Data development is critically important for the emerging polymer informatics ecosystems. Two primary challenges of preparing a computational polymers database are which polymer (out of a huge number of possible candidates) should be considered and how to obtain a reasonable polymer model, from which calculations for relevant can be performed. We have developed an autonomous computational workflow, involving an active-learning guidance step and an efficient Polymer Structure Prediction strategy, to solve these problems. Our scheme has been implemented in the Computational Autonomy for Materials Discovery package and used to significantly enlarge the computational polymer database that power the Polymer Genome platform (https://www.polymergenome.org/). Machine-learning models trained on our computational data for predicting various polymer properties are available in https://www.polymergenome.org/. This work is supported by Toyota Research Institute and Office of Naval Research.
