Research direction
Human-supervised robotic science that can reason about evidence, detect when experiments or equipment deviate from plan, and recover or escalate safely.
Autonomous science should be more than a robot executing a fixed protocol or an optimiser selecting the next condition. I study how models, laboratory robotics and scientists can operate as one auditable loop in which experimental evidence informs what happens next and physical execution is monitored as part of the scientific reasoning process.
My emphasis is trustworthy hybrid autonomy. Humans retain authority over scientific intent, constraints and escalation boundaries; AI and models help propose and interpret experiments; robots execute and monitor physical procedures; and the system must recognise when uncertainty, equipment behaviour or protocol deviations no longer justify autonomous continuation.
This includes a resilience problem that is often hidden inside laboratory automation: an anomalous result can reflect new science, measurement error, a robot failure or a protocol violation. A useful robotic laboratory should gather diagnostic evidence, repeat or reconfigure a step when justified, preserve provenance, and know when to return the decision to a human scientist.
These publications are methodological foundations for model-based, auditable and reliable closed-loop experimentation; the autonomous-laboratory programme itself is now expanding through current projects.
Demonstrates a core scientific loop: use controlled physical interaction to infer hidden parameters, quantify a model of the system, and use that model for downstream decisions.
Combines identification, optimisation and physical execution in a closed-loop workflow that provides a methodological bridge toward robotic experimentation.
Addresses numerical reliability and reproducibility when differentiable models are used inside optimisation and inference loops — essential when model outputs guide real experiments.