Learning from history: Adaptive calibration of 'tilting spine' fiber positioners

TL;DR

This paper introduces an adaptive calibration method for 'tilting spine' fiber positioners that uses location-specific historical data to improve accuracy in astronomical fiber positioning systems.

Contribution

It presents a novel adaptive calibration approach that updates parameters based on observed fiber movements, incorporating location-specific learning for better accuracy.

Findings

Significant reduction in positioning error with the new method

Performance varies with fiber location, requiring adaptive calibration

Prototype results demonstrate improved accuracy over traditional methods

Abstract

This paper discusses a new approach for determining the calibration parameters of independently-actuated optical fibers in multi-object astronomical fiber positioning systems. This work comes from the development of a new type of piezoelectric motor intended to enhance the 'tilting spine' fiber positioning technology originally created by the Australian Astronomical Observatory. Testing has shown that the motor's performance can vary depending on the fiber's location within its accessible field, meaning that an individual fiber is difficult calibrate with a one-time routine. Better performance has resulted from constantly updating calibration parameters based on the observed movements of the fiber during normal closed-loop positioning. Over time, location-specific historical data is amassed that can be used to better predict the results of a future fiber movement. This is similar to a technique previously proposed by the Australian Astronomical Observatory, but with the addition of location-specific learning. Results from a prototype system are presented, showing a significant reduction in overall positioning error when using this new approach.