- What exactly does “autonomous” mean for a drone?
- It is a legal term, not a feature. Article 2 of Implementing Regulation (EU) 2019/947 defines autonomous operation as an operation during which the unmanned aircraft operates without the remote pilot being able to intervene. By that measure almost all industrially deployed systems fly highly automated rather than autonomously: as long as a remote pilot can override or abort over an existing radio link, responsibility in air law stays with the human. Genuine autonomy begins only where that fallback level is absent.
- Where does onboard autonomy actually run today, and what is a demonstration?
- It holds up where human intervention is physically impossible: in fully enclosed spaces and assets such as tanks, silos, boilers, ballast tanks and sewers. There, LiDAR-based SLAM combined with optical flow is the industry standard for positioning without GNSS. Sense-and-avoid outdoors, fusing radar, optical and LiDAR sensors, is regulatorily demanding and heavy on onboard compute — that is not routine operation, and we do not describe it as one.
- We only fly indoors. Does the EU frame apply at all?
- Not inside the building. The German Federal Aviation Office states in its FAQ on UAS operation that these requirements do not apply to the operation of UAS inside buildings. But practice is rarely purely indoors: according to manufacturer figures from Flyability, up to 70 percent of flights with inspection systems such as the Elios platform take place at least partly outdoors. Even the walk from the set-down point across the site to a manway or sewer shaft activates the full legal frame. Those transition zones belong in the assessment from the outset.
- What changes in the operating authorisation when decisions move onboard?
- What limits the operation is not flight capability but evidence. Complex inspection flights with autonomous or automated trajectory guidance fall into the specific category and therefore into a systematic risk assessment under SORA 2.5 — developed by JARUS, introduced by EASA with ED Decision 2025/018/R of 29 September 2025 and incorporated into the June 2026 revision of the Easy Access Rules. Containment is assessed there at Step 8, before the operational safety objectives are set at Step 9. What is assessed is the residual risk that the system leaves the defined operational volume through malfunction or navigation error.
- What do we actually have to be able to show an authority?
- Four things, none of them a manufacturer brochure. First, deterministic abort criteria: evidence that on sensor failure or contradictory attitude determination the onboard computer triggers defined safety manoeuvres rather than unpredictable corrections. Second, validated geofencing: evidence that physical boundaries are stored tamper-proof in the flight controller. Third, gapless telemetry: complete recording of raw sensor data, control commands and state decisions for later audit. Fourth, documented failure modes. Settle contractually as well who is the registered operator, who provides the remote pilots and who maintains the SORA documentation.
- Does the drone analyse the imagery as well?
- That is a different processing step, and this page deliberately does not cover it. What is treated here is exclusively flight and navigation decisions taken onboard in real time. Automated image analysis, crack detection and thermographic damage classification belong inside a structured inspection programme and are handled separately there.
- And on the ground?
- The same question arises for ground robots, under a different rulebook: what changes in specification, acceptance and change control when part of the behaviour comes out of a learned model? That is covered by the werob Physical AI solution.