Scalable Thermoset Informatics Integrated with Autonomous Experimentation for On-Demand Discovery

As photopolymer additive manufacturing matures, there is an increasing need for on-demand, application-specific materials. A major bottleneck is the limited exploration of the vast design space for new thermoset polymer formulations, which must satisfy both material property requirements and processing constraints. This challenge arises from the combinatorial complexity of formulations and the high cost of experimental evaluation. In this work, we present an end-to-end pipeline that integrates a partial self-driving laboratory (p-SDL) with informatics-driven optimization to discover and validate thermoset acrylate formulations. The pipeline begins with a thermoset acrylate database that consolidates multiple data sources, including in-house experiments, molecular dynamics simulations, and literature data. These data sets are used to train single-task and multitask predictive models, which interface directly with the p-SDL to guide experiments and measure mechanical and processing-related properties. Each component is validated independently and then deployed together to optimize a target elastomeric material with over 90% reduction in human effort. The complete optimization required only ∼40 h of active experimental time, demonstrating rapid identification of formulations that satisfy multiple target properties without exhaustive screening. This work establishes a scalable framework for embedding multifidelity machine learning within laboratory workflows, enabling autonomous, data-driven thermoset formulation discovery.