TORUS does not just detect, it decides. Machine learning runs where the data is born: on the node, at the gateway, and in command. Each tier turns raw signal into a smaller, more certain piece of intelligence, so what travels the network and reaches the operator is a classified track, not a raw feed or a bare alarm.
Footfall, vehicle, or digging; voice, engine, or breaking glass. The system reports what it is, with a confidence, not merely that something happened.
A seismic cue plus an acoustic cue from a neighbouring node is a stronger track than either alone. Fusion is where confidence is earned.
Wind, rain, traffic, and wildlife are the real enemy of an unattended sensor. On-device models learn the background and hold their fire.
AI is distributed, not centralised. Every tier runs the inference its power and vantage allow, and passes a smaller, richer product to the next.
Geophone, microphone, thermal, LiDAR, RF at the node.
Node labels the event and its confidence; discards background.
Corroborate across the local ring; run visual inference on cue.
Form tracks across the field; suppress false alarms.
A classified track with location and confidence, ready to act on.
Each stage reduces data and raises certainty. A raw waveform becomes a labelled event, a set of events becomes a track, and a track becomes a decision, so the network carries kilobytes of intelligence rather than megabytes of feed.
| Tier | Runtime | Approach |
|---|---|---|
| Edge | ESP32-S3, no accelerator | Small quantized classifiers (TinyML) sized to RAM and the battery budget; wake-gated so inference runs only on a real cue |
| Gateway | NVIDIA Jetson Orin | Accelerated detection / tracking on thermal and visual streams; on-site so raw imagery stays local |
| Command | CCISRT server | Rule-based correlation today (explainable, auditable); learned multi-target tracking and classification fusion drop in behind the same track interface |
| All tiers | Isaac Sim | Scenarios generate labelled data and validate detection and false-alarm rates before field trials |
Explainable first. The command correlator ships as transparent rules so an operator can see why a track formed. Learned models are introduced behind that interface, with the rule-based path retained as a fallback and a sanity check.
Audio and imagery are processed where captured; only compact labelled events cross the network.
Sending decisions instead of feeds cuts bandwidth, power, and RF signature, which matters for a passive posture.
Track formation and command actions are logged, so an analyst can reconstruct why the system decided what it did.
| Now | Next |
|---|---|
| On-device seismic / acoustic classification at the node | Larger vocabularies and per-site adaptation via OTA model updates |
| Rule-based correlation and track formation at command | Learned multi-target tracking and cross-domain classification fusion behind the same interface |
| Isaac Sim scenario data for validation | Active learning from operator confirmations to cut false alarms over time |
TORUS is an independent platform. Company names and product model numbers referenced in this document (including but not limited to Espressif ESP32-S3 and NVIDIA Jetson, Orin, and Isaac Sim) are used solely for engineering and bill-of-materials identification and imply no affiliation with or endorsement by those companies. Supply of any such third-party product to the designers, integrators, or evaluators of the TORUS platform remains at the sole discretion of the respective owning company, organisation, or legal entity. "FOR EVALUATION" is a document-handling marking only and denotes no government classification. All trademarks are the property of their respective owners.