Agentic LLM Reasoning in a Self-Driving Laboratory for Air-Sensitive Lithium Halide Spinel Conductors
Authors: Yuxing Fei, Bernardus Rendy, Xiaochen Yang, Junhee Woo, Xu Huang, Chang Li, Shilong Wang, David Milsted, Yan Zeng, Gerbrand Ceder
Organizations: Department of Materials Science & Engineering, University of California, Berkeley, Berkeley, California 94720, United States · Materials Sciences Division, Lawrence Berkeley National Laboratory, Berkeley, California 94720, United States · Energy Storage Research Alliance, Argonne National Laboratory, Lemont, Illinois 60439, United States · Department of Materials Science and Engineering, Korea Advanced Institute of Science and Technology, Daejeon 34141, Republic of Korea · Department of Chemistry and Biochemistry, Florida State University, Tallahassee, Florida 32306, United States
Self-driving laboratories promise to accelerate materials discovery. Yet current automated solid-state synthesis platforms are limited to ambient conditions, thereby precluding their use for air-sensitive materials. Here, we present A-Lab for Glovebox Powder Solid-state Synthesis (A-Lab GPSS), a robotic platform capable of synthesizing and characterizing air-sensitive inorganic materials under strict air-free conditions. By integrating an agentic AI framework into the A-Lab GPSS platform, we structure autonomous experimental design through abductive and inductive reasoning. We deploy this platform to explore the vast compositional space of lithium halide spinel solid-state ionic conductors. Across a synthesis campaign comprising 352 samples with diverse compositions, the system explores a broad chemical space, experimentally realizing 72% of the 171 possible pairwise combinations among the 19 metals considered in this study. Over the course of the campaign, the fraction of compositions exhibiting both good ionic conductivity (> 0.05 mS/cm) and high halide spinel phase purity increases from 1.33% in the first 75 agent-proposed samples to 5.33% in the final 75. Furthermore, by inspecting the AI's reasoning processes, we reveal distinct yet complementary discovery strategies: abductive reasoning interrogates abnormal observations within already explored regions, whereas inductive reasoning expands the search into broader, previously unvisited chemical space. This work establishes a scalable platform for the autonomous discovery of complex, air-sensitive solid-state materials.
Figures & tables
Figure 1: Overview of the A-Lab GPSS platform for air-sensitive solid-state synthesis, characterization, and impedance measurement. (a) Photograph of the customized NX2 -filled glovebox system housing the A-Lab GPSS setup. (b) Three-dimensional schematic of the GPSS glovebox layout, with individual modules color-coded and labeled on the right. (c) Synthesis and characterization workflow, comprising automated powder dispensing, mixing, high-temperature heating, grinding, and XRD sample preparation, followed by semi-automated powder XRD and electrochemical impedance spectroscopy (EIS) measurements. Multiple samples are connected to the heating tasks, as one heating cycle can handle multiple samples (up to 6). (d) Manual transfer step for sealing prepared XRD samples with Kapton film prior to XRD measurement outside the glovebox. (e) Customized electrochemical station for EIS measurements. The hydraulic press (right) is electrically connected to an external electrochemical workstation. During pressing, the pressing rod and the plate form a closed electrical circuit. The inset (left) shows the custom test cell, in which the sample is pelletized between two stainless-steel rod electrodes.
Figure 2: Agentic workflow for experimental design. (a-c) Experimental design agent workflows, each with different system prompts to focus on different aspects of scientific reasoning, using the same underlying LLM (OpenAI GPT-5). (a) The abnormality-detection agent first calls the LLM to analyze existing experimental data to identify anomalous samples whose XRD-identified phases or ionic conductivity deviate significantly from other chemically similar samples. For each detected anomaly, a subsequent LLM call generates a hypothesis explaining the deviation and proposes a targeted follow-up experiment to test the hypothesis. (b) The pattern-finding agent compares multiple samples to extract recurring trends and correlations in the dataset, and designs new experiments by extrapolating or applying these learned patterns. (c) The Bayesian-optimization- (BO-)assisted pattern-finding agent extends the pattern-finding workflow by first training a BO model on the full dataset to propose candidate compositions. These candidates are then evaluated by the LLM, which selects compositions that are consistent with the identified experimental patterns and introduces modifications only when necessary. (d) Agent outputs include all key parameters required for experiment execution. For the abnormality-detection agent, both the target composition and synthesis temperature may be adjusted. For the two pattern-finding agents, only the target composition is modified, while the synthesis temperature is estimated based on the melting points of the precursor materials. (e) The proposed samples are subsequently sent to A-Lab GPSS for experimental validation. The resulting XRD phase information and ionic conductivity are analyzed and fed back to the agents, which then run the next experimental design cycle.
Figure 3: Statistical analysis of the experimental results. (a) Co-occurrence matrix of cations across the campaign. The upper triangle reports the number of samples containing each cation pair, while the lower triangle reports the mean ionic conductivity for that pair. (b) Ionic conductivity versus lithium deficiency x in the target composition of the synthesis, within formula LiX2–x\/MX1\textpmy\/ClX4 . M represents the metal species other than Li. y denotes the degree of metal-site off-stoichiometry. Only samples with high spinel content (spinel phase weight fraction >80% and no clear unindexed peaks with relative intensity >0.3 normalized by the highest peak in the XRD pattern) are shown in this plot. (c) Boxplots of ionic conductivity grouped by the number of distinct metal species in the sample. Only high-spinel-content samples are shown. The center line denotes the median; the box edges correspond to the 25th and 75th percentiles. Circle markers indicate outliers beyond 1.5 times the interquartile range (IQR). (d) Temporal evolution of ionic conductivity across the campaign. Hollow markers indicate samples containing clearly identifiable non-spinel phases, either having a spinel phase weight fraction <80% or exhibiting a clear unindexed peak with relative intensity >0.3 compared to the most intense peak in the XRD pattern. The dashed line represents the running Pareto front of ionic conductivity.
Figure 4: Behavior analysis of experimental design agents. (a) Principal component analysis (PCA) of strategy embeddings derived from agents’ experiment-design justification. Four dominant strategy clusters emerge and are labeled accordingly: temperature tuning (shortened to “Temp tuning” in the figure), stoichiometric refinement, single-element doping, and multi-element co-doping. (b) The max (solid line) and mean (dashed line) values of all samples’ ionic conductivities over batches. (c) Shannon surprise of samples proposed by each agent; the mean surprise ( μ ) is annotated beside each violin plot. Surprise values are clipped at 10 to mitigate numerical instabilities arising from the logarithmic operation. (d) Ratio of external claims (statements not directly grounded in prior experimental data) versus batch number. Across panels, color indicates sample provenance: red (abnormality-detection agent), green (pattern-finding agent), and orange (BO-assisted pattern-finding agent).
Self-driving laboratories can explore synthesis conditions autonomously, but their decision-making layer is typically a black-box optimizer, and the output is a set of optimized samples, with the measurements reduced to predefined scalar objectives and the reasons behind success left unarticulated. Here we present SynAgent, a framework in which large language model agents operate an automated experimental system and maintain an explicit, revisable understanding of the synthesis process as the campaign's primary output. Starting with no predefined analysis pipeline, SynAgent adaptively generates analysis skills for newly acquired data and evolves this understanding through multimodal reasoning over experimental data such as X-ray diffraction patterns and electron micrographs. The evolution is guided by a verify-falsify scheme, in which the agent deliberately challenges its own hypotheses by testing conditions predicted to fail as well as those predicted to succeed. In a single campaign of 18 autonomous experiments using LiCoO2 (001) thin-film deposition as a testbed, SynAgent synthesized highly crystalline films and evolved an understanding of how the substrate temperature governs crystallization, discovering an abrupt threshold and a narrow optimal growth window at 650-690 °C. These results extend autonomous experimentation beyond optimized samples to testable, human-readable understanding.
Izumi Takahara, Kazunori Nishio, Akira Aiba +4
Institute of Industrial Science, The University of Tokyo, Tokyo 153-8505, Japan · Department of Chemistry, The University of Tokyo, Tokyo 113-0033, Japan · Rigaku Corporation, Tokyo 196-8666, Japan +1
Large language models (LLMs) have enabled agentic AI systems for scientific discovery, but most approaches remain limited to textbased reasoning without automated experimental verification. We propose MIND, an LLM-driven framework for automated hypothesis validation in materials research. MIND organizes the scientific discovery process into hypothesis refinement, experimentation, and debate-based validation within a multi-agent pipeline. For experimental verification, the system integrates Machine Learning Interatomic Potentials, particularly SevenNet-Omni, enabling scalable in-silico experiments. We also provide a web-based user interface for automated hypothesis testing. The modular design allows additional experimental modules to be integrated, making the framework adaptable to broader scientific workflows. The code is available at: https://github.com/IMMS-Ewha/MIND, and a demonstration video at: https://youtu.be/lqiFe1OQzN4.
Geonhee Ahn, Donghyun Lee, Hayoung Doo +3
Institute for Multiscale Matter and Systems (IMMS), Ewha Womans University, Seoul 03760, Republic of Korea
The current structure-centric paradigm in artificial intelligence (AI)-driven materials discovery, despite delivering thousands of candidate structures, is stalling at a critical barrier: the synthesizability gap. We argue that closing this gap demands a pivot to a synthesis-first paradigm in which executable synthesis protocols, not just atomic configurations, are treated as primary design variables. We outline a roadmap built on three pillars: (i) representing synthesis procedures as machine-readable protocols, (ii) deploying generative and inverse-design models to propose actionable reaction pathways and recipes, and (iii) integrating closed-loop optimisation to refine protocols against experimental realities and sustainability constraints. Framed in terms of the causal backbone P->X->y from protocol P to structure X and properties y, this perspective sets out methodological building blocks, standards needs and self-driving laboratory (SDL) integration strategies to accelerate reproducible, data-first materials discovery.
Guillaume Lambard
Data-driven materials design group, Center for Basic Research on Materials (CBRM) National Institute for Materials Science (NIMS), Namiki 1-1, Tsukuba, Ibaraki, 305-0044, Japan