Every autonomous decision starts with a sensor.
Sensors degrade for years before anything flags it.
Know when your sensors stop telling the truth.
obsurver names the sensor that's degrading, how far gone it is and when it hits its limit, while the self-test still says OK. Read-only software, for any sensor, on the stack you already ship.
Detecting degradation the moment it happens.
A camera loses focus, its window soils, a repair leaves it a degree off. The self-test keeps saying OK, because it only asks whether the camera is alive, not what it sees. This is the picture your stack gets. Drag the slider to age it.
risk events a year on EU roads by 2029, from reduced lane-keeping performance alone.
TÜV Rheinland with TRL, 2021.
- Edges softenat baselineHeat and age shift the focus. The person on the crossing blurs out at 60 m first, then at 40.
- A soiled patchnoneRoad film builds on the window. Behind it the picture goes dim and smeared, and returns are lost.
- Off axisalignedA windscreen swap leaves the camera a degree off. Everything it reports sits a little beside where it is.
Illustrative. DEKRA 2023: a front camera misaligned below the self-diagnosis threshold, emergency braking could not prevent impact at 20 km/h. Pandey et al. 2025: aging defocus, precision unchanged, recall below 0.5 beyond 60 m.
obsurver knows what every sensor should see.
On day one obsurver records what each sensor delivers. Every drive after that is compared with that record. Cross a line you agreed on and your function supervisor gets a recommendation: hold the automation level, or lower it. Never raise it.
one sensor, one integrity metric, normalized to its own baseline
what obsurver recommends, as the measurement crosses each line
-
passwithin tolerance
No change. The function runs as designed and evidence accrues with every drive.
-
watch10 % below
Driver notice to clean or check the named sensor. The unit is flagged in the fleet view with its first event.
-
warn25 % below
Restrict the dependent function. A work order goes out with the date the limit will be reached.
-
fail50 % below
Function unavailable until the sensor is recalibrated or replaced. Time-stamped proof of when it degraded and why.
recommended maximum automation levelmaintain
The lowest matching level wins. A measurement obsurver can't compare is flagged for review, never counted as a pass. obsurver advises, your function supervisor decides.
Works with camera, lidar, radar and infrared. Any supplier.
The architecture.
obsurver sits beside your stack, not inside it. It reads sensor frames before perception and object lists after it, compares both with each sensor's own history, and hands your function supervisor an advisory. Nothing is written back into the vehicle.
YOUR STACK
both taps feed obsurver; the advisory result returns to function supervision
KPI library frames in, measurements out
sharpness, noise, contrast, SNR, spatial, geometry, timing
Detection engine measurements in, classification out
within tolerance, weighted over evidence, slope projected forward. References: own baseline, fleet cohort, ground truth. Tolerances: absolute, age-related, drift.
Advisory result integrity per sensor, degradation event with severity, evidence record
Same problem on every platform.
Automotive is where the rules bind first, so that's where we start. Everywhere else the problem is the same: a sensor, its own baseline, and a self-test that never notices.
Motorway roadworks at night. The right lane closes behind cones, a crash-cushion truck sits in it, workers stand behind the barrier. The front camera has to get this picture right for fifteen years. Regulation now asks for monitoring across that whole life, not a test at type approval.
entry pointrecorded drive logs from a programme you already run
A reconnaissance drone over a convoy. The EO/IR gimbal looks down at three armoured vehicles on a dirt track and a walled compound to the right. Sand abrades the optic over hundreds of flight hours, and a degraded sector looks like a quiet sector until the mission depends on it.
entry pointrecorded sensor data from the platform, no integration
A humanoid at a packing bench. Its head camera looks down at three bins, its own hands in frame, a colleague at the line behind. The depth camera wears gradually and fails without warning, and here the consequence stands a metre away.
entry pointrecorded logs from the robots already in the hall
A tractor in maize. The cab camera looks over the bonnet along the rows to the headland while a combine works the next field. Heat, humidity and dust wear field sensors faster than anything on a road, and today the only fix is manual recalibration.
entry pointrecorded data from one season
Regulation demands change.
Catching sensor degradation in service used to be good practice. In automotive it's turning into a condition of approval and a question of who is liable.
UN Regulation on ADS adopted June 2026
A safety case and in-service monitoring across the vehicle's life. Sensing limits have to be evidenced, not assumed.
ISO 21448, SOTIF
The performance limits of sensing have to be known and covered, including how they move over the years.
ISO/TS 5083
Sets the evidence bar for automated driving systems, including how they are monitored in service.
EU Product Liability Directive 2024/2853 applies from 9 December 2026
Software is a product. When a sensor degrades in the field, the record of when and why decides who pays.
Built by people who helped write the standard.
obsurver was founded in Sindelfingen, in the middle of the German car industry, by three founders who have spent their careers in autonomous sensing. Two of them co-founded IEEE P2020, the working group on automotive image quality, and the team has published peer-reviewed work on sensor degradation since 2021.
We build obsurver together with stack providers, suppliers and research partners, so it fits real architectures, not idealized ones.
Fabian Schmidt
CEO and co-founder
Serial founder with 7+ years of tech leadership and an M.Sc. in Entrepreneurship from Babson College. Built companies across consulting, supply chain and mobility, and is one of Germany's youngest guest lecturers for AI and entrepreneurship.
LinkedIn
Benjamin May
CTO and co-founder
M.Sc. in Physics, University of Greifswald. Spent 15+ years leading ADAS and autonomous-driving system development for global OEMs, and co-founded the IEEE P2020 working group on automotive image quality.
LinkedIn
Dr. Sven Fleck
CSO and co-founder
M.S. and Ph.D. in Computer Science, University of Tübingen. Advises premium OEMs on imaging, is co-founder and vice-chair of IEEE P2020, and has 35+ scientific publications alongside expert-reviewer work for the European Commission.
LinkedIn
Backed by
Next Mobility Labs
obsurver was built inside Next Mobility Labs, the global mobility venture studio. NML acted as institutional co-founder: strategic guidance and access to a mobility and industrial network across Europe.
nextmobilitylabs.com
Göran GöhringMD Next Mobility Labs, angel investor
Peter MertensFormer Audi board and Volvo Cars CTO
Joachim LangenwalterBoard dSPACE, former NVIDIA director
Sabina JeschkeAdvisory board Rheinmetall and Aumovio
Dirk WollschlägerFormer General Manager, IBM Global Automotive
Bram SchotFormer Audi CEO and board of Volkswagen
Start on a drive log, not an integration.
30 minutes with the founders. Bring a recording if you have one, we bring the software.
