biological plausibility. Thus, reinforcement learning method is utilized for the learning part of the mass model. Although, TDL ensures the biological plausibility, it is a rule based model anyway. So, though it is possible to merge TDL with point neuron based models, spike timing dependent plasticity (STDP), which is more convenient from the biological aspect, is utilized for the learning part of the point neuron based action selection model. The investigation of the mass model shows that it is possible to obtain meaningful results from the biological aspect using the computational models. Another result of this thesis is that it is possible to implement different scales of computational models for cognitive processes into robots and run in real time applications. So, the results show that, using these computational models to realize complex tasks in future will infer further results. As a result, this thesis is a step to reach evaluating such cognitive models for the complex tasks in real environment and also, that it is possible in near future.