Visualization and Analysis of Musculature Prediction in Simulation via Nonlinear Autoregressive Exogenous-Input Networks
Nicholas Benson
Massachusetts Institute of Technology
6.UR Undergraduate Research, Spring 2016
##Abstract
NARX networks operate on time-domain signal windows and output feedback windows to mimic the dynamics of a given system. I implement and test these networks to evaluate their feasibility for solving motion planning problems for a generic 18-muscle appendage analog developed in simulation using Unity, with an emphasis on visualizing the training and path response data to understand the potential and limits of the system as a whole. While drifting and trapping behaviors indicate flaws in either the training data or network scheme, the system is able to guide the appendage with some degree of reliability, and lays the groundwork for more sophisticated network paradigms for musculature prediction in the future.
##Full Text
The full text of the Vamps paper can be found [here.] (http://omono.me/vamps)