Autonomous battery research: Principles of heuristic operando experimentation

Emily Lu, Gabriel Perez, Peter Baker, Daniel Irving, Santosh Kumar, Veronica Celorrio, Sylvia Britto, Thomas F. Headen, Miguel Gomez-Gonzalez, Connor Wright, Calum Green, Robert Scott Young, Oleg Kirichek, Ali Mortazavi, Sarah Day, Isabel Antony, Zoe Wright, Thomas Wood, Tim Snow, Jeyan Thiyagalinga · 2026

Read Autonomous battery research: Principles of heuristic operando experimentation by Emily Lu, Gabriel Perez, Peter Baker, Daniel Irving, Santosh Kumar, Veronica Celorrio, Sylvia Britto, Thomas F. Headen, Miguel Gomez-Gonzalez, Connor Wright, Calum Green, Robert Scott Young, Oleg Kirichek, Ali Mortazavi, Sarah Day, Isabel Antony, Zoe Wright, Thomas Wood, Tim Snow, Jeyan Thiyagalinga (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-00851.txt.

Introduction

Emily Lu and a twenty-four-author team propose letting an AI pilot run battery experiments at large research facilities. Understanding battery degradation is critical to the energy transition, but experiments that watch batteries while they work suffer from weak reliability, representativeness, and reproducibility. Bespoke hardware and pre-programmed routines miss fleeting failure events such as the first growth of a dendrite, a metal spike that can short a cell.

Using a multi-modal toolkit at the Rutherford Appleton Laboratory as a case study, the authors show conventional experiments systematically miss such transient phenomena. Their heuristic operando framework instead steers the beamline actively, guided by physics-based digital twins, to predict and capture rare events deliberately rather than by luck.

Catalog record

CREATOREmily Lu, Gabriel Perez, Peter Baker, Daniel Irving, Santosh Kumar, Veronica Celorrio, Sylvia Britto, Thomas F. Headen, Miguel Gomez-Gonzalez, Connor Wright, Calum Green, Robert Scott Young, Oleg Kirichek, Ali Mortazavi, Sarah Day, Isabel Antony, Zoe Wright, Thomas Wood, Tim Snow, Jeyan Thiyagalinga
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE113 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-00851

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Autonomous battery research: Principles of heuristic operando experimentation Emily Lu, Gabriel Perez, Peter Baker, Daniel Irving, Santosh Kumar, Veronica Celorrio, Sylvia Britto, Thomas F. Headen, Miguel Gomez-Gonzalez, Connor Wright, Calum Green, Robert Scott Young, Oleg Kirichek, Ali Mortazavi, Sarah Day, Isabel Antony, Zoe Wright, Thomas Wood, Tim Snow, Jeyan Thiyagalingam, Paul Quinn, Martin Owen Jones, William David, James Le Houx Abstract Unravelling the complex processes governing battery degradation is critical to the energy transition, yet the efficacy of operando characterisation is severely constrained by a lack of Reliability, Representativeness, and Reproducibility (the 3Rs). Current methods rely on bespoke hardware and passive, pre-programmed methodologies that are ill-equipped to capture stochastic failure events. Here, using the Rutherford Appleton Laboratory’s multi-modal toolkit as a case study, we expose the systemic inability of conventional experiments to capture transient phenomena like dendrite initiation. To address this, we propose Heuristic Operando experiments: a framework where an AI pilot leverages physics-based digital twins to actively steer the beamline to predict and deterministically capture these rare events. Distinct from uncertainty-driven active learning, this proactive search anticipates failure precursors, redefining experimental efficiency via an entropy-based metric that prioritises scientific insight per photon, neutron, or muon. By focusing measurements only on mechanistically decisive moments, this framework simultaneously mitigates beam damage and drastically reduces data redundancy. When integrated with FAIR data principles, this approach serves as a blueprint for the trusted autonomous battery laboratories of the future. Keywords: Autonomous Experimentation, Operando Characterisation, Battery degradation, Scientific AI, Physics-Informed Machine Learning Introduction Next-generation energy storage technologies, particularly batteries, are indispensable for the global transition to clean energy [ 22 ] . Realising their full potential, however, is hindered by a fundamental bottleneck: the need to understand and mitigate the complex physical and chemical processes, many of them transient and stochastic, that govern performance, safety, and lifecycle. Operando characterisation, which probes battery materials during live electrochemical cycling, has become the primary tool for unravelling these dynamics [ 59 ] . Nevertheless, this vital capability is severely constrained. A recent comment by Drnec and Lyonnard [ 15 ] formalised widespread community concerns, identifying systemic failures in Reliability, Representativeness, and Reproducibility (the 3Rs) as a grand challenge for the entire battery characterisation field. Resolving the 3R crisis extends far beyond the laboratory; it is the rate-limiting step in deploying the safe, next-generation storage technologies essential for a sustainable future. The urgency of the 3R challenge is compounded by concurrent shifts in experimental capabilities. Next-generation synchrotron and neutron sources, such as the Diamond-II upgrade [ 9 ] , are designed for significantly higher data acquisition rates. Without corresponding methodological advancements, this increased data throughput will inevitably exacerbate the ’data deluge’, overwhelming researchers with vast quantities of data that are difficult to store and analyse, while compounding existing difficulties in obtaining reproducible results [ 2 ] . Simultaneously, recent advances in predictive AI algorithms, the advent of exascale computing [ 38 ] and the increased accessibility of high-performance computing (HPC) provide the necessary tools for real-time processing of complex data streams. The conventional, brute-force approach of passive data collection is not only insensitive to transient events but often logistically impossible. As recent work by Corrao et al. [ 11 ] highlights, a full, multi-modal map of a single sample wafer could take 2 days by XRD but a staggering 4 months by XAFS. This unsustainable reality demands a methodological shift from passive data collection to an active, intelligent search. Together, these developments provide the technical motivation for fundamentally redesigning operando battery characterisation. In response to the 3R challenge, this paper first provides a critical review of the state-of-the-art, multi-modal operando electrochemical toolkit at the Rutherford Appleton Laboratory (RAL). This analysis, detailed in the Supplementary Information, confirms that the difficulties in achieving the 3Rs are often systemic and linked to the inherent limitations of current, highly specialised hardware. Notably, while specialised cells are highly representative of fundamental material electrochemistry (low Technology Readiness Level (TRL)), they often fail to represent the coupled thermal and mechanical constraints of commercial devices (high TRL). However, our analysis also reveals a deeper methodological problem extending beyond hardware standardisation. We argue that standardisation, while necessary, is an insufficient solution because the fundamental experimental approach itself is flawed. Conventional operando experiments typically rely on pre-programmed, fixed-cadence measurements, rendering them inherently passive. As studies on dynamic processes have shown, this creates a fundamental trade-off: fast scans suffer from poor signal-to-noise, while slow scans lack the temporal resolution to capture short-lived intermediates [ 51 ] . Consequently, this approach is blind to unpredictable, stochastic events—such as the nucleation of new phases, the onset of critical interfacial changes (e.g., SEI breakdown), or transient shifts in local ion dynamics—that frequently govern irreversible performance loss and degradation [ 17 ] . By definition, a pre-programmed scan is not synchronised with the unpredictable event; it is statistically likely to miss the critical onset of failure. While the burgeoning field of autonomous experimentation (AE) offers potential solutions [ 50 ] , current approaches predominantly focus on active learning strategies, often using Bayesian optimisation or Gaussian processes to efficiently map static parameter spaces [ 36 ] or optimise material properties [ 7 ] . These methods excel at reactive optimisation—learning from what has already happened. However, they are fundamentally ill-suited to the challenge addressed here: proactively predicting and intercepting rare, non-stationary, stochastic events governed by hidden chemo-mechanical precursors. A Multi-Modal, Multi-Scale Toolkit Complementary Probes for a Complex System Understanding battery performance and degradation requires probing complex, coupled processes that span multiple length and time scales, from atomic arrangements to electrode-level heterogeneities. Consequently, no single experimental probe can comprehensively capture this intricate behaviour. A multi-modal approach is therefore essential, leveraging the complementary strengths of techniques available at large-scale facilities [ 59 ] . X-ray methods provide high-resolution structural information—ranging from diffraction (XRD) for long-range crystal order to pair distribution function analysis (XPDF) for local, disordered structures. Spectroscopy techniques provide chemical information, such as average bulk oxidation states from X-ray Absorption Spectroscopy (XAS) and surface composition from X-ray Photoelectron Spectroscopy (XPS) [ 31 ] . Neutron techniques offer complementary contrast because, unlike X-rays, neutrons interact with the atomic nucleus [ 25 ] . This fundamental difference provides unique capabilities, including sensitivity to light elements (like Li and H), the ability to distinguish between isotopes (e.g., {}^{6}Li/^{7}Li ), and the ability to differentiate elements with similar electron configurations, such as neighboring transition metals [ 41 ] . Muon spectroscopy ( \mu SR) is used as a sensitive local probe capable of quantifying ion diffusion dynamics, offering atomic-level insights complementary to neutron and X-ray methods [ 4 ] . Electron microscopy (SEM/TEM) provides unparalleled spatial resolution for micro- and nanostructure, while laser-based vibrational spectroscopy (e.g., Raman, IR) and neutron molecular spectroscopy adds chemical sensitivity, probing molecular bonds and interfacial species. Integrating these diverse techniques is vital for a holistic understanding. However, this necessity inherently drives the requirement for a varied and complex suite of specialised operando electrochemical cells, presenting significant practical challenges. X-ray techniques Synchrotron facilities provide high-brilliance, tunable-energy X-ray beams essential for probing battery materials. The wide energy range, from soft X-rays (typically 0.1–2 keV) to high-energy hard X-rays (exceeding 100 keV), enables access to complementary structural

Read the complete text