Self-Driving Scanning Probe Microscopy: From Acceleration to Discovery and Manipulation.

Scanning probe microscopy offers a powerful suite of techniques for investigating meso-, micro- and nanoscale phenomena ─ enabling the exploration of advanced materials, heterostructures, quantum materials, as well as providing insight into surface energetics and reactions. Additional spectroscopic capabilities combined with multimodal imaging allow for spatially resolved characterization of functional material properties. Despite its central role, microscopy is still predominantly conducted through a sequential, operator-driven workflow in which data acquisition, interpretation, and subsequent experimental decisions occur in discrete steps. This approach not only reduces experimental throughput but also limits the effective utilization of increasingly complex multimodal data sets, where critical information may remain unrecognized across real-time data streams. Recent advances in artificial intelligence (AI) and machine learning (ML) have enabled the concept of self-driving laboratories (SDL) which promise to accelerate scientific research by closing the loop between data acquisition, analysis, and experimental control. Within this broader paradigm, self-driving microscopy (SDM) has emerged as a powerful approach to overcome the limitations of traditional, operator-driven microscopy workflows. In this Account, we discuss the evolution of SDM from automated experimental platforms to fully autonomous systems capable of adaptive, decision-driven operation. We highlight its implementation for real-time data processing and optimization, feature identification, and strategies to efficiently explore across multidimensional parameter spaces. Advanced optimization strategies, including Bayesian optimization and reinforcement learning, allow SDM systems to iteratively refine experimental parameters based on prior observations, prioritizing measurements that maximize information gain. This adaptive approach reduces the total number of experiments required while improving the efficiency of exploration. Collectively, we showcase how SDM accelerates microscopy through AI/ML integration, enables real-time analysis from high-dimensional data acquisition, while simultaneously guiding the experiments toward discovery. In addition to characterization, the autonomous workflows allow for active material manipulation to create nanoscale artificial structures on demand. We highlight that formalizing experimental strategies into algorithmic workflows enhances integration and scalability across instruments and laboratories. These advances position SDM as a broadly applicable paradigm at the forefront of next-generation experimental science.