This talk will outline how AI models, when trained on large datasets of experimental electrochemical results, can address the limitations of traditional approaches in autonomous labs, and discuss the potential of this methodology to overcome current limitations in scaling up autonomous laboratories for broader energy technology applications.
Abstract
The development of self-driving laboratories in electrochemistry represents a transformative shift in how energy storage materials are discovered and optimized. While conventional AI models in autonomous labs have largely relied on computational data, significant challenges remain in bridging the gap between simulation-driven predictions and real-world experimental outcomes. This talk will explore a novel approach: leveraging AI models trained on experimental data to improve the accuracy and reliability of autonomous laboratory systems, with a particular focus on energy storage electrochemical materials.
Conventional AI models often rely heavily on computational simulations, which may not fully capture the complexities and nuances of experimental data. In contrast, models trained directly on high-throughput experimental results can better account for electrochemical behavior under real-world conditions, improving predictive power and guiding more effective material discovery processes. By integrating experimental data from diverse electrochemical analysis techniques, such as cyclic voltammetry, impedance spectroscopy, and battery cycling tests, AI can identify key trends, optimize experimental workflows, and accelerate the discovery of novel materials for energy storage applications.
This talk will outline how AI models, when trained on large datasets of experimental electrochemical results, can address the limitations of traditional approaches in autonomous labs. Through case studies, we will demonstrate how this data-driven approach can enhance the efficiency of material screening, reduce the time and resources needed for experimentation, and provide actionable insights into optimizing energy storage electrochemical materials. Furthermore, we will discuss the potential of this methodology to overcome current limitations in scaling up autonomous laboratories for broader energy technology applications.
By combining the power of AI with high-throughput experimental data, this approach holds the potential to revolutionize the discovery and optimization of electrochemical materials for energy storage, driving forward a more sustainable and energy-efficient future.
Lithium-ion batteries are among the most complex energy storage technologies, with performance driven by intricate interactions across diverse materials and operating conditions. Despite advances in modeling, predicting battery behavior from individual components remains a challenge, making physical experimentation essential. To accelerate discovery, Wildcat has developed a high-throughput research platform that generates high-quality, reproducible, and cross-comparable data; more importantly, including both successful and unsuccessful results. This comprehensive dataset provides a strong foundation for predictive modeling. Building on this, we are integrating artificial intelligence to uncover patterns, optimize designs, and enhance research efficiency. In this presentation, we will share Wildcat’s data-driven approach to battery innovation, highlight the capabilities of our high-throughput system, and discuss key insights from deploying machine learning models to transform Lithium-ion battery research.
Electrochemical energy conversion and storage sit at a decisive inflection point: urgency of breakthrough is high, yet the science remains dominated by multiscale physics, interfacial complexity, sparse high-fidelity data, and long validation cycles for safety and lifetime. I have a pragmatic thesis: AI does not “replace” electrochemistry; it accelerates it when models are anchored to mechanistic constraints, uncertainty is taken care of, and workflows are designed to close loops between simulation, experiment, and optimization. I will discuss a disciplined operating model for scientific machine learning in batteries: defining the decision that must be improved (screening and discovery), choosing representations that encode physics, and quantify where the model can and cannot be trusted under distribution shift. We want to compress the “electrons-to-devices” pipeline by replacing repeated expensive computations and measurements with calibrated surrogates, while using active learning and multi-fidelity strategies to allocate scarce high-quality labels where they matter most.
A core effort addresses electronic-structure bottlenecks that propagate into materials screening, interfacial chemistry, and kinetics. We introduce equivariant graph neural network frameworks that predict electron density across molecules, liquids, and solids at speeds orders of magnitude beyond density functional theory (DFT). To reduce data load and improve practicality, we shift the learning target from dense real-space grids to sparse, SE(3)-equivariant density matrices, reaching high accuracy with far smaller datasets while enabling uncertainty diagnostics tied to charge error and self-consistency. We have extended compact electronic representations further by learning “floating orbital” placements and coefficients through an equivariant Cartesian tensor network with controlled symmetry breaking; beyond accuracy, these models deliver immediate computational value by reducing self-consistent-field iterations when used to initialize DFT. Scaling upward, we develop and operationalize uncertainty-aware machine-learned interatomic potentials (MLIPs), including general-purpose foundation models for stable molecular dynamics across diverse chemistries, alongside evidence-based limits of universality in energy materials. A case study in polyanion sodium cathodes demonstrates that out-of-distribution performance often requires system-specific datasets and charge-aware architectures; fine-tuning universal MLIPs can recover much of the deficit, but domain-specialized models can remain faster and more accurate. To make specialization economical, we have done autonomous batch active learning, integrating uncertainty, batch selection, and efficient force/stress evaluation, to reduce labeling cost and expert intervention. We also advance data efficiency at the quantum-chemistry level through minimal multilevel machine learning which optimally allocates training across fidelity hierarchies to minimize combined prediction error and acquisition wall-time. Finally, to break the equilibrium bias that has limited reactive modeling, we curate pathway-rich datasets and show that equivariant models trained on such data can serve as practical surrogate potentials for reaction-path search, enabling kinetics at scales previously infeasible; these concepts are demonstrated in electrochemical interface kinetics and in long-timescale cathode dynamics where nanosecond molecular dynamics becomes accessible while retaining DFT-level fidelity.
Our effort moves from forward prediction to inverse design and autonomy, where the true payoff is not a faster model, but a faster, safer, more reliable decision. For battery interphases such as the solid-electrolyte interphase (SEI), we outline a blueprint for inverse design that couples multi-scale, multi-fidelity simulations with operando characterization, high-throughput synthesis/testing, and semi-supervised generative learning. The essential ingredient is uncertainty tracking across all layers, experimental, computational, and generative, so fidelity can be improved deliberately, and “failed” experiments can be used to reduce bias and expand coverage. More broadly, we develop and deploy generative methods for molecules and materials, diffusion models for organometallic catalysts and crystalline structures, symmetry-consistent generation for disordered crystals, and data-free or amortized global optimization—paired with hierarchical screening (“materials funnels”) that combine generative proposal, fast learned surrogates, MLIPs, and final DFT validation. Because acceleration without trust is fragile, we place calibrated uncertainty at the center: ensemble-based probabilistic predictors unify aleatoric and epistemic components and are recalibrated on unseen data; early-aging models forecast full degradation trajectories from limited battery cycling with uncertainty and explainability sufficient to support uncertainty-guided truncation of long tests; and sensitivity analysis of physics-based degradation models via differentiable surrogates identifies the parameters that truly govern capacity fade. We take an integrated view of distributed, intention-agnostic materials acceleration platforms that orchestrate asynchronous experiments and simulations across institutions. The overarching message is operational: when physically grounded representations, calibrated uncertainty, data-efficient multi-fidelity learning, and closed-loop autonomy are engineered as a single system, electrochemical science can reduce time-to-insight and time-to-validation, maintaining interpretability and reproducibility.
The development of a self-driving laboratory (SDL) is driving electrocatalysis research from traditional trial-and-error approaches toward automation, high throughput, and intelligence. As an autonomous experimental system tailored to electrochemical scenarios, the AI-eChemist Laboratory integrates front-end intelligent decision-making, automated high-throughput experimentation, multimodal characterization, and data-driven analysis, providing a new paradigm for the discovery, mechanistic understanding, and application validation of complex electrocatalytic materials. This review first summarizes recent advances in SDL from two perspectives: front-end intelligence and autonomous experimental platforms. On this basis, we further focus on three key technical routes established in AI-eChemist: high-throughput synthesis and screening of model catalysts, high-throughput synthesis and screening of practical powder catalysts, and emerging screening strategies targeting intrinsic catalytic activity. These routes promote the construction of a closed-loop research system in AI-eChemist, spanning materials screening and mechanistic investigation to device validation, through standardized data acquisition, practical materials discovery, and intrinsic activity evaluation. Finally, in view of the demands of AI-eChemist for practical applications and autonomous development, we discuss future directions including multimodal characterization, automated function islands, scalable fabrication, and multi-agent collaboration, aiming to provide systematic insights for the intelligent discovery and application-oriented translation of advanced energy materials.
Y. Tan, Lunbo Chen, Xiangyi Shan et al.· AI Agent· 0 citations
The Energy Storage Research Alliance (ESRA, https://energystoragera.org/) aims at tackling fundamental science challenges in beyond-Li-ion energy storage systems. Underlying the scientific thrusts are the crosscutting efforts which aim at developing cutting-edge capabilities. In particular, the Materials Acceleration Platform (MAP) crosscut leverages and contributes towards the advances in artificial intelligence/machine learning (AI/ML) and automation to accelerate the discovery, synthesis, and characterization of new and existing energy storage materials, especially for Na and Zn. Under the MAP crosscut, we use AI to enable and accelerate physics-constrained information extraction from characterization, combine computational discovery and design with autonomous synthesis with closed-loop feedback, create autonomous electrochemical characterization laboratories, and make possible large scale predictions of dynamic systems with ML interatomic potentials (MLIPs). We will discuss efforts in ESRA-MAP to allow extraction of new insights from microscopy and spectroscopy data, to accelerate the design and synthesis of new solid state ion conductors, and to simulate complex reactions at interfaces.
This work is funded by the Energy Storage Research Alliance "ESRA" (DE-AC02-06CH11357), an Energy Innovation Hub funded by the U.S. Department of Energy, Office of Science, Basic Energy Sciences.
Maria K. Y. Chan· ECS Meeting Abstracts· 0 citations
The rapid development of next-generation energy storage technologies increasingly depends on the ability to efficiently explore high-dimensional materials and formulation spaces. Automation combined with data-driven optimization provides a powerful pathway to accelerate discovery while reducing experimental cost and human bias. This presentation highlights our recent advances in self-driving laboratory frameworks for battery electrolyte engineering, with a focus on Bayesian optimization (BO) integrated with an automated coin-cell assembly and electrochemical testing platform (ODACell 2) across both lithium- and zinc-based aqueous battery chemistries.
In lithium-based systems, we demonstrate a closed-loop experimental workflow that integrates automated coin-cell assembly, galvanostatic cycling of LiFePO
4
|| Li
4
Ti
5
O
12
organic–aqueous full cells, and BO-driven experiment selection. This approach addresses a longstanding disconnect between highly automated cell assembly and data-driven decision-making in battery research. By navigating a multi-component electrolyte design space comprising four organic co-solvents and two lithium salts, the framework rapidly identified formulations achieving ≥ 94% Coulombic efficiency. Coupling the autonomous workflow with online electrochemical mass spectrometry further enabled mechanistic insight, revealing that optimized co-solvent combinations effectively suppress parasitic hydrogen evolution.
Complementary advances are presented for aqueous zinc batteries. Coulombic efficiency was optimized in Cu || Zn cells across a five-dimensional electrolyte space comprising ZnCl
2
, ZnSO
4
, and multiple functional additives. The data-efficient optimization revealed strong salt-dependent behavior, including pronounced sensitivity of ZnCl
2
electrolytes to concentration, stabilizing effects of ZnSO
4
-based systems, and nuanced additive–salt interactions. While BO proved highly effective for rapid screening and performance mapping, limitations in model interpretability near design-space boundaries underscore the need for careful experimental design and validation strategies.
Together, these studies establish Bayesian optimization as a practical, scalable tool for electrolyte discovery when tightly integrated with laboratory automation. By spanning distinct battery chemistries and experimental objectives, this work illustrates both the opportunities and challenges of deploying data-driven methods in real-world electrochemical systems. The presented frameworks provide a foundation for extending autonomous, data-efficient optimization toward predictive electrolyte design and broader adoption in advanced energy storage research.
The discovery of advanced alloys capable of withstanding extreme environmental conditions such as high temperatures, intense radiation, corrosive atmospheres, and mechanical stress is critical for applications in aerospace, nuclear energy, deep-sea exploration, and space missions. Traditional experimental and computational approaches to alloy design are often time-consuming and resource-intensive. Recent advances in artificial intelligence (AI) and machine learning (ML) offer powerful tools to accelerate the discovery process by enabling high-throughput screening, property prediction, and design optimization. This paper presents a comprehensive review and methodology for AI-assisted alloy discovery, focusing on the integration of data-driven models with physical principles, high-fidelity simulations, and experimental validation. We highlight successful case studies, discuss the challenges of data scarcity and model interpretability, and propose a framework for closed-loop design that incorporates generative models and active learning. This AI-driven approach represents a paradigm shift toward faster, more cost-effective discovery of next-generation materials for extreme environments.
Venkatesh Iyer, Nandhini Ravi· International Journal of Mod...· 0 citations