Multimodal Co-orchestration for Uncovering Structure–Property Relationships in Combinatorial Libraries

The rapid progress of high-throughput automated synthesis has greatly accelerated materials discovery; however, downstream characterization remains a critical bottleneck, as it requires diverse and resource-intensive measurements across electrical, mechanical, chemical, and structural properties. We introduce a Multimodal Co-orchestration workflow that integrates multiple experimental techniques into a unified automated framework for the characterization of combinatorial libraries. This workflow employs Bayesian optimization in the latent space of variational autoencoders (VAEs), which perform unsupervised feature extraction directly from raw data. Multi-task Gaussian processes (GPs) serve as surrogate models, guiding the selection of optimal experimental trajectories within the combined modality–composition space. The efficacy of this approach was demonstrated through the co-orchestration of piezoresponse force microscopy and Raman spectroscopy for the characterization of a Sm–BiFeO₃ combinatorial library. This generalizable framework provides a flexible and robust algorithm for the efficient, autonomous exploration of combinatorial libraries and related material systems.