UC Berkeley

The limited throughput of experimental validation increasingly constrains the promise of data-driven approaches in inorganic materials discovery. Self-driving laboratories have emerged as a promising solution to mitigate this bottleneck. In this thesis, I present my Ph.D. work on an integrated framework for autonomous solid-state synthesis and characterization aimed at enabling scalable exploration of complex materials spaces. By unifying robotics, workflow orchestration, automated characterization analysis, and AI-driven experiment planning, the self-driving closed-loop systems developed here increase experimental throughput, reduce human intervention, and enhance the quality, reproducibility, and interpretability of data. I detail the design and implementation of each component, their integration into a cohesive architecture, and the results from multiple discovery campaigns that demonstrate the approach. This thesis establishes a practical pathway toward scalable autonomous solid-state research, integrating synthesis, characterization, and machine learning to accelerate discovery in complex inorganic systems.Chapter 2 describes A-Lab, an autonomous platform for inorganic powder synthesis to which I contributed during my Ph.D. A-Lab integrates text-mined synthesis priors, robotic execution, automated powder X-ray diffraction (XRD) analysis, and active learning. In a 17-day continuous campaign, the platform successfully synthesized 36 of 57 target materials (63% success rate), spanning 33 elements and 40 structural prototypes, demonstrating reliable end-to-end autonomous synthesis for inorganic crystals. Chapter 3 introduces AlabOS, a reconfigurable workflow orchestration framework that enables robust operation of complex autonomous laboratories. AlabOS employs graph-based experiment representation, on-the-fly resource allocation, and built-in monitoring and error-recovery mechanisms. Deployed within A-Lab, it has orchestrated over 3,500 synthesized and characterized samples, demonstrating scalability and operational reliability in real-world autonomous materials research. Chapter 4 presents Dara, a data-driven framework for automatically analyzing powder XRD phase identification. Dara integrates tree search, peak matching, and Rietveld refinement to explicitly construct and evaluate competing phase hypotheses. In production use, it has processed 2,453 diffraction patterns with strong goodness-of-fit and practical runtime performance. Beyond automation, Dara quantifies intrinsic ambiguity in powder XRD interpretation arising from incomplete compositional information and enables targeted experiment redesign to systematically reduce uncertainty. Chapter 5 introduces A-Lab GPSS, an autonomous glovebox platform for air-sensitive solid-state synthesis coupled with large language model (LLM)-driven agentic experiment design. In a closed-loop campaign, the system synthesized and characterized 352 lithium halospinel compositions across 19 metals. Agent-driven exploration increased the hit rate for ionic conductivities above 0.05 mS/cm from 1.33% in the first 75 AI-agent submitted samples to 5.33% of the last 75 samples, while substantially expanding coverage of the chemical design space. Collectively, these systems demonstrate that tightly integrated autonomy, spanning synthesis, characterization, orchestration, and machine reasoning, provides a scalable and reproducible paradigm for accelerating discovery in solid-state inorganic materials.