ETD system

Electronic theses and dissertations repository


Tesi etd-11142019-145645

Thesis type
Tesi di laurea magistrale
Learning and Modulating Motor Commands using FORCE-trained Spiking Neural Networks
Corso di studi
relatore Prof.ssa Laschi, Cecilia
tutor Dott. Morin, Fabrice
relatore Dott. Falotico, Egidio
correlatore Dott. Vannucci, Lorenzo
Parole chiave
  • spiking neural networks
  • reti neurali a spike
  • FORCE learning
  • apprendimento FORCE
  • Izhikevich model
  • modello di Izhikevich
  • robotics
  • robotica
  • motor control
  • controllo motorio
Data inizio appello
Secretata d'ufficio
Data di rilascio
Riassunto analitico
The objective of this project is to make a step toward achieving human-robot collaboration using neurocontrollers. The focus is mainly on motor control by means of a recurrent spiking neural network, trained with a technique called FORCE learning.
Moreover, the challenge of having the robot operating at different speeds is addressed using a special input to the neural network, referred as high-dimensional temporal signals (HDTS), which is inspired by the synchronized input pattern that songbirds use for learning and replay, proposed as an integration to the FORCE learning method.
The aim is to use a single neural network to learn a set of primitives that would allow an industrial robotic arm to reach several points in a workspace, performing the trajectories at different velocities, and return to a home position.
To this end, several techniques will be explored to adapt the FORCE learning method to learn low-frequency signals, which are the main interest of collaborative robotics. The successful techniques will then be applied in the context of behaviourally-relevant motor tasks, to learn four trajectories of joints positions.
Finally, the neural network trained on the joint trajectories will be integrated in a robotic arm simulated in the HBP Neurorobotics Platform (NRP).