Guidance and Control Networks with Periodic Activation Functions

TL;DR

This paper introduces a modified Guidance & Control Networks (G&CNETs) using periodic activation functions, inspired by SIRENs, which train faster and achieve lower errors across multiple control scenarios.

Contribution

The paper presents a novel G&CNET variant with periodic activations, demonstrating improved training speed and accuracy over previous versions.

Findings

Faster training convergence with periodic activations.

Lower overall training error in control tasks.

Preliminary analysis explains the superior performance of SIREN-based G&CNETs.

Abstract

Inspired by the versatility of sinusoidal representation networks (SIRENs), we present a modified Guidance & Control Networks (G&CNETs) variant using periodic activation functions in the hidden layers. We demonstrate that the resulting G&CNETs train faster and achieve a lower overall training error on three different control scenarios on which G&CNETs have been tested previously. A preliminary analysis is presented in an attempt to explain the superior performance of the SIREN architecture for the particular types of tasks that G&CNETs excel on.