The Camera · The Bridge Layer
The smart camera for every behavioral assay you already run. Patent-pending cross-polarized NIR optics reject background in hardware, real-time pose estimation runs on the device itself — and it drops into your existing protocol without changing a single step.
The Bridge Layer
There are roughly 200 published rodent behavioral assays. Dedicated instruments exist for only a handful of them. PolarEye is designed to capture any of them — point it at the paradigm you already run, with zero protocol change, and stream standardized, analysis-ready video into MouseOS.
Your legacy assays aren't a compromise. Decades of validated endpoints are exactly what makes the next generation of behavioral models possible.
No new arena, no new workflow, no re-validation of your paradigm. PolarEye drops into the assay as it exists today.
Cross-polarized NIR illumination cancels glare and background before a single line of software runs. Clean signal in, better models out.
Real-time pose estimation runs on the camera itself — edge AI that turns raw frames into kinematics as they're captured.
The Complete Readout
An animal in pain and a healthy animal can look nearly identical — the same silhouette, the same track through the arena. The signal that separates them is subtle: micro-adjustments in posture and gait, and the facial expressions researchers quantify with grimace scales. Legacy systems miss it. PolarEye measures it.
Full-body pose across body, head, tail, and face at 95 frames per second — 8,075 postural keypoints measured every second, on the device itself.
Grimace-scale facial expression analysis and facial recognition for individual-subject ID — no ear notching, no tail marks, no RFID, minimal handling stress.
PolarEye images at 850 nm, beyond the visual range of mice and rats. No visible light, no observer in the room, no behavioral artifact from the act of measurement itself.
How It Works
PolarEye illuminates the scene with polarized 850 nm near-infrared light — invisible to rodents — and images it through cross-oriented polarizing optics. Glare and background cancel in the optical path itself, rendering the animal as a clean, high-contrast form.
Contrast is generated by the physics of the light path, before a single line of software runs. Machine-learning models downstream inherit clean, consistent, high-signal input — which is precisely why they perform. On-device pose estimation then turns those frames into kinematics in real time.
The cross-polarized NIR optics are patent-pending. They are the foundation of the entire Virtual Labs capture stack.
Built for Machine Vision, Not for Humans
Consumer and legacy scientific cameras are designed to produce images that look good to people. PolarEye is designed to produce data that trains and runs models. The differences are not cosmetic.
Every pixel in every frame is exposed simultaneously. Rolling-shutter cameras smear fast limb movement across the readout — a silent killer of gait metrics. PolarEye never rolls.
A narrow bandpass filter on AR-coated optics rejects ambient room light entirely. Your data looks the same at noon and at midnight, lights on or lights off.
Frame rates up to 95 fps at full 5 MP resolution — fast enough to resolve the individual footfall dynamics of a running mouse, with the temporal resolution that stride-level kinematics demand.
The optical path delivers at least 3 pixels per millimeter at the imaging plane across the full field of view — enough to segment individual digits on a mouse paw.
Dedicated trigger input with sub-millisecond multi-camera sync. When multiple PolarEye units are arrayed together, every frame across every camera is captured in the same instant — a hardware guarantee, not a software timestamp.
Every unit ships with a machine-readable hardware profile — sensor, optics, filter, firmware, calibration record — and every frame is tagged with it. Your methods section writes itself.
Real-Time, Closed-Loop
Because scoring happens on the device in real time, PolarEye can act on behavior the moment it occurs — detecting scratching, rearing, grooming, freezing, paw lifting, convulsive behavior (flagged for welfare alerts), nocifensive behaviors, and social interaction as they happen.
Deliver von Frey or Hargreaves stimuli only when the animal is in the correct posture — gating validated by the camera, not the experimenter's eye.
Trigger recording or stimulation the instant a target behavior appears — including real-time place preference paradigms with no physical partitions at all.
Run staircase and Bayesian-adaptive designs that converge in fewer trials, and set welfare alerts that catch distress behaviors for faster intervention.
Hardware-accurate synchronization with your existing equipment — TTL/GPIO trigger in and out, frame-timing pulses.
Context Is Data
Behavior doesn't happen in a vacuum — it happens in a room. Every PolarEye session automatically records its own experimental context and stores it with the data.
Ambient light level, background sound, temperature, pressure, and humidity — logged continuously alongside every frame.
Ultrasonic vocalizations across the full murine and rat range, captured in sync with video — an affective signal most rigs never record.
Cloud-sourced context — weather, air quality, even thunder events that startle animals — joined to your sessions by time and place.
Uncontrolled variables become measured covariates. The variance you could never explain becomes variance you can model — and remove.
Reproducibility Is a Hardware Property
When your recordings arrive in MouseOS, the platform knows precisely what hardware produced them — so analyses are matched to hardware, cross-site data stays comparable, and behavioral results hold up across labs, across sites, and across years. PolarEye does it by default.
Specifications
Target specifications. Final performance figures are verified on production units.
*Academic pricing. Commit at SfN 2026 and receive 30% off as a Founding Customer — meet us in Washington, DC →
Works Alone. Built for More.
PolarEye is the first instrument in a larger platform. The next one debuts at SfN 2026 in Washington, DC — November 14–18. Be there when the covers come off.