The development of advanced materials has been a major driving force behind technological innovation and plays a critical role in addressing global societal challenges. Breakthroughs in materials science are needed to combat climate change by enabling greener batteries and more efficient catalysts that convert carbon captured from the atmosphere back into fuels, chemicals, and other value-added products. Despite significant progress in predicting new materials with tailored functional properties using advanced materials modeling and machine learning, the rate of discovery remains constrained by the often time-consuming and resource-intensive experimental realization and validation of new materials. These observations hold especially also for the validation of battery materials and electrocatalysts, where the challenge extends beyond identifying individual materials, requiring materials to operate synergistically in combination with other materials in an electrochemical cell.

In my talk, I will provide an overview of our efforts to establish and validate the two automated high-throughput experimental platforms Aurora [1] and Ophelia [2]. Our battery robot Aurora automates electrolyte formulation, electrode balancing, battery cell assembly, and electrochemical cycling, while Ophelia enables automated, parallel assessment of electrocatalysts for CO₂ reduction with real-time gas and liquid product monitoring via online chromatography. Both platforms are supported by extensive open-source software that streamline the definition of experimental protocols, monitoring of experiments, and data analysis. I will further discuss our linked data structure, which relies on semantically annotated metadata aligning with the battery and electrochemistry domains of the Elementary Multiperspective Ontology, establishing a new reference for reporting of scientific data compliant with the FAIR data principles [3,4]. I will conclude with an outlook on our vision for progressing from automated to fully autonomous materials research platforms.

[1] E. Svaluto-Ferro, G. Kimbell, Y. Kim, N. Plainpan, B. Kunz, L. Scholz, R. Läubli, M. Becker, D. Reber, R.-S. Kühnel, P. Kraus, C. Battaglia, Toward an autonomous robotic battery materials research platform powered by automated workflow and ontologized findable, accessible, interoperable, and reusable data management, Batteries & Supercaps 2025, 202500151

[2] A. Senocrate, F. Bernasconi, P. Kraus, N. Plainpan, J. Trafkowski, F. Tolle, T. Weber, U. Sauter, and C. Battagliia, Parallel experiments in electrochemical CO2 reduction enabled by standardized analytics, Nature Catalysis, 2024, 7, 742

[3] N. Plainpan, S. Clark, C. Battaglia, BattINFO converter: an automated tool for semantic annotation of battery cell metadata, Batteries & Supercaps, 2025, 202500151 (see also https://battinfoconverter.streamlit.app/)

[4] G. Kimbell, R.-S. Kühnel, C. Battaglia, submitted

The transition from laboratory discovery to commercial manufacturing remains a critical challenge in photovoltaic technology development. This talk presents our comprehensive vision for self-driving laboratories (SDLs) that not only accelerate development but fundamentally reshape how we approach photovoltaic innovation. We begin by demonstrating our recent achievements in roll-to-roll manufacturing of perovskite solar cells, where our SDL platform enabled the fabrication of modules with 11% efficiency under ambient conditions, processing over 11,800 devices in 24 hours. This unprecedented throughput, combined with our cost projections of ~0.7 USD/W, establishes a new benchmark for scalable photovoltaic manufacturing. However, the true potential of SDLs extends far beyond automation. We will discuss our perspective on the evolution of SDL paradigms - from current focused optimization approaches to future systems capable of autonomous hypothesis generation and testing. This includes our ongoing work towards hybrid perovskite solar cells (HPSCs) through cross-domain optimization, where SDLs simultaneously consider materials properties, device architectures, and manufacturing constraints. Our latest innovations in integrating biomass-derived materials and developing novel device structures demonstrate how SDLs can unlock previously unexplored pathways in photovoltaic development. We propose a roadmap for future SDL development that emphasizes the democratization of these technologies, integration of manufacturing concepts from the outset, and the potential for autonomous agents to accelerate discovery.The transition from laboratory discovery to commercial manufacturing remains a critical challenge in photovoltaic technology development. This talk presents our comprehensive vision for self-driving laboratories (SDLs) that not only accelerate development but fundamentally reshape how we approach photovoltaic innovation. We begin by demonstrating our recent achievements in roll-to-roll manufacturing of perovskite solar cells, where our SDL platform enabled the fabrication of modules with 11% efficiency under ambient conditions, processing over 11,800 devices in 24 hours. This unprecedented throughput, combined with our cost projections of ~0.7 USD/W, establishes a new benchmark for scalable photovoltaic manufacturing. However, the true potential of SDLs extends far beyond automation. We will discuss our perspective on the evolution of SDL paradigms - from current focused optimization approaches to future systems capable of autonomous hypothesis generation and testing. This includes our ongoing work towards hybrid perovskite solar cells (HPSCs) through cross-domain optimization, where SDLs simultaneously consider materials properties, device architectures, and manufacturing constraints. Our latest innovations in integrating biomass-derived materials and developing novel device structures demonstrate how SDLs can unlock previously unexplored pathways in photovoltaic development. We propose a roadmap for future SDL development that emphasizes the democratization of these technologies, integration of manufacturing concepts from the outset, and the potential for autonomous agents to accelerate discovery.

Functional materials for next-generation energy devices such as batteries, fuel cells, electrolyzers are fabricated by powder film formation; mixture, disperse, apply, and dry. As with powdered dishes, nano- and micro-scale complex phenomena in each process. Process Informatics is one solution to the requirement of high-throughput exploration of process parameters. This talk introduces a topic of autonomous exploration of drying process, minimizing the defect ratio of catalyst layer of proton exchange membrane fuel cells (PEMFCs). The first prototype in Fig. (a) demonstrated the optimization of the drying parameter minimizing the defect ratio of carbon catalyst layer. The system reached the optimization with only 40 trial-and-error Bayesian optimization among 40,000 candidates [1]. The scalable process exploration method, ROPES (Robotic Objective Process Exploration System) in Fig. (b) integrated the process unit; the die-coating, the four-step zone-heating, and the evaluation unit; the camera, the interferometer, and the electronic conductivity testing. The second prototype realized the high-speed automatic experiments and high-throughput autonomous exploration of drying process, which also contributes the scaleup to factory-size zone-heating process.

Keywords：Process Informatics; powder film formation; autonomous exploration; Scaleup

In modern electrocatalysis research, initial screening of catalyst performance is typically carried out in batch electrochemical cells such as rotating disk electrode (RDE) half-cells [1]. Although these setups are efficient and economical compared to full device testing, they are not suited for the rapid evaluation of large material libraries containing hundreds or thousands of samples. To overcome this limitation, various scanning electrochemical cells have been developed. When combined with high-throughput synthesis, characterization, quality control, data acquisition, and automated analysis, these platforms form the basis of autonomous electrocatalysis workflows and ultimately self-driving laboratories. One example is the scanning flow cell (SFC), where continuous electrolyte flow allows direct coupling to external analytical techniques such as inductively coupled plasma mass spectrometry (ICP-MS).

This talk will introduce the setups and present representative examples of how SFC-based systems can accelerate the evaluation of electrocatalysts across multiple stages of development. One example includes a grid-based search using an SFC coupled to ICP-MS to screen a NiFeCo material library for highly active and stable oxygen evolution reaction (OER) catalysts in neutral media [2]. Faster, more accessible workflows can also be achieved by employing purely electrochemical approaches, along with stability proxies and active learning algorithms [3]. At a more advanced stage, SFC configurations capable of testing gas diffusion electrodes (GDEs) allow catalyst layer screening under conditions approaching those in real devices [4].

Overall, this presentation highlights recent progress in high-throughput electrocatalyst testing, demonstrating how automated SFC-based approaches can significantly accelerate discovery and mechanistic insight [5]. Looking ahead, continued development in automation, data-driven decision making, and integration with advanced synthesis and characterization platforms is poised to enable fully autonomous electrocatalysis laboratories that shorten development cycles and expand the search space for next-generation catalyst materials.

High-throughput (HT) experimentation is increasingly shaping the way electrocatalysts are discovered and assessed, offering the ability to rapidly map composition–property relationships across vast chemical spaces. Yet, as in many areas of electrochemical energy research, practical limitations introduce an inherent tradeoff: fast activity screening enables broad sampling at scale, but often emphasizes throughput over detailed mechanistic insight. Conversely, deeper multimodal characterization can deliver deep mechanistic understanding but naturally limits experimental throughput and slows down knowledge acquisition.[1] Navigating this balance is especially important for modern, data-centric materials discovery, where robust structure–activity–stability correlations are essential for training reliable predictive machine-learning models.[2]

This talk will introduce an approach aimed at addressing the throughput–knowledge tradeoff by integrating Gaussian process regression (GPR) with comprehensive HT electrochemical and structural characterization. Using IrCoTi mixed-oxide thin-film libraries reactively sputtered at room temperature and 500 °C as a model platform, it will be demonstrated how synthesis conditions, crystallinity, and phase segregation can influence the acidic oxygen evolution reaction performance and durability. Special emphasis will be placed on operando dissolution analysis using a scanning flow cell coupled to inductively coupled plasma mass spectrometry (SFC-ICP-MS), which provides insight into how instability can shape observed activity trends and underscores the importance of including stability considerations for interpreting HT electrocatalyst activity screening outcomes.

Drawing on results from our recent work and related studies, the talk will highlight how coupling GPR with HT multimodal workflows can enhance data generation rates without sacrificing mechanistic depth.[3] This combined strategy provides a transferable blueprint for AI-guided catalyst discovery, offering a pathway toward the digitally accelerated development of energy-relevant materials and devices.

We have been developing computational software towards assisting in the discovery of organic-based materials with targeted structures and properties. Here, we will explore the different ways in which both artificial intelligence (AI) and molecular simulation can assist in the discovery and understanding of a variety of materials for energy devices. This will include exploration of generative AI for the discovery of ionic liquids[1,2], the use of molecular simulation and machine learning forcefields to predict the structure and ion conductivity of polymer membranes for redox flow batteries,[3,4] and the use of optimisation algorithms to accelerate the discovery of new electrode materials.

[1] “Deep learning-enabled discovery of low-melting-point ionic liquids”, G. Ren, A. Mroz, F. Philippi, T. Welton, K. E. Jelfs, ChemRxiv, 2025.

[2] “Expanding the chemical space of ionic liquids using conditional variational autoencoders”, G. Ren, A. Mroz, F. Philippi, T. Welton, K.E. Jelfs, ChemRxiv, 2025.

[3] “Sulfonated poly(ether-ether-ketone) membranes with intrinsic microporosity enable efficient redox flow batteries for energy storage”, T. Wong, Y. Yang, R. Tan, A. Wang, Z. Zhou, Z. Yuan, J. Li, D. Liu, A. Alvarez-Fernandez, C. Ye, M. Sankey, D. Ainsworth, S. Guldin, F. Foglia, N. B. McKeown, K. E. Jelfs,* X. Li,* Q. Song,* Joule (2025), 9 (2), 101795.

[4] “Selective ion transport through hydrated micropores in polymer membranes”, A. Wang, C. Breakwell, F. Foglia, R. Tan, L. Lovell, X. Wei, T. Wong, N. Meng, H. Li, A. Seel, M. Sarter, K. Smith, A. Alvarez‐Fernandez, M. Furedi, S. Guldin, M. M. Britton, N. B. McKeown, K. E. Jelfs & Q. Song, Nature (2024), 635, 353.

Design-Make-Test-Learn (DMTL) is the cycle through which materials are designed, synthesized, tested and evolve into the next iteration.  This process is typically used in early drug discovery and relies primarily on synthesis and screening workflows of hundreds of thousands of drug candidates which take from months to year or longer. When chemists / scientists are empowered by automation and digitalization of the workflows, it significantly accelerates and standardizes the discovery processes. Furthermore, autonomous labs in combination with AI and cloud computing propel innovation, evolutionary learning and high quality open, AI ready FAIR data.  This is not limited to pharmaceutical research anymore and extends into energy research such as electrochemical storage and conversion technologies owing to society’s ambitious commitment to transition towards climate-neutral economy.

We, Chemspeed part of Bruker Corporation, develop and manufacture vender agnostic solutions for R&D and QC workflows automation.

For power-to-x/clean tech, we provide modularly by design end-to-end solutions from synthesis, formulation, application, curing, analysis to testing. In this talk, I will present selected snapshot use cases:

1. End-to-end electrocatalysts development for green hydrogen production
2. DMTL of photocatalysts
3. End-to-end automated, digitalized solutions for photovoltaics
4. High throughput and high-quality coin cell assembly with a throughput of 20 cells per hour
5. Electrode paste/slurry formulations and screening with benchtop Rheometer, PSD and more analytical devices
6. Solid phase synthesis of electrode materials and online validation with XRD, XRF
7. Screening of catalysts for CO2 conversion to value added chemicals like methanol

Autonomous experiments that integrate machine learning and robotics are reshaping materials research. By automating experimental workflows and efficiently searching high-dimensional parameter spaces, these approaches markedly accelerate materials discovery and process optimization.

Here, we report a modular self-driving laboratory (SDL) for solids and thin films [1–4]. The SDL orchestrates all stages of the experimental cycle—including sample transfer, synthesis, characterization, and iterative optimization. Data acquisition spans X-ray diffraction, scanning electron microscopy, Raman spectroscopy, and optical transmittance measurements. A Bayesian optimization enables autonomous exploration of the parameter space and rapid identification of optimal conditions.

We demonstrate the platform by synthesizing thin films of TiO₂ and LiCoO₂. We further show that the same workflow supports the discovery of new ionic conductors. These results highlight the potential of autonomous experimentation to accelerate research in solid-state materials. Ongoing efforts extend the SDL to bulk-materials synthesis, aiming to unify thin-film and bulk workflows within a single autonomous framework.

Practical electrocatalyst deployment requires balancing performance across multiple, often conflicting objectives. We demonstrate this for the acidic oxygen reduction reaction (ORR) where tradeoffs between activity, stability, and material cost must be considered. To navigate this, we employ a multi-objective Bayesian optimization framework that utilizes the continuous design space of high-entropy alloys (HEAs). The activity and stability are assessed through our existing machine-learning driven models for catalytic activity[1] and electrochemical dissolution[2]. In simulating the activity, we will present a machine-learning model for adsorption energy inference on alloys finetuned on an extensive density functional theory HEA dataset that covers 12 elements and 9 adsorbates. Within our framework, we uncover a novel activity-stability-cost Pareto front for ORR for the Ag-Au-Cu-Ir-Pd-Pt-Rh-Ru composition space. We find that alloying expands the hypervolume spanned by the Pareto front, and that the front is constituted of low-entropic alloys consisting primarily of Ag, Au, Cu, Pd, and Pt. Moreover, we present a new approach to analyzing these optimal trade-offs by investigating the hypervolume degradation upon removing critical elements (Au, Pd, and Pt), revealing their individual roles in the Pareto front. This work highlights the necessity of considering all relevant objectives in catalyst optimization and the advantage of HEAs as a platform for multi-objective catalyst discovery.

Autonomous systems have achieved remarkable progress in domains such as self-driving vehicles, game-playing computers, and robotics, largely powered by reinforcement learning. However, their potential in electrochemical systems remains largely unexplored. In electrochemistry, pulsed operation often enhances conversion rates or power output, suggesting an opportunity for adaptive control strategies. This work presents the complete development process of a reinforcement learning-based AI controller designed to maximize the conversion rate of the glycerol oxidation reaction.[1] The process begins with a simple microkinetic model to design the initial controller architecture, followed by deployment and iterative refinement in real laboratory conditions. To accelerate algorithm refinement and reduce laboratory time, data collected during laboratory testing was used to construct a recurrent neural network-based digital twin. This virtual model enables rapid hypothesis testing through simulation of the electrochemical system, facilitating the evaluation of different improved AI controllers. Finally, the controller’s decision-making process is analysed and visualized by performing a time-series analysis using Markov chain modelling. Together, this work showcases a comprehensive AI-based approach that combines reinforcement learning, digital twinning, and time-series analysis to perform, simulate and analyse complex electrochemical experiments.

AI-driven scientific innovation is often blocked by scarce, fragmented, and biased data. To solve this, a global infrastructure of cloud-connected, autonomous laboratories has been proposed, generating high-quality, reproducible datasets via robotic experiments. This vision is structured into five hierarchical levels: G1 (Process Automation), G2 (Theory-Experiment Iterative Loop), G3 (Large Model-Driven), G4 (Multi-Platform, Multi-Task), G5 (Autonomous Scientific Discovery).
Our G4-level large-model-driven autonomous platform coordinates domain-specific small AI models and robotic experiments, enabling intelligent scheduling and real-time fine-tuning of pre-trained models based on experimental data, thereby creating a dynamic, synergistic feedback loop. Major breakthroughs in novel material creation (e.g., catalysts, polymers, COFs, and proteins) have compressed discovery cycles from over a century to mere months.
The global infrastructure transforms isolated scientific efforts into a collaborative and efficient exploration engine. It democratizes access to high-quality data, breaks down geographical and institutional barriers to innovation, ultimately accelerates industrial-scale scientific and technological advancements.

Many industrial chemicals are currently derived from non-renewable petroleum feedstocks, but electrochemical conversion of renewable biomass waste offers a more sustainable alternative. We have developed a high-throughput, automated electrolyser system designed to optimize the electrocatalytic production of lactic acid from glycerol at the anode, while simultaneously generating clean hydrogen at the cathode. The custom-built system can autonomously run up to 16 reactions, systematically varying key parameters of temperature, applied current, electrolyte flow rate, and concentration. Reaction products are analysed using a 96-vial HPLC autosampler, integrated with a data analysis program to quantify product yields. These results are fed into a Bayesian optimization algorithm (PHYSBO), which employs a Gaussian process to predict and select the next set of experimental conditions.¹ The chosen conditions are automatically updated into the experimental run program, initiating the next round of testing. This self-driving electrocatalysis platform tailors reaction conditions to each catalyst, ensuring optimal performance and a foundation for material comparison.

Controlling the solid-liquid interface and understanding catalyst-electrolyte interactions are central challenges for the rational design of electrocatalysts, particularly for the nitrate reduction reaction (NO₃RR) to ammonia under realistic wastewater conditions where diverse coexisting ions strongly impact activity and stability. In this work, a decentralized self‑driving laboratory (SDL) is established between the University of Toronto and ICFO to autonomously explore the combined space of catalyst and electrolyte compositions for NO₃RR. The platform integrates in situ catalyst fabrication with programmable compositional gradients, high‑throughput electrochemical screening in electrolytes of controlled multi‑ion composition, and automated analysis of nitrate‑to‑ammonia performance. A Bayesian optimization engine closes the loop by iteratively proposing new catalyst formulations conditioned on the specific electrolyte matrix, enabling the system to self‑optimize catalyst composition for a given wastewater‑relevant environment. This approach provides a general framework to disentangle and exploit catalyst-electrolyte interactions, enabling accelerated optimization of both catalyst composition and electrolyte formulation, and supporting decentralized strategies for electrocatalyst development that can be extended to other electrolyte‑engineered systems.

Organic crystalline materials are potential candidates for photocatalytic overall water splitting (OWS) which is of significant industrial importance for the production of green hydrogen for use as a fuel or chemical feedstock.^1–3  Extended metal-free crystalline materials such as covalent organic frameworks and graphitic carbon nitrides have been heavily investigated for OWS, where the latter, denser materials exhibit significantly higher hydrogen evolution rates than more porous COFs.^1,4 While organic crystals of single molecules have been heavily investigated for organic electronic applications such as OLEDs and solar cells, there has been comparatively much less research into OWS in these materials.^5,6 Organic molecular crystals are a promising platform due to their chemical diversity and large range of crystal structures and densities.³ The key question for the design of OWS photocatalysts is to what extent do chemical groups, extended vs molecular structures, crystal packing and density affect the electronic properties? Optical absorption, charge-transport properties, ionisation energy/electron affinity and dispersibility in water are key considerations and are influenced both by molecular properties and crystal structure, making computational modelling challenging.^7–9 In this work we shed light on these structure-property relationships by firstly investigating a series of widely known organic electronic materials including molecular crystals and COFs which have published crystal structures, using periodic Density Functional Theory (with the HSE06 functional). Here we analyse the effects of the above structural features on the calculated electronic properties such as band gap, band alignments, optical absorption and charge transport. We then devise a series of both molecular and bulk property descriptors and computationally screen organic materials from crystallographic databases based on these descriptors to highlight new potential OWS candidates.

Rapid discovery of performant electrocatalysts is vital for a carbon-neutral economy. In drug development and structural biology, acceleration in discovery¹ has been driven by machine-learning (ML) models trained on vast, standardised experimental datasets that map structure to function.² Although computational screening of electrocatalysts and other electrochemical technologies has advanced similarly,³ there is still no experimental database of comparable scale or quality to validate predictions and power next-generation generative models. The obstacle is the disordered, nanometre-scale nature of electrocatalyst interfaces and the fact that many interfaces form during operation, meaning physical parameters which are predictive of performance must be measured in-situ. In oxygen evolution reaction (OER) performance descriptors are usually extracted only from idealised single-crystal systems that do not represent deployable materials.⁴ Of particular interest are binding energies – a descriptor for OER catalyst performance and the strength of adsorbate interactions which produce a coverage dependent perturbation to these values. Previously, we have shown that these values can be extracted by accurate measurements of minute changes in optical absorption that arise as a result of the generation of surface bound OER intermediates.^5–8 In this talk, I will present HiSpEC—the first high-throughput operando optical spectroelectrochemistry platform capable of extracting binding energy and interaction strengths at scale. HiSpEC pairs modular, hierarchical experiment orchestration with a physics-constrained, unsupervised ML workflow, enabling stable, automated measurements, with a throughput of roughly 100 compositions per day. The scale of the data, coupled to improved sensitivity and robust automated analysis reveals detailed trends in materials properties that were not previously observable, such as changes in binding energy and interactions driven by surface disorder. This dataset closes the experimental gap, allows rigorous testing of computational theories, and lays the foundation for generative ML models that can design and optimise next-generation energy materials at unprecedented speed.

(1) Jayatunga, M. K., et. al. Drug Discov. Today 2024, 29 (6), 104009.

(2) Jumper, J., et. al. Nature 2021, 596 (7873), 583–589.

(3) Horton, M. K., et. al Nat. Mater. 2025, 1–11.

(4) Kuo, D.Y., et. al. J. Am. Chem. Soc. 2018, 140 (50), 17597–17605.

(5) Moss, B.; et. al J. Am. Chem. Soc. 2024, 146 (13), 8915–8927.

(6) Liang, C., et. al., J. Am. Chem. Soc. 2024, 146 (13), 8928–8938.

(7) Liang, C.; et. al., Nat. Catal. 2024, 7 (7), 763–775.

(8) Moss, B., et. al. Nat. Rev. Met. Primers (In press)

To move towards wide-spread adoption of renewable energy sources we need sufficient energy storage technologies. To meet growing demand and continue to diversify supply chains, it is necessary not only to optimise existing battery technologies but also explore next-generation battery materials. However, it is not typically possible to test a wide variety of materials and chemistries simultaneously. Materials synthesis is traditionally a time-intensive process, particularly with respect to optimisation of novel materials. Additionally, coin cell assembly and subsequent electrochemical data suffer due to human error and inconsistencies during cell assembly, often resulting in large variation between datasets.

The DIGIBAT facility at Imperial College London provides tools to automate both material synthesis and cell assembly, removing these roadblocks. Automation allows for continuous experimentation done in a reliably reproducible way, enabling the generation of large, consistent, datasets. Thereby, the research process is accelerated while simultaneously improving results.

This presentation will compare initial results from manual and automated coin cell assembly, probing how true these statements about automation are, discussing both the advantages and any limitations encountered so far. Additionally, the automated synthesis of hard carbon (as the anode for sodium ion batteries) will be explored, with both the advantages and challenges when creating an automated workflow being discussed.

Hard carbons are currently the most promising candidates for anode materials in sodium-ion batteries (SIBs). [1] The disordered structure of hard carbons featuring high interlayer spacing, defects and pores provides suitable sites for sodium ion storage. The relationship between the structural features of glucose derived hard carbons and their electrochemical behaviour was previously reported. [2,3] The interplay of these features suggests there may be an optimal hard carbon microstructure which may result in improved electrochemical performance. Therefore, strategies for improving the electrochemical performance of hard carbons include tuning their intrinsic properties via changing the synthetic conditions. [4] In the study of glucose derived hard carbons the synthetic method consists of initial hydrothermal carbonisation of glucose followed by high temperature pyrolysis. Previously, only the pyrolysis temperature was varied in the synthetic process. This showed for example that increasing the pyrolysis temperature results in greater pore size which is beneficial for sodium storage but decreases the defect concentration and interlayer spacing which is detrimental to sodium storage. [2] Exploring the large space of remaining possible experimental conditions in this synthetic process is investigated using a data-driven approach with the goal of increasing the efficiency at which optimal conditions are found. A closed loop Bayesian optimisation workflow for investigating synthetic conditions to obtain hard carbons with improved electrochemical performance is being developed. The workflow consists of training a surrogate model on a dataset developed from archived experimental data to make predictions on new synthetic conditions. Based on these predictions the next experimental condition is selected for investigation. The results from the synthesis and characterisation at the suggested conditions are then collected and processed in an automated way to update the dataset and the surrogate model. A custom toolkit for the automated analysis and extraction of important features from both the electrochemical and microstructure characterisation is implemented. Furthermore, the use of a research data management system in the developed process allows for the effective integration of data from the experimental workflow with the computational workflow. The developed process provides a more efficient and data-driven approach to experimental work in the field of hard carbon materials for SIBs.

The focus of our work is the development of computational tools to accelerate the design of organic electrolytes for aqueous organic redox flow batteries (AORFBs). In our paper, we present the construction of a computational dataset of organic redox-active molecules for aqueous electrolytes. This dataset underpins the development of machine learning (ML) models for property prediction, with a longer-term vision of leveraging Generative AI to propose novel electrolyte candidates and accelerate sustainable materials discovery.

The growing demand for renewable energy emphasises the need for efficient and scalable energy storage. AORFBs are promising systems due to their sustainability and tunability. However, identifying suitable organic electrolytes remains a major challenge. [1] Molecules must satisfy strict criteria including appropriate redox potentials, solubility and chemical stability. Current discovery approaches rely heavily on experimental screening, which is both time-consuming and resource intensive.

This study aims to advance electrolyte discovery for AORFBs by integrating high-throughput computational methods. Specific goals include: (i) identifying molecular properties critical for electrolyte performance, (ii) benchmarking computational approaches to guide design and (iii) applying ML to accelerate computational screening and propose novel candidates.

We implemented a virtual screening workflow to evaluate existing ML models for key electrolyte properties, such as redox potential and solubility. A custom RDKit-based enumeration tool was developed to generate combinatorial derivative libraries of organic backbones through user-defined substitutions, generating over 10,000 unique molecules. Several state-of-the-art ML models were benchmarked, including MolGAT for redox potential, AqSolPred for solubility, and GASA for synthetic accessibility. [2-4] The enumeration tool enabled flexible, high-throughput exploration of structural modifications, while the ML benchmarking revealed both the strengths and limitations of current predictive models for redox-active systems.

We employed the enumeration tool with an automated DFT workflow to model the redox reactions of 7,500 organic redox-active molecules, yielding a consistent dataset of redox potentials and solvation free energies across diverse organic scaffolds. We take this dataset to train a ML model to predict redox potentials with DFT-level accuracy at a significantly lower computational cost. Ongoing investigations explore the use of transfer learning strategies for cross-domain redox potential predictions using graph neural networks (GNNs).

Overall, this study establishes a computational framework for aqueous organic electrolyte discovery, integrating molecular enumeration, automated quantum chemical calculations and AI. The resulting dataset provides a robust foundation for developing advanced AI models to support sustainable materials discovery. Future work will couple these predictive models with generative AI approaches to design novel electrolyte candidates.

Aqueous batteries provide a safe, low-cost and sustainable option for large-scale energy storage, but their performance is often limited by the instability of the electrode–electrolyte interphase (EEI). The EEI plays a central role in regulating ion transport, nucleation, parasitic reactions and long-term cycling behaviour, yet its structure and evolution depend on a complex interplay of factors – electrolyte composition, additive chemistry, temperature, current density and cycling conditions. This creates a large, nonlinear design space that is difficult to probe through traditional, manually performed experiments, which are typically slow, path-dependent and hard to reproduce at scale. Self-driving laboratory (SDL) architectures offer a way to overcome these challenges by combining automated experimentation, standardized measurement workflows and machine-learning-guided decision making. Within the Inorganic SDL at the Acceleration Consortium, automated electrochemical testing, high-throughput characterization and active-learning experiment planners are integrated to systematically explore this landscape. The current platform uses aqueous Zn batteries as a model system, enabling closed-loop studies of electrolyte formulations and operating conditions supported by rapid measurements of Coulombic efficiency, impedance and in-situ optical microscopy of interfacial evolution. These datasets are further enriched by SEM and XPS analyses, linking additive chemistry to interfacial morphology and surface composition.

By uniting statistically robust, reproducible experimentation with adaptive machine-learning strategies, this SDL approach is positioned to provide quantitative insight into the structure–function relationships that govern EEI formation, evolution and degradation. Such mechanistic insight accelerates the development of interphases for next-generation aqueous batteries and establishes a scalable paradigm for autonomous discovery in electrochemical materials research.

Porous battery electrodes exhibit complex, multiscale architectures that strongly influence rate capability, degradation pathways, and manufacturability. Yet the mapping from processing parameters to emergent microstructure remains notoriously difficult to predict with physics-based models alone. This talk introduces a suite of generative AI tools developed at Imperial College London that address this challenge by learning microstructural distributions directly from data and enabling rapid exploration of manufacturable design spaces. Building on techniques such as GAN-based dimensionality expansion for synthesising statistically representative 3D volumes from single 2D micrographs, we demonstrate how data-driven generative models can overcome bottlenecks inherent to conventional image-based simulation workflows.

A central part of the talk examines the generation, analysis, and optimisation of microstructures conditioned on realistic processing parameters such as porosity, active-material fraction, and binder behaviour. By embedding these generative models within Bayesian optimisation loops, we show how large numbers of statistically meaningful microstructures can be produced and evaluated efficiently. This enables the identification of high-performance manufacturing conditions without requiring either full 3D imaging campaigns or computationally intensive virtual fabrication pipelines. While generative models can, in principle, hallucinate physically implausible features, we highlight how incorporating tortuosity depth-profiles, phase topology constraints, and microstructural representativeness tests mitigates this risk and yields structures consistent with experimental XCT trends.

The talk then focuses on quantifying tortuosity factor as a function of electrode depth in graded electrodes. Measuring tortuosity as a depth-resolved field offers a unifying perspective that reconciles XCT-derived tortuosity with impedance measurements from symmetric cells containing blocking electrolytes. We show how this depth-dependent tortuosity can parameterise Doyle–Fuller–Newman models more faithfully for graded electrodes, avoiding oversimplified scalar tortuosity assumptions that misrepresent transport limitations in layered architectures.

To close, we present how our spin-out company, Polaron, is deploying these tools to engineers worldwide for microstructural characterisation, optimisation, and design. We also discuss work from the FULL-MAP project on AI agents for autonomous laboratories, including recent results interrogating how large language models encode chemically structured knowledge such as the periodic table, revealing geometric and layered representations that raise both opportunities and cautions for scientific use. 

This combination of generative modelling, rigorous microstructural physics, and autonomous experimentation sketches a path toward rapid, data-driven materials optimisation pipelines for next-generation electrochemical technologies.

In this talk, I will present our recent work on medium-throughput electrocatalyst discovery enabled by machine-learning tools such as Bayesian Optimization and Gaussian Process Regression (GPR). High-entropy materials provide a versatile platform for identifying optimal performance across both single-objective targets - such as catalytic activity - and multi-objective challenges involving activity–stability trade-offs. Our group employs a range of synthesis strategies, including incipient wetness impregnation and electrodeposition [1], to generate diverse catalyst libraries. These materials are evaluated with respect to several key electrochemical reactions, including the oxygen reduction (ORR) [2], oxygen evolution (OER) [3], and glucose oxidation reactions.

In our newest workflow we apply multi-objective optimization, where we use Pareto analyses to navigate competing performance metrics. By constructing Pareto fronts from GPR models of experimentally measured data, we identify catalyst compositions that represent optimal compromises rather than absolute maxima in any single metric. This provides a rigorous framework for quantifying trade-offs, guiding subsequent exploration, and ultimately accelerating the convergence toward practically relevant electrocatalysts. I will discuss the statistical considerations behind these approaches and highlight recent case studies demonstrating how data-driven strategies can meaningfully enhance digital discovery pipelines in electrocatalysis.

Self-Driving Laboratories (SDLs), which realize autonomous materials exploration through the integration of black-box optimization (BBO) methods (leveraging machine learning/artificial intelligence) and robotic experimental systems, propose and execute experiments under promising, yet unexplored, conditions without human intervention.

To easily facilitate this integration, we developed NIMO, an open-source middleware. NIMO treats each experimental system and BBO method as interchangeable modules, enabling the flexible realization of diverse SDLs through arbitrary combinations. NIMO standardly implements multiple advanced BBO algorithms, including Bayesian optimization (BO) variants, phase diagram construction, and objective-free search. Furthermore, its modular design significantly streamlines the implementation of new algorithms, allowing newly developed exploration strategies to be rapidly adopted in robotic experiments.

This talk introduces the NIMO framework, showcases its modular architecture, and presents concrete examples of self-driving laboratories realized using NIMO. We will also demonstrate successful application results in materials exploration, highlighting NIMO's capability to accelerate autonomous discovery. NIMO is available at https://github.com/NIMS-DA/nimo.