Data-driven acquisition offset correction for telescope networks using machine learning and sensitivity analysis
Abstract
Reliable target acquisition is essential for maximizing observing efficiency, particularly for robotic telescope networks that operate across many sites, instruments, and environmental conditions. Although these facilities generate large volumes of operational and observational data, acquisition procedures often continue to rely on static pointing models and manual adjustments by observatory staff. These approaches can be effective, but they may not fully capture changing behavior associated with telescope configuration, observing conditions, instrument state, or other time-dependent effects. We present a data-driven machine-learning pipeline for automatically predicting and correcting telescope acquisition offsets, demonstrated using archival operational logs from the Las Cumbres Observatory network. Our approach applies both traditional machine-learning methods and deep-learning regression models to learn relationships between measured acquisition offsets and a range of observing, telescope, and instrumentation conditions. By using historical performance data, the pipeline is designed to complement existing pointing models and reduce the need for repeated manual correction. We also present an open-source, Dockerized software service that makes the trained models accessible through a browser-based graphical interface. The service is intended to support straightforward, real-time use by observatory staff while remaining portable across computing environments. Together, the prediction pipeline and deployment framework provide a practical path toward more adaptive, automated, and data-informed telescope acquisition.
Related
- The impact of seasonality over the sensitivity of Einstein Telescope and the SNR of CBC signals at the Sardinia candidate site
- aRieL - Reinforcement Learning for the Ariel Space Telescope Scheduling
- Newly Identified Planetary Nebula Candidates in M31 using Deep-Learning Search of Hubble Space Telescope Images
- Chromospheric sensitivity of stellar spectral lines: an unsupervised machine learning classification
- C3NN-SBI: Learning hierarchies of N -point statistics from cosmological fields with physics-informed neural networks
BibTeX
@misc{zariski2026datadriven,
title = {{Data-driven acquisition offset correction for telescope networks using machine learning and sensitivity analysis}},
author = {Jackson Zariski and Kaitlin Kratter and Mark Bowman and Nandini Dharwadkar},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.34719},
eprint = {2610.34719},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.34719},
note = {CC BY-NC-ND 4.0},
}
RIS
TY - UNPB
TI - Data-driven acquisition offset correction for telescope networks using machine learning and sensitivity analysis
AU - Jackson Zariski
AU - Kaitlin Kratter
AU - Mark Bowman
AU - Nandini Dharwadkar
PY - 2026
DA - 2026/10/26
PB - arXiv
DO - 10.48550/arxiv.2610.34719
UR - https://arxiv.org/abs/2610.34719
AB - Reliable target acquisition is essential for maximizing observing efficiency, particularly for robotic telescope networks that operate across many sites, instruments, and environmental conditions. Although these facilities generate large volumes of operational and observational data, acquisition procedures often continue to rely on static pointing models and manual adjustments by observatory staff. These approaches can be effective, but they may not fully capture changing behavior associated with telescope configuration, observing conditions, instrument state, or other time-dependent effects. We present a data-driven machine-learning pipeline for automatically predicting and correcting telescope acquisition offsets, demonstrated using archival operational logs from the Las Cumbres Observatory network. Our approach applies both traditional machine-learning methods and deep-learning regression models to learn relationships between measured acquisition offsets and a range of observing, telescope, and instrumentation conditions. By using historical performance data, the pipeline is designed to complement existing pointing models and reduce the need for repeated manual correction. We also present an open-source, Dockerized software service that makes the trained models accessible through a browser-based graphical interface. The service is intended to support straightforward, real-time use by observatory staff while remaining portable across computing environments. Together, the prediction pipeline and deployment framework provide a practical path toward more adaptive, automated, and data-informed telescope acquisition.
ER -