Sensor fusion that accounts for every report
When sensors produce data faster than a processor can fuse it, today's systems drop the lower-priority reports and move on. Accountable fusion records what happened to every report it accepts, gives the same answer whether it runs on one processor core or many, and says so plainly when it falls behind.
The problem
Aircraft, ships, satellites and autonomous vehicles fuse reports from radar, electro-optical, infrared and navigation sensors into one picture. Those sensors keep getting faster and more numerous, and most fusion software still processes their reports largely one at a time.
When the reports arrive faster than they can be fused, the usual answer is to rank them and drop the ones that rank lowest. The dropped reports leave no trace, so nobody can say afterwards which data the picture was built from, or whether a lost track was lost to the sensor or to the processor.
Adding processor cores does not fix this by itself. Work split across cores finishes in a different order every time, so two runs over the same data can associate reports differently. A system that gives different answers to the same input cannot be tested, explained or certified with confidence.
Reports also arrive late. A report delayed on a slow link describes the world as it was when it was measured, not when it arrived, and treating it as current quietly distorts the picture.
How it works
Four properties, held every processing cycle.
Account for every report.
Every report the system accepts receives exactly one recorded outcome: fused into the picture, deferred to a later cycle, or set aside with its reason. A report the system cannot take is refused at the door and recorded as refused, never silently lost.
Place each report by when it was measured.
A report is fused according to the moment it describes, not the moment it arrived. A late report corrects the retained picture going forward, and the correction is recorded next to the original decision instead of overwriting it.
Same answer on any number of cores.
The work is spread across processor cores, and the result does not depend on how many cores ran it or which finished first. A single-threaded replay of the same recorded input reproduces the same result, so the parallel system can be checked against a simple one.
Say when it falls behind, and show where two runs differ.
When the load exceeds what the system can carry, it reports that as a degraded state the operator can see, rather than presenting an incomplete picture as a complete one. When two runs of the same data disagree, the record shows where: which cycle, and which report or track.
Design properties
Complete
Every accepted report has a recorded outcome. Nothing is dropped without a trace.
Reproducible
The same recorded input gives the same result on one core or many.
Honest under load
Falling behind is declared as degradation, not hidden.
Time-correct
Late reports are placed by measurement time, and corrections never rewrite the record.
Checkable
The record can be replayed and compared to locate any difference.
Algorithm-neutral
Works around the association and estimation methods an integrator already uses, rather than replacing them.
Where it applies
Military aircraft and ships
Fusing high-rate radar, distributed-aperture and targeting-sensor reports on multi-core mission processors, without trading lost tracks for throughput.
Space-based tracking and warning
Fusing observations from many satellites in several orbits, where operators need to know which observations were used and that the same data always gives the same answer.
Autonomous vehicles and drones
Perception fusion with no operator in the loop, where a silently dropped sensor report is a safety question.
Avionics certification
Reproducible results and a complete record of every input are the kind of evidence safety assessment and certification work relies on.
Maritime bridge systems
Integrating radar, AIS and navigation feeds on commercial bridges, where congestion and interference drive input rates up exactly when the watch is busiest.
Regulated data pipelines
Any system that consolidates several asynchronous feeds and must later prove that nothing was lost, duplicated or reordered.
Applications beyond the naval and aviation focus are directions for licensing and partnership, not current deployments.
How it relates to SENTINEL
SENTINEL decides which navigation sources can be trusted and records why. Accountable fusion applies the same discipline, deterministic behavior and a verifiable record, to fusing many sensor reports at high rate. The two run side by side: SENTINEL judges source integrity, and accountable fusion accounts for every report that goes into the picture.
Patent
Patent pending. U.S. provisional application 64/159,476, Accountable Deterministic Fusion of Asynchronous Multi-Source Measurements With Per-Report Disposition Accounting, filed September 2026 by RV Technology Consulting. It builds on U.S. provisional applications 64/144,562 and 64/149,078, which cover SENTINEL.
Status
U.S. provisional patent application filed September 2026. Patent pending. Company-funded reference implementations, including a C11 implementation that allocates no memory at run time. Submitted as the basis of a U.S. Navy SBIR Phase I proposal on multi-core sensor fusion, September 2026.
Who is behind it
Rajan Vasudevan, Founder & Principal Investigator. 35 years in software architecture and design, including real-time payment validation and fraud detection, distributed data processing and public-sector modernization. Founder and CEO of OceanManager for 12 years, delivering maritime safety and compliance software to more than 1,200 cargo vessels.
Work with us
We are open to conversations with fusion-software integrators, avionics and mission-system suppliers, research partners, and organizations that need to prove what their sensor picture was built from.
Patent pending. Technical details beyond this page are shared under a non-disclosure agreement.