Autonomous survey systems

Flying is the easy part

Route planning, station keeping, obstacle avoidance, returning home on a low battery — these are close enough to solved that they can be bought. The airframe is not the bottleneck and has not been for some time.

The bottleneck is everything after landing. A single survey flight over a modest site produces a few thousand frames. Somebody has to look at them, or something does, and the honest state of most operations is that the imagery is captured diligently and reviewed sporadically.

The pipeline between capture and decision

Set out plainly, the chain is: capture, transfer, store, register against previous flights, detect what changed, decide whether the change matters, tell someone who can act. Most attention and most money goes to the first three. The value is entirely in the last three.

The gap is not usually a modelling problem. Change detection over registered imagery is a well-understood technique. The gap is that nothing owns the step where a detected change becomes a work order, so detections accumulate in a dashboard nobody has a reason to open.

On-board triage

The alternative we are most interested in is moving the first filter onto the aircraft. If the machine can decide during the flight that ninety per cent of what it is seeing is unchanged, the amount that has to be transferred, stored and reviewed drops to something a person can genuinely look at the same afternoon.

This trades a hard problem for a different hard problem. Anything discarded in the air is discarded permanently, and the aircraft is making that call with less context than the ground system would have. The interesting design question is what the aircraft is allowed to throw away, and what it must keep regardless of how boring it looks.

• Keep everything on the first flight of a site — you have no baseline to compare against.
• Keep anything near a boundary, an access point or previously flagged region, whether or not it changed.
• Keep a low-resolution version of everything discarded, so a later question can at least be answered with “nothing was visible there”.

Anything the aircraft discards is discarded permanently. That is a governance decision, not a compression setting.

What has to be kept

Survey data is evidence. On an infrastructure or construction site it may be the only record of what a location looked like on a particular morning, and it can be asked for years later by people who were not thinking about storage costs.

Retention therefore has to be decided before the pipeline is built, not after the disks fill. It is the least glamorous part of this work and the part most likely to be the reason a system is trusted or quietly abandoned.