Agentic Self-Driving Lab Accelerates Scientific Discovery and Validation

While AI-for-Science agents can automate ideation, planning, and analysis, the crucial step of experimental validation often remains a bottleneck, limited by the number of experiments required or the high cost of each trial. This research proposes an agentic self-driving lab (SDL) designed to tackle these physical constraints. The SDL integrates two key agentic components. First, a prior-aware agentic Design of Experiments (DOE) loop leverages domain knowledge and past results to intelligently propose the most informative next experiments, thereby reducing the total number of trials needed to reach a target. Second, a cost-aware surrogate agent predicts high-resolution, high-cost measurements from more affordable, low-resolution data. This agent intelligently decides whether to perform a high-cost or low-cost measurement based on predicted uncertainty. By combining these elements, the single agent aims to significantly accelerate the SDL loop by optimizing both the quantity and cost of experiments in scientific discovery.