An active-learning self-driving lab for polymer blend phase maps
Current work at Berkeley Lab — a multimodal self-driving lab that uses active learning to map polymer blend morphology and phase behavior, choosing each next sample where the map is least certain.
- Headline result
- NowALS Doctoral Fellowship, Berkeley Lab, 2026–27
- Period
- April 2026 – presentActive
- With
- Lawrence Berkeley National Laboratory · Advanced Light Source
From hitting a target to mapping the space
The closed-loop work searched for one structure. This project asks a broader question: what does the whole morphology landscape of a polymer blend look like? Instead of optimizing toward a single target, an active-learning loop builds a phase map, placing each new sample where the current map is least certain — typically along the boundaries between phases, where a fixed grid wastes most of its points.
Multimodal by design
Scattering alone does not settle every phase assignment. The loop combines GISAXS with complementary measurements so that each sample’s morphology is classified from more than one view, and the active learner works from that combined picture.
NOVA ties it together
NOVA is the central control software I built to run the loop as one system: it drives the robots and data acquisition, runs the reduction, updates the model, and dispatches the next sample within a single campaign. More on NOVA.
Work in progress. This project is active and unpublished; results will be added as they are released.