Cerebellar Model Articulation Controller with introspective voting weight updates for quadrotor application

This paper explores control of a four rotor aerial vehicle, or quadrotor. The quadrotor exhibits under-damped and under-actuated dynamics, so designing a feedback control that can provide robustness to disturbances such as wind and adapt to payloads provides a research challenge. The Cerebellar Model Articulation Controller (CMAC) is a type of neural network that can provide the basis for a stable adaptive neurocontrol. However, traditional robust weight update schemes for the CMAC must often sacrifice performance to eliminate bursting caused by weight drift. This work applies an introspective voting weight update scheme to a CMAC to avoid busting, without reducing performance. The proposed scheme achieves stable, adaptive, and robust control in both simulation and experiment.