Skip to content
Work
UAV / SecurityPublic anonymized

UAV telemetry and field systems

CybersecurityInfrastructure & DevOpsAI / Computer VisionBackend Engineering

Resilient telemetry, edge inference, and control-path security

Architecture, telemetry, and security engineering

Context

UAV telemetry and command paths fail in ways that look like “the radio is fine.” Spoofed identifiers, delayed packets, and unverified control messages are architecture problems, not UI problems.

System boundary

In scope

  • vehicle sensors
  • transport
  • ground ingestion
  • validation
  • time-series store
  • operator control

Out of scope

airframe design

Architecture

The vehicle is untrusted relative to the ground station. Telemetry is authenticated and validated before it is stored. Control is a separate path with a tighter trust boundary than observation.

Field architecture for unmanned systems: vehicle telemetry, transport, validation, persistence, and a monitored control surface with explicit trust boundaries.
  1. Vehicle
  2. Transport
  3. Ingestion
  4. Validation
  5. Persistence
  6. Monitor / control
  • Dashed = trust boundary
  • Control is a separate path
Published sequence
  1. Vehicle
  2. Transport
  3. Ingestion
  4. Validation
  5. Persistence
  6. Monitor / control
Stages on each row
  • Row 13Vehicle · Ingestion · Persistence
  • Row 23Transport · Validation · Monitor / control
Roles in the diagram
  • external1
  • service1
  • process1
  • store1
  • interface2

Counts from the published record. They describe structure, not a measured result.

What crosses the trust boundary?

Untrusted field

  • externalVehicle

transit

  • interfaceTransport

Ground trust

  • serviceIngestion
  • processValidation
  • storePersistence

Control plane

  • interfaceMonitor / control
  1. VehicleTransport
  2. TransportIngestion
  3. IngestionValidation
  4. ValidationPersistence
  5. PersistenceMonitor / control

Critical flows

  1. Vehicle emits telemetry

  2. transport (NATS/MAVLink-class)

  3. ingestion

  4. schema and anomaly validation

  5. Timescale persistence

  6. observe/control surfaces. Command traffic does not share the telemetry write path.

Stages

  • Telemetry ingest
  • Trust boundaries
  • Time-series store
  • Edge inference
  • Control policy

Decisions

01

Separate observe and command

A compromised telemetry writer must not become a command issuer. Boundaries are in the topology, not in comments.

02

Validate before persist

Storage is not a dump. Identifier, clock, and envelope checks happen on the way in.

03

Edge inference is a peer, not a plugin

EO/IR models run at the edge with their own health, not as a side effect of the GCS UI.

When it fails

ConditionDetectionResponse
Identity or envelope violationIngest validatorDrop; increment security counter; do not store
Clock skew / replayMonotonic timestamp windowQuarantine stream; alert control
Transport partitionHeartbeat lossHold last validated state; freeze command unless policy allows
Store write failureCommit errorLocal WAL on ingest; backfill when store returns

Infrastructure

  • Field radio
  • NATS
  • ingest services
  • TimescaleDB
  • edge GPU/CPU
  • ground control host
  • audit log

Outcome

A field topology where telemetry can be distrusted, validated, and stored without collapsing into a single “GCS application.”

Current state

Active field architecture. Security and telemetry research continues against this topology.

  • NATS
  • TimescaleDB
  • Rust
  • Python
  • EO/IR
  • Edge inference
  • MAVLink
Contact