Autonomous AI Agents in ORR Electrocatalyst Discovery: A Review of Closed-Loop Workflows, Materials Bottlenecks, and Energy System Translation

Abstract

The commercialization of fuel cells and metal-air batteries is hindered by the slow kinetics of the oxygen reduction reaction (ORR) at the cathode and the high cost of the platinum supported on carbon (Pt/C) catalyst. Although Pt/C catalysts remain the benchmark, their reliance on scarce, expensive platinum and limited durability under acidic conditions underscores the urgent need for alternative materials. Promising candidates, including 2D materials, M-N-C single-atom catalysts, and high-entropy alloys, occupy vast and complex compositional spaces, making traditional trial-anderror approaches too slow to fully explore them. While machine learning has accelerated the identification of potential catalysts, challenges, such as long-term stability, practical synthesizability, and experimental validation remain unresolved. In this Perspective, we highlight the emerging role of autonomous AI agents in ORR catalyst discovery. By integrating large language models, structured knowledge graphs, and robotic high-throughput experimentation, these systems can autonomously design, execute, analyze, and refine experiments with minimal human input. This review examines key challenges to realizing this vision, including the need for standardized operando datasets, explainable AI frameworks that provide chemical insight, and robust strategies for synthesizing complex materials like 2D sheets.Finally, we propose a roadmap for the development and implementation of these autonomous systems, aiming to accelerate the discovery of the next generation of ORR electrocatalysts.