While some materials can be discovered and engineered using standalone self-driving workflows, coordinating multiple stakeholders and workflows toward a common goal could advance autonomous experimentation (AE) for accelerated materials discovery (AMD). Here, we describe a scalable AMD paradigm based on AE and "collaborative learning". Collaborative learning using a novel consensus Bayesian optimization (BO) model enabled the rapid discovery of mechanically optimized composite polysaccharide hydrogels. The collaborative workflow outperformed a non-collaborating AMD workflow scaled by independent learning based on the trend of mechanical property evolution over eight experimental iterations, corresponding to a budget limit. After five iterations, four collaborating clients obtained notable material performance (i.e., composition discovery). Collaborative learning by consensus BO can enable scaling and performance optimization for a range of self-driving materials research workflows driven by optimally cooperating humans and machines that share a material design objective.
Keywords: Bayesian optimization; active learning; autonomous experimentation; glycomaterials; materials genome initiative.