APAI Airport Physical AI Readiness Index Start the assessment

An independent diagnostic framework · v1.0

Is your airport ready to host machines that move?

Changi Airport put driverless baggage tractors into live airside service in January 2026, after nearly a year of trials and more than 5,000 test trips, with 24 vehicles planned by 2027. That is an operation, not a demonstration.

Most airports could not follow tomorrow. Not because the technology is unavailable — increasingly it is off the shelf — but because the airport itself is not yet a viable operating environment for it.

Free · about 10 minutes · scored report at the end · no commercial data required

6Domains
30Criteria
5Maturity levels
40%Weighted to enablers

The diagnosis

Airports don't fail at physical AI in the robots

Having looked closely at why airport automation programmes stall, the pattern is remarkably consistent — and it is almost never the machine. It is the environment the machine was dropped into.

01

The apron was never mapped to survey grade. An autonomous vehicle navigates against a map. CAD drawings and stand plans are not one.

02

Positioning is accurate to five metres. Autonomy needs centimetres, in rain, at night, next to a moving aircraft.

03

The safety case has no author. Nobody in the building has written one for an autonomous system, so the vendor writes it — which makes it a liability rather than an asset.

04

The regulator has no precedent and no one opened the conversation early enough for one to be created.

05

The unions found out when the vendor arrived. Sequence, here, is the entire game.

06

Nobody owns the gap between "the pilot worked" and "this is now an operation." So nothing crosses it.

The failure state, named

Pilot purgatory

A portfolio of trials, most of which technically worked, none of which became an operation. It happens because each pilot had to build its own connectivity, its own positioning fix, its own integration and its own safety argument — so the third project cost roughly what the first one cost. In a real platform, each deployment is cheaper than the last. When your marginal cost is flat, you don't have a platform. You have unrelated projects sharing a budget line.

The framework

Six domains, split into the ones that matter differently

Four operational domains describe where physical AI actually lands. Two enabling domains decide whether anything that lands there ever scales. The split is the framework's argument, and the weighting is deliberate.

Where the machines land

60%

Four operational domains

A

Airside & ramp autonomy

Apron mapping, GSE telematics, turnaround capture, AV operating envelopes, ramp safety data.

B

Terminal & passenger autonomy

Robot-traversable buildings, flow sensing, service robots, PRM automation, checkpoint automation.

C

Baggage, cargo & logistics

BHS data openness, robotic handling, bag-level tracking, cargo automation, inter-facility transport.

D

Infrastructure & airfield integrity

Asset registers, drone operations, automated FOD detection, predictive maintenance, digital twin.

What lets them land

40%

Two enabling domains

E

Data, connectivity & digital backbone

Airside network coverage, precision positioning, integration architecture, data governance, cyber-physical security.

F

Governance, workforce & safety assurance

Regulatory pathway, safety-case capability, labour engagement, funding and procurement, stage-gate ownership.

Two domains carrying 40% of a six-domain index is intentional. An autonomous machine is a networked, positioned, integrated, governed thing — remove any one and it does not operate. The consequence is that an airport with impressive deployments and weak foundations cannot score highly. That is the correct behaviour: such an airport is not ready, it is exposed.

How it works

Ten minutes, thirty questions, one honest answer

Step one

Describe the airport

Passenger volume, operating model, your role. Scale changes what "ready" means, so results are read against a reference band for airports of your size.

Step two

Answer thirty criteria

Five per domain, each with five concrete level descriptors rather than agree-or-disagree scales. You pick the one that describes your operation. No numbers to look up.

Step three

Get the diagnosis

A weighted score, a maturity band, your enablement deficit, and a prioritised sequence of moves — printable, and written to survive a leadership meeting.

The report

A score is a headline. The report is the argument.

Everything is generated from your own answers. Nothing is generic.

Domain-by-domain diagnosis

What each score means for your operation specifically, and the move your answers point to in each domain.

Your enablement deficit

The gap between what you deploy and what supports it — the metric that distinguishes two airports with identical scores and opposite problems.

Three highest-leverage moves

Ranked by weighted gap, not by lowest score. A level-1 criterion in a heavily weighted domain outranks a level-0 in a light one.

A 90-day / 12-month / 36-month sequence

Matched to your maturity band. Order matters more than speed, and most airports attempt these in the wrong order.

Position against your size tier

Read against a reference band for airports of your scale — clearly labelled as a modelled expectation, not a survey mean.

A printable document

Save it as a PDF and circulate it internally. It is designed to be read by people who were not in the room when you filled it in.

Who built this

An independent framework, published openly

I'm Jinger Jiang. I work on physical AI for airports — translating what is genuinely happening in robotics and autonomous systems into decisions airport leadership teams can actually make. Not vendor selection: the question of what an airport has to become before any of it works.

This framework is free and published in full, including its methodology, its weighting rationale and its limitations. It is not affiliated with any regulator, trade body or supplier, and it does not assess or recommend specific products.

If your result raises questions the report can't answer — which of your gaps is actually binding, what your authority is likely to accept, where a first deployment should go — that's a conversation worth having.

What this is not

  • A vendor comparison or procurement guide
  • A digital transformation maturity model with robots bolted on
  • An argument that every airport should automate
  • A survey whose benchmark bands are presented as measured data

Find out where you actually stand

Answer as your airport is today, not as it is planned to be. The index is only useful if it is honest.

Start the assessment →