So far, a spiking cerebellar modular network has been embedded in a closed-loop control system for controlling a robot by learning its inverse or forward internal models. The student will have to define a method for combining both inverse and forward modules in order to implement a composite architecture. Then he/she will have to study how to learn multiple inverse and forward models for different manipulated objects by a robot, how to retain the learning and switch between the models. As a result of this a gaiting network will be implemented and trained to generate a weighted prediction of the multiple inverse and forward models, in order to determine the locally responsible models. Finally, the student will have to design two processes for switching between the models, based on feedforward and feedback outcome. Test and validation phases with a simulated robot will be run in the Neuro-robotics platform developed within the European Human Brain Project. Read the pdf attached for having a better idea of the field as well as papers from our group. For more information, send an email for appointment.
Knowledge in Robotics, Control theory, Programming in Python and/or C, Neural Networks, Machine Learning, Ros