In animals, achieving an efficient and accurate movement is complicated. It needs coordinated activity from a group of regions within the nervous system. In this thesis, we studied motor control by developing a Spiking Neural Network (SSN) controller to show efficient motor control. The cerebellar circuit and neuron models responsible for generating motor commands and planning trajectories are presented in previous work. However, previous versions of the controller circuit did not have proper synaptic plasticity mechanisms. In this thesis, we advanced the implementation presented by introducing novel advanced features. We introduced supervised learning to the motor cortex to enhance learning capabilities, as well as STDP-driven plasticity mechanisms in the cerebellar components. For the main inputs of the internal cerebellar model, a radial basis function approach was used to encode the input signals. The EBRAINS Neurorobotics Platform (NRP) and MUSIC platform were leveraged for the orchestration of simulations encompassing both neural and robotic parts. To preserve scalability and portability for larger-scale neural networks, the new design also utilizes HPC clusters. To ensure the biological plausibility of system behavior, the system also provides a realistic portrayal of sensory-motor delays across key cerebral pathways. From the previous prototype, this new architecture is a major step toward a modern, scalable, adaptive neuromorphic controller that can function well in virtual robotic environments.
Negli animali, ottenere un movimento efficiente e preciso è complicato. Richiede l'attività coordinata di un gruppo di regioni del sistema nervoso. In questa tesi, abbiamo studiato il controllo motorio sviluppando un controller Spiking Neural Network (SSN) per dimostrare un controllo motorio efficiente. Il circuito cerebellare e i modelli neuronali responsabili della generazione dei comandi motori e della pianificazione delle traiettorie sono trattati in lavori precedenti. Tuttavia, le versioni precedenti del circuito di controllo non disponevano di meccanismi di plasticità sinaptica adeguati. In questa tesi abbiamo migliorato l'implementazione presentata, introducendo nuove funzionalità avanzate. Abbiamo introdotto l'apprendimento supervisionato nella corteccia motoria per migliorare le capacità di apprendimento, nonché meccanismi di plasticità guidati da STDP nelle componenti cerebellari. Per gli input principali del modello cerebellare interno, è stato utilizzato un approccio basato su funzioni di base radiali per codificare i segnali di input. La piattaforma di neurorobotica EBRAINS (NRP) e la piattaforma MUSIC sono state utilizzate per l'orchestrazione di simulazioni che includono sia componenti neurali sia robotiche. Per preservare la scalabilità e la portabilità delle reti neurali su larga scala, il nuovo design utilizza anche cluster HPC. Per garantire la plausibilità biologica del comportamento, il sistema fornisce anche una rappresentazione realistica dei ritardi sensomotori lungo i principali percorsi cerebrali. Rispetto al prototipo precedente, questa nuova architettura rappresenta un importante passo avanti verso un controllore neuromorfico moderno, scalabile e adattivo, in grado di operare efficacemente in ambienti robotici virtuali.
Development and testing of closed-loop cerebellar models for human reaching motor control
Mazlum, Volkan
2024/2025
Abstract
In animals, achieving an efficient and accurate movement is complicated. It needs coordinated activity from a group of regions within the nervous system. In this thesis, we studied motor control by developing a Spiking Neural Network (SSN) controller to show efficient motor control. The cerebellar circuit and neuron models responsible for generating motor commands and planning trajectories are presented in previous work. However, previous versions of the controller circuit did not have proper synaptic plasticity mechanisms. In this thesis, we advanced the implementation presented by introducing novel advanced features. We introduced supervised learning to the motor cortex to enhance learning capabilities, as well as STDP-driven plasticity mechanisms in the cerebellar components. For the main inputs of the internal cerebellar model, a radial basis function approach was used to encode the input signals. The EBRAINS Neurorobotics Platform (NRP) and MUSIC platform were leveraged for the orchestration of simulations encompassing both neural and robotic parts. To preserve scalability and portability for larger-scale neural networks, the new design also utilizes HPC clusters. To ensure the biological plausibility of system behavior, the system also provides a realistic portrayal of sensory-motor delays across key cerebral pathways. From the previous prototype, this new architecture is a major step toward a modern, scalable, adaptive neuromorphic controller that can function well in virtual robotic environments.| File | Dimensione | Formato | |
|---|---|---|---|
|
2025_12_Mazlum_Executive_Summary_02.pdf
accessibile in internet per tutti a partire dal 18/11/2026
Descrizione: executive summary of the thesis
Dimensione
3.37 MB
Formato
Adobe PDF
|
3.37 MB | Adobe PDF | Visualizza/Apri |
|
2025_12_Mazlum_Thesis_01.pdf
accessibile in internet per tutti a partire dal 18/11/2026
Descrizione: thesis text
Dimensione
26 MB
Formato
Adobe PDF
|
26 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/247209