SensorPipe is the infrastructure layer that makes sensor pipeline health, calibration quality, and perception reliability visible — before they cause a field failure.
We've been inside sensor-dependent products long enough to see where the infrastructure is missing — not from research, from building.
A robot is calibrated and deployed. Over days or weeks, thermal drift, vibration, and firmware updates silently degrade sensor quality. Teams find out when perception accuracy drops — not before. The health score never updated. There was no health score.
A team runs Kalibr. Gets a matrix. Copies it into config. Months later: the robot misjudged a shelf. Which calibration was active? Which algorithm version? What was the reprojection error? What were the conditions? Nobody knows. There is no record.
A firmware update ships. A sensor driver's timing changes slightly. The calibration was run on old firmware. The CI pipeline checks model accuracy. It does not check whether the calibration is still valid. It isn't.
SensorPipe is software that sits between raw sensor data and a final report.
It takes sensor recordings as input, runs systematic checks at each stage,
and produces structured output — including a tamper-evident audit trail
and compliance evidence documentation.
Raw materials (sensor data) go in one end. At each station, a specific quality
check runs: sensor health, timing alignment, calibration quality, cross-sensor
consistency. The output at the other end is a validated, documented,
evidence-backed report of sensor pipeline health.
SensorPipe does not implement calibration algorithms. It orchestrates external
algorithms — Kalibr, OpenCalib, and others — validates their results against
configurable quality gates, and captures the full evidence trail automatically.
Run against a dataset. Produce a health report. Check calibration quality. Generate SOTIF-aligned evidence. CI gate pass/fail.
Run continuously against live sensor data. Health score per sensor, updated in real time. Trend detection. Alerts before thresholds breach.
Introduce a known fault into a clean dataset. Run the pipeline. Verify your detection catches it. Generate a fault detection certificate.
The format is open. The CLI is open. The software that analyzes it is commercial. Every calibration run, every health check, every sync violation, every quality gate result — captured automatically with SHA-256 batch hash chain tamper evidence.
$ sensorpipe trace verify session.sptrace → Chain integrity: OK [ session complete · no gaps detected ]
$ sensorpipe trace report --compliance SOTIF session.sptrace → SOTIF §8.3 — PASS (calibration verification: reprojection within threshold) → SOTIF §8.4 — PASS (sensor health: all sensors within tolerance at session start) → SOTIF §8.5 — FAIL (sync validation: camera↔lidar drift 43ms, threshold 10ms)
Format specification: github.com/ritzylab-dev/sensorpipe
The gaps SensorPipe addresses are documented in peer-reviewed research. Not invented as a product pitch — confirmed before writing the first line of code.
Sensor health that degrades silently between calibration sessions. Calibration with no audit trail, no version tracking, no comparison across algorithms. No systematic tool to verify sensors meet spec when received from a vendor. Calibration environments that produce bad results with no warning before running. Payload swaps on drones and modular robots with no field verification step. Radiometric calibration for multispectral and inspection sensors done manually. Cross-robot calibration consistency that nobody verifies at fleet scale. Environmental conditions that invalidate calibration silently.
The full problem set and research citations are in the specification document: github.com/ritzylab-dev/sensorpipe
What SensorPipe produces — structured event records, tamper-evident audit trails, typed calibration schemas with algorithm version and git commit — is directly useful as evidence in safety cases under:
The certification of your system is your responsibility. The evidence is what SensorPipe produces, automatically, without instrumentation.
EmbedIQ is RitzyLab's open-source firmware and gateway framework — the AI-first application layer that runs from MCU firmware through Linux gateways to edge AI devices. Apache 2.0. FreeRTOS, Zephyr, POSIX. The same architectural pattern across the full embedded stack. Tamper-evident event records on every target. Production deployed on real products.
SensorPipe inherits the same architectural DNA: open format specification, closed commercial software, compliance-aware by design, observable by default, zero instrumentation required.
If you want to understand how we think about structured observability for safety-critical systems — read the EmbedIQ ARCHITECTURE.md. It's public. Every design decision is documented and reasoned. No marketing.
Five roles, each with a different entry point to the same underlying problem.
Perception regression in the field. Days spent ruling out the software stack. Need to rule out sensor health and calibration drift in minutes, not days.
Ran the calibration. Got a matrix. Needs to know if it's trustworthy, how it compares to last month, and whether it will pass a CI gate before it ships.
Owns the CI pipeline. Needs a calibration quality gate the same way they have model accuracy gates — automated, reproducible, binary pass/fail.
Facing a SOTIF audit. Needs structured evidence mapped to specific clauses. Doesn't care how it's generated. Cares that it's tamper-evident and traceable.
Running a fleet of robots. Wants to know when a sensor is degrading — before it causes a failure. Wants a Slack alert, not a field incident report.
No demo. No pitch deck. Tell us what you're building, what's failing, and what you've tried. If SensorPipe is the right tool, we'll tell you. If it isn't, we'll tell you that too.
RitzyLab delivers calibration health audits and SOTIF evidence packages as a consulting service — independent of whether you install anything. The service exists. The conversation is free.