MarutAI | The Science of Autonomy

The challenge

Testing every candidate formulation physically is cost-prohibitive. Teams iterate slowly through a small fraction of viable compositions, often missing better-performing regions entirely.

The approach

A process twin calibrated from experimental data models formulation chemistry and property responses. VATS generates regimes across the property space and constructs decision models that guide experimentation toward high-value candidates.

Expected outcome

Reduce physical testing by using model rankings to choose which formulations enter the lab.

The challenge

Predicting how a new composition will behave requires expensive characterization. Each round of synthesis and testing adds weeks and cost before a property profile is available.

The approach

Twin-calibrated models predict mechanical, thermal, and chemical properties from composition data. Predictions are bounded by the twin abstraction and flagged when outside the calibrated regime.

Expected outcome

Estimate property ranges before synthesis so lab work can focus on candidates within the calibrated regime.

The challenge

The number of possible experiments grows combinatorially with the number of variables. Running all of them is not feasible, and choosing which to run is itself a hard problem.

The approach

VATS-constructed models identify the experiments with the highest information value relative to the current twin calibration. Each run refines the twin and reshapes the next batch of experiments.

Expected outcome

Choose experiments by expected information value, then use each result to update the next batch.