Feasibility Study of Dendritic Gated Networks for Upper Limb Prosthetic Control

This study provides insight into the synergy between neuroprosthesis control and dendritic gated networks (DGNs)—a newly introduced type of artificial neural network. These networks have demonstrated their potential in rich, dynamic environments but have yet to be deployed in the real world. We conducted an extensive offline analysis of DGNs on forearm prosthesis classification and regression tasks, and studied the influence of different hyperparameters on prediction quality. Our results suggest that DGNs are capable of learning usable predictions quickly and efficiently across different limb positions, highlighting their ability to learn in the presence of changing contexts and settings of use. Based on these findings, we recommend further investigation into dendritic gated networks for wearable robotic settings.