Ragnarok Solutions / Robotics / Sensors Sensing lines

No single sensor survives every setting.

Cameras fail in glare and smoke. Lidar fails in dust. Radar sees through weather and then sees ghosts. Sound fails when the machine changes. We test each sensor where it is weakest, and we test the layer that cross-checks them, so the system says when it cannot trust what it sees.

Six sensing lines

what it is for, where it fails, what we test
Light

Lasers and photonics

Vibration read from a distance for machine health. Gas leaks and plumes read by laser. Lidar for shape and range. Laser scanning for structural change. They fail in dust, smoke, rain and under a spoofed or dazzled return. We test lidar and laser readers under obscurants and attack, and we score plume and vibration AI against known truth.

Radio

Radar

Sees through fog, dust, sleet and snow when cameras and lidar cannot. Finds voids and rebar behind concrete. Tracks drones near a facility. It fails on clutter, ghost echoes, interference and ice on the cover. We build sealed radar test sets and score how a system handles a jammed or spoofed return.

Sound and touch

Acoustic and vibration

Machine health from sound and vibration. Leaks heard before they are seen. Sonar in murky water. Tap testing of structures. The failure is a model that learned the machine instead of the fault, and a robot that cannot hear past its own noise. Our test sets hold out whole machines, so a score means the fault was learned.

Heat and spectrum

Thermal and multispectral

Thermal for heat loss, hot spots and night work. Shortwave infrared and gated imaging through haze. Polarization against glare. Event cameras for fast motion. They fail on reflections, thermal crossover and a dirty window. We score imaging AI by condition with the true temperature or state known for every item.

Position

Navigation without GPS

Underground, indoors, under water, in a jammed area, off-world. Drift is expected. The hazard is a robot that drifts at full confidence. We score whether the system knows when its position can no longer be trusted, and how fast it says so.

Test data

Synthetic sensor test sets

Real base data with physics injected on top, or scenes rendered with known truth, so every item has an exact answer. Keys are sealed by hash before any run. Zero-signal items count false alarms. The synthetic set is checked against a real bench before it is used to score anyone.

The layer that cross-checks

Fusion is only as good as the referee.

A robot that carries five sensors still has one picture of the world. Something has to decide which sensor to believe when they disagree, when to re-check, and when to stop and say so. We test that layer directly: one sensor is fed a wrong return and the others are not, and we score what the system does next.

The same test tells you whether a fleet can be spoofed from the outside, and how far a robot gets before it notices.

How a sensor test runs

four steps, each one checkable
Step 1

Name the conditions

Your site, your sensors, the conditions that matter: dust density, light, temperature, motion, interference. Written down before any item is built.

Step 2

Build and seal

Items with exact truth across those conditions. Keys hashed and timestamped. The hash goes to you before the run.

Step 3

Run

Your system reads the items. We log every reading, every abstention and every false alarm.

Step 4

Score and keep

One score per condition, the misses with the truth beside them, and the test set kept for the next version to pass.

What an untested sensor costs

Published accuracy was measured on a clean day.

Your plant does not have clean days. A sensor that was never scored in dust, glare or cold is a sensor whose first real test is an incident. A sealed test is a fixed fee and a few weeks.

Tell us the sensor and the worst day it will see.

The sensor or sensor suite, what it reads, and the conditions you worry about. We reply within one working day with how we would test it.

Photos: valve and pressure gauge by Shixart1985, CC BY 2.0, cropped. Laser scanner, U.S. Air Force courtesy photo, cropped and edited. Ground radar in Greenland by Capricorn4049, CC BY-SA 4.0, cropped, this crop under the same license.