Sensor Fusion Under Bounded Latency
This research domain develops sensor fusion methodologies, multi-modal input normalization techniques, and real-time processing architectures capable of handling diverse physical and digital input streams with bounded latency — a requirement often treated as secondary in academic sensor fusion research, but decisive for production operation.
From Research to Measured Practice
The practical validation of this research sits in Henri's nine parallel, independent anomaly detection skills: each skill processes a distinct signal class with no shared state, running a three-stage pipeline of signal fingerprinting, cross-cycle correlation, and classification. In the current production deployment, measured classification latency runs under 40 milliseconds per cycle, entirely on embedded hardware.
Why Parallel, Stateless Processing
A key research finding that flowed directly into the architecture: a misclassification in one signal class must not contaminate classification in another. That led to the design decision to run the nine detection skills in parallel with no shared state, rather than merging them into a single monolithic model.