Develop and validate autonomous decision-making methods that select both the next materials experiment and the appropriate measurement modality

The National Institute of Standards and Technology awarded a $150,000 cooperative agreement to Theiss Research on February 1, 2027, to develop and validate autonomous decision-making methods that select both the next materials experiment and the appropriate measurement modality under the Measurement and Engineering Research and Standards program (CFDA 11.609).

Theiss Research will develop a non-myopic, measurement-modality-aware active-learning framework that combines multi-fidelity active learning with large language model-derived scientific priors to improve the efficiency and reliability of advanced materials characterization. The recipient will implement a multi-task surrogate model linking lower-fidelity UV-VIS measurements with higher-fidelity small-angle X-ray scattering measurements, develop an LLM pipeline to extract and apply literature-based relationships between measurement modalities, and demonstrate the resulting framework through autonomous phase mapping of peptide-guided gold nanoparticle synthesis using NIST Autonomous Formulation Laboratory infrastructure. The recipient will benchmark measurement efficiency against single-fidelity approaches and disseminate methods, software, data, and scientific findings.

Expected outcomes include a validated algorithmic framework for hierarchical measurement scheduling; an LLM-assisted method for incorporating prior scientific knowledge into autonomous measurement decisions; a high-resolution phase map of a peptide-guided gold nanoparticle system obtained with fewer costly measurements; quantitative benchmarks of time and measurement savings; and openly available software, data, documentation, and publications enabling reuse by other autonomous laboratory programs.

Performance occurs at Theiss Research's La Jolla, California facility. The period of performance runs from February 1, 2027, through June 1, 2028.

Purpose

The purpose of this cooperative agreement is to develop and validate autonomous decision-making methods that select both the next materials experiment and the appropriate measurement modality. The work will combine multi-fidelity active learning with large language model-derived scientific priors to improve the efficiency and reliability of advanced materials characterization.

Activities

The recipient will develop a non-myopic, measurement-modality-aware active-learning framework; implement a multi-task surrogate model linking lower-fidelity UV-VIS measurements with higher-fidelity small-angle x-RAY scattering measurements; develop an LLM pipeline to extract and apply literature-based relationships between measurement modalities; and demonstrate the resulting framework through autonomous phase mapping of peptide-guided gold nanoparticle synthesis using NIST autonomous formulation laboratory infrastructure. The recipient will benchmark measurement efficiency against single-fidelity approaches and disseminate the resulting methods, software, data, and scientific findings.

Expected outcomes

Expected outcomes include a validated algorithmic framework for hierarchical measurement scheduling; an LLM-assisted method for incorporating prior scientific knowledge into autonomous measurement decisions; a high-resolution phase map of a peptide-guided gold nanoparticle system obtained with fewer costly measurements; quantitative benchmarks of time and measurement savings; and openly available software, data, documentation, and publications that enable reuse by other autonomous laboratory programs.

Intended beneficiaries

Academic, government, and industrial researchers developing autonomous laboratories and advanced materials will benefit from more efficient use of costly characterization resources. The public may benefit through faster and more cost-effective research supporting innovation in manufacturing, energy, healthcare, and other materials-dependent technologies.

Subrecipient activities

The recipient does not intend to subaward funds.