The self-driving lab

The Undergrad turns your existing lab into a self-driving lab.

A robot that physically moves between the instruments you already own — connecting every station — and, together with an orchestration layer, turns your bench into an autonomous, self-driving lab.

How the Undergrad works

One robot, the whole bench

Glides along a rail to serve every station — no lab redesign.

Vision-verified at every step

Cameras confirm each transfer; errors are caught at the bench, not the reader.

Runs day & night

Load once and the workflow repeats — load, run, unload — hands-free.

Proven result

A real cloning protocol, run end-to-end.

Robot on Rails and Red Queen Bio ran a real wet-lab molecular cloning protocol on the Undergrad — a 79× increase in cloning efficiency, measured on a real bench, not simulated.

Read the case study →

DNA cloning workflow, run end-to-end on the Undergrad.
Perception, not repetition

It sees what it's doing.

Automation has always been able to repeat; it has never been able to perceive. The Undergrad sees the actual deck, understands the goal, and adapts to a real bench — instead of blindly replaying a script that assumes the world is perfect.

Most failed runs are silent handling errors — a missed tip pickup, a misaligned plate, an incomplete aspiration — found only later in the data. Onboard vision checks each step as it happens and flags or re-seats an exception mid-run, so a clean result means the steps actually happened.

Errors caught in real time

The vision loop checks each atomic action against what it expects to see, so a problem is caught at the bench, not at the reader.

Walk-away confidence

Because it verifies its own work, you can leave it running — and trust that a clean result means the steps actually happened.

Brain + body

A mind that understands your science, a body that runs it

Traditional automation is all body and no mind — it replays a fixed script and assumes the world is perfect. Physical AI joins the two: perception and reasoning that understand intent, driving a robot that acts and verifies on a real bench.

Plain-language reasoning + vision

You describe the protocol the way you'd brief a colleague. The system interprets that intent into an executable plan, and onboard vision lets it perceive the actual deck — plates, tips, liquids — rather than assuming where everything is.

The rail robot + your instruments

A precise rail-mounted robot operates on your existing bench and the instruments you already trust — thermocyclers, readers, magnets, the Accuris AutoMATE 96 — instead of replacing them or demanding a dedicated room.

Architecture

Four layers, from intent to verified action

You work at the top, in plain language. The system handles everything below — down to camera-verified motion.

↑ You work herePlain language
1
IntentWhat you want, in natural language — no scripting.
2
ProtocolStructured, checkable steps: volumes, labware, conditions, order.
3
Robot actionsGrouped operations: bind, wash, transfer, incubate, read.
4
Atomic actionsPrecise instrument-level moves — each confirmed on camera.
↓ System handles thisVision-verified
Learning flywheel

Better with every use

Every run is training data. Doing the science is what makes the science get easier.

Intentrecords Visiontruth Anomalysignal Simulation

Intent records capture the goal; cameras capture ground truth; every re-seat and flag is labelled anomaly signal; and runs mirror into simulation so gains compound across every deployment — removing run-to-run variation so downstream differences reflect biology, not who pipetted that day.

Breadth is depth

One general-purpose bench operator

Fixed automation gets faster at one task. The Undergrad gets broader — because perception and plain language are general, every new workflow is a description, not an integration project. Breadth is what compounds into depth.

Sees the bench

Onboard vision lets it work on a real, unstructured deck — expanding what it can act on without re-engineering for every layout.

Speaks your protocols

Plain-language understanding turns any described workflow into action — a new assay is a new sentence, not a new build.

Across your workflows

Confirmed integration today: the Accuris AutoMATE 96.

GenomicsImmunoassaysLiquid handlingmRNA / RNACell biology

Browse the full protocol library →

System specifications for the Undergrad: benchtop rail form factor, 6-DOF manipulation, multi-camera vision system, LLM-powered intelligence, equipment-agnostic compatibility, and full deployment support.
Trust

Transparency, not a black box

A clear line separates your experimental IP from our system intelligence. We use what we learn to make your robot better — never to compete with your science or share what you've built.

Your protocols are yours

What you build on the Undergrad belongs to you.

Introspectable logs

Readable, audit-ready records of every step — no opaque decisions.

A partner, not a vendor

Always-on support and iteration, in the lab with you.

Questions

Frequently asked questions

What is Robot on Rails?

Robot on Rails builds Physical AI Lab Operators. The Undergrad physically runs experimental protocols from plain-language instructions, vision-verified at every step, so scientists spend their time on science instead of repetitive bench work.

What is a Physical AI Lab Operator?

AI that doesn't just plan on a screen — it physically acts in the lab. It interprets a plain-language protocol, converts it into precise actions, and carries them out on real instruments with vision-guided verification.

Do I need to know how to code to use it?

No. You describe your protocol in plain language. The system converts it into atomic steps, simulates them for your review, then runs them physically — no scripting required.

Does it work with my existing lab instruments?

Yes. The Undergrad is equipment-agnostic and installs on a standard bench without facility modification. It operates the instruments and labware you already use, with no vendor lock-in.

What proof exists that it works?

Robot on Rails delivered a 79× improvement on a molecular cloning protocol, validated with its customer Red Queen Bio. The company has reached $2M+ in revenue, bootstrapped, with a committed long-term contract with Red Queen Bio.

The next experiment runs itself

Bring physical AI to your existing instruments.

Book a demo Browse protocols