Chemistry Labs

Emerging interdisciplinary directions

Self-driving laboratories

How automated instruments, robotics, machine learning, and human expertise can form closed-loop experimental systems that choose, run, and learn from chemical experiments.

IntuitionAn experiment loop that learns

A self-driving laboratory repeatedly proposes an experiment, executes it with instruments or robots, measures the outcome, and uses that result to choose what to do next. Automation can increase throughput, but reliable chemistry still depends on sound protocols, calibration, safety, and interpretation.

Interactive topic simulation illustrating the cycle from experiment selection through execution and measurement to the next decision.

SchoolThe connected laboratory

Definition: Closed-loop experimentation

A closed loop connects a decision to an experiment and its measured result, then feeds that result back into the next decision. Unlike a fixed batch, the sequence can adapt as evidence accumulates.

The system may combine liquid handling, reactors, analytical instruments, sample tracking, and software. Each interface must preserve identity and units: a mislabeled sample or unit conversion error can make an automated data stream confidently wrong.

UndergraduateChoosing the next informative experiment

xt+1=arg⁡max⁡x∈Xa(x∣Dt)x_{t+1}=\arg\max_{x\in\mathcal{X}} a(x\mid\mathcal{D}_t)

An acquisition rule aa ranks possible experiments xx using current data Dt\mathcal{D}_t. It may favor expected improvement, information gain, or a balance of both. The chosen experiment must also satisfy operational constraints such as instrument availability, sample volume, time, and safety.

Definition: Design of experiments

Design of experiments selects combinations of controllable factors—such as temperature, concentration, or catalyst loading—to estimate effects efficiently. Randomization and replicates help distinguish real effects from drift and measurement noise.

AdvancedAutomation is a measurement system

A robot’s repeatability does not guarantee accuracy. Sensors require calibration, analytical methods require validation, and batch effects or carryover can bias results. Quality controls, blanks, standards, and periodic human review help reveal failures before they steer the optimization loop.

Chemical safety must be engineered into the workflow: compatible materials, containment, interlocks, waste handling, and human override. An optimizer must not be allowed to select an experiment outside an approved safety envelope simply because its predicted score is high.

ResearchResearch frontier

A rigorous autonomous experiment is traceable end to end: define the hypothesis and measurement, record instrument state and protocol, preserve raw data, quantify uncertainty, and report deviations. Human scientists remain responsible for deciding what the evidence warrants.

References

  • A mobile robotic chemist · B. Burger et al., 2020
  • Self-driving laboratories for chemistry and materials science · B. Häse, L. M. Roche, A. Aspuru-Guzik, 2019