Vision

A staged path from prototype to validated field use

Where we want to take agri-edge, in the order we intend to earn it. Stages, not dates. A stage is only marked complete when it has been demonstrated.

Roadmap

Read top to bottom. Solid stages are demonstrated; hatched stages are planned and not started.

Stage1Complete

Detection models benchmarked

Detection models are evaluated on public datasets so we know what the pipeline can and cannot do.

  • Sheep model AP@.50 0.905
  • ImageWeeds about 0.457 (species), 0.673 (binary weed)
  • weed_crop 0.749: pipeline proof only, not field-aerial
Stage2Complete in simulation

Ops Station requirements met in simulation

The operations interface is built and exercised against simulated scenarios.

  • 88/88 requirements (§§79–166) simulation-complete
  • Not field-complete
  • Demo uses synthetic data only
Stage3Planned, not started

Hardware and loop validation on the bench

Prove the companion computer, Detection API and SITL adapter work together before anything flies with them.

  • Hailo/Pi, Detection API, SITL adapter: not yet validated
  • Full-loop validation: not completed
Stage4Planned, not started

Supervised field pilots with partners

Scoped trials with defined acceptance criteria, run alongside the operator and agreed with them in advance.

  • Operator reviews every recommendation
  • Treatment needs validated_locally and a geofence
  • No results exist yet
Stage5Intent only

Wider use, once evidence supports it

Only after earlier stages are demonstrated would we widen scope. Nothing about this stage is promised.

  • Scope set by what validation shows
  • Regulatory requirements are the operator’s and ours to meet, step by step

Guardrails that do not move

Whatever the stage, the safety design stays the same. Features are added around it, not through it.

Always true

  • The Pi never flies.
  • MAVLink is read-only toward the aircraft.
  • Treatment requires validated_locally plus a geofence.
  • The phone companion at /app stays read-only.