Autonomous vehicles

Driverless transport is simultaneously a mature, unremarkable technology and a decade-long pilot programme — depending entirely on whether it runs on a guideway or in traffic. Almost every disappointed expectation in this field comes from conflating the two.

The atlas holds 25 documented entries across 19 cities — but only 10 countries and two continents. That concentration in East Asia and the Gulf is real and worth stating plainly: driverless deployment at scale is not currently a global phenomenon.

The distinction that explains everything

Rails: solved, and quietly enormous

Unattended metro operation (GoA4) is not experimental. Dubai's driverless metro has carried over 2.4 billion riders since 2009. Shanghai reported 167 km of driverless route across five lines by the end of 2021. Beijing's Yanfang line proved home-grown signalling in 2017; state media traced from it a wave reaching about 50 fully automated lines across 21 Chinese cities — roughly 1,480 km — by January 2025.

Chengdu's Line 9 contributes the most portable engineering lesson: launching straight at unattended operation, rather than retrofitting automation onto a driver-run line later, let obstacle and derailment sensing be designed into the fleet from the start. Retrofitting autonomy is consistently harder than starting with it.

Astana's LRT supplies the governance lesson. It opened driverless — after a decade of delay and a landmark corruption case in which the abandoned viaduct became a national symbol of graft. Automation was never the hard part of that project.

Streets: ten years of pilots

The contrast in commitment is stark, and both numbers are in the atlas. Beijing's Yizhuang demonstration area reports 38 million kilometres of cumulative autonomous test mileage, 35 test companies and 1,098 test plates as of mid-2025. The flagship EU-funded Sohjoa Baltic shuttle programme, across six cities and 19 months, carried 19,437 passengers in total — its Helsinki pilot 3,932.

Neither figure is embarrassing on its own terms. Together they explain why European cities keep concluding that shuttles "are not ready" while Chinese and Gulf cities report commercial services: the sample sizes differ by four orders of magnitude.

Even the newest street deployments stay bounded. Singapore's Punggol shuttle service runs with safety operators aboard. Tokyo's Marunouchi trial logged 420 rides across 69 runs in five days on a pedestrianised street. Amsterdam's Roboat, which set out to put autonomous vessels on the canals, is graded scaled-back: the research programme ended and the surviving commercial work is far more modest than the original vision.

The evidence problem

Not one street-autonomy entry in this atlas carries independent or peer-reviewed evidence. Robotaxi ride counts come from operators; test mileage comes from the authority promoting the zone; terminal productivity comes from the port and its technology partner, explicitly not independently audited. This is normal for a young field — but it means comparative claims about safety or efficiency currently rest on the interested party's own accounting.

NEOM's The Line marks the outer edge of that problem: a 170-kilometre autonomous-mobility megastructure announced before engineering and financing were proven, now suspended at a 2.4-kilometre stub.

Questions to ask before commissioning

• Guideway or traffic? If it is segregated, this is proven procurement. If it is mixed traffic, it is a research programme with a service brochure.
• New line or retrofit? Chengdu's record argues strongly for designing automation in from the start.
• Is a safety operator on board? Ask what would have to change to remove them, and when.
• What is the passenger count, not the mileage? Test kilometres measure investment; boardings measure usefulness.
• Who audited the figures? In this field the honest answer is usually nobody — plan the evaluation yourself.
• What is the fallback service? Several pilots here ended and left nothing behind; a shuttle route should not be the only service on that corridor.

The practical takeaway: if a city wants driverless transport that carries people at scale today, the answer in this atlas is a metro. Street autonomy is advancing fastest where a single authority created a large fenced testbed and committed for years — and it remains, everywhere, a programme to evaluate rather than a service to buy.