How Automata is Developing Adaptive Lab Execution With NVIDIA Nemotron
By enabling laboratory workflows to adapt when execution goes off plan, Automata aims to reduce manual intervention and advance autonomous science.
From Automata R&D: This article describes an internal experiment, not a currently available LINQ feature. The workflow drafts were checked and simulated against a virtual workcell configuration; no physical instruments were run.
Scientific AI is expanding what researchers can imagine and design.
Automata is developing adaptive lab execution using NVIDIA Nemotron and its LINQ lab orchestration platform. When conditions in the lab diverge from the plan, experiments can stall until an operator intervenes. Automata is exploring how agentic reasoning, simulation and orchestration can identify the problem, determine a recovery and coordinate next steps - keeping experiments moving with less manual intervention.
Reasoning through a laboratory disruption
The Automata team recently conducted research involving a plate placement scenario. A plate was set down on an instrument's nest at a 12° tilt and 6 mm off position (where it should sit flat and centred). In simulation in Automata's digital twin, a camera inspection after the placement measured how the plate was seated.
NVIDIA Nemotron then decided how to recover from that finding, and proposed the recovery plan to LINQ. Automata uses Nemotron 3 Super, served as an NVIDIA NIM microservice, running in the NVIDIA Agent Toolkit. It received the camera finding through LINQ's tools: the plate and instrument, the measured lean (8.8° in the camera image) and offset (11.7 mm), the 1.0° tolerance, the evidence, and the recovery options. It used those inputs to choose a recovery and explain its selection.
In the demonstrated runs it proposed reseating the plate; LINQ checked and scheduled the reseat. The plate was re-inspected at 0.0° and 0.0 mm, and Nemotron reported the outcome, which matched LINQ's own record.
This is where agentic reasoning enters the workflow: Nemotron read what the camera saw and proposed the recovery, with its reason; LINQ oversaw every decision that moved the hardware. When an operator asked Nemotron to move the plate to a second instrument that already held another plate, LINQ refused the proposal, naming the conflict; Nemotron proposed a reseat instead.
In the demonstrated scenario, LINQ checked each proposal before anything moved: whether the deviation exceeded the tolerance, whether the evidence might drive actuation, and whether the destination was free. It scheduled the accepted reseat like any other task, and the plate was re-inspected before the run continued.
Nemotron’s established contribution above is distinct from broader workflow components still under development. This scenario ran in Automata's simulation environment, against the digital model of the assay and instruments - no physical hardware was actuated.
A step toward autonomous science
Recovering from a plate-placement problem is a focused challenge with broader significance. As AI plays a greater role in proposing and designing experiments, labs will need to reliably translate that intelligence into physical execution - and adapt when reality diverges from the plan.
Automata’s goal is to make LINQ the operational layer connecting scientific intelligence with the physical laboratory. This work explores a practical step toward that vision: enabling lab execution to understand when something has changed, adapt the path forward, and keep experiments moving.
Explore the LINQ orchestration platform behind this research and what it enables for labs today.
