Pony.ai and Verne have begun carrying passengers in fully driverless test rides on public roads in Zagreb, Croatia, without an onboard safety operator. The trial uses a 22-kilometer route connecting Zagreb with Zagreb Airport, creating a notable European test case for autonomous transport on a practical city-to-airport corridor rather than a short, tightly contained demonstration loop.[1]
The distinction matters. These are test rides, not an unrestricted commercial robotaxi launch, and the announcement does not establish a timetable for public availability at scale. But operating without a safety driver on an airport link raises the bar beyond vehicle autonomy alone: it requires an operational model that can handle public-road traffic, passenger service, roadside incidents, route constraints and coordination with local authorities. It is evidence that driverless systems are beginning to move from market-specific pilots toward exportable transport infrastructure.
By the numbers
- 22 kilometers: Length of the reported Zagreb-to-airport test route.
- 100 million+ kilometers: Autonomous-driving distance Pony.ai says its vehicles have accumulated worldwide.
- 40 million+ kilometers: Distance within that total that Pony.ai says was accumulated in fully driverless mode.
A city-to-airport route is a more consequential test
Airport corridors are attractive early use cases for autonomous passenger vehicles because they combine a clear passenger need with repeatable travel patterns. Riders have a defined origin and destination, trips are often longer than neighborhood shuttle journeys, and the service can be designed around known pickup, drop-off and curb-management locations. A successful airport connection can also demonstrate a use case with an obvious alternative to private-car parking, taxi queues and fixed-schedule transit.
At the same time, a 22-kilometer public-road route is not equivalent to a geofenced campus deployment. The vehicle must operate in a changing road environment that can include mixed traffic, vulnerable road users, temporary roadworks, variable weather and airport-area traffic flows. The important milestone is therefore the absence of an onboard safety operator during passenger testing, not simply the fact that an autonomous vehicle is carrying riders.
Pony.ai and Verne have not presented the Zagreb activity as an open-ended consumer service. That limitation is important. A test program can use selected passengers, defined operating conditions, restricted hours, operational support and a narrow service area. Those controls are standard ways to validate a driverless service before an operator attempts broader commercial deployment.
What “fully driverless” does—and does not—mean
In this announcement, fully driverless means there is no safety operator seated in the vehicle to take control during the passenger ride. It does not mean the system operates without human organization. Driverless fleets still require dispatching, vehicle inspection, cleaning, maintenance, customer support, incident response and processes for unusual situations. A remote operations center may monitor fleet status or assist passengers, but such support is not the same thing as a human driver continuously controlling the vehicle.
For an autonomous-driving company, the technical task is to keep the vehicle operating safely within a defined operational design domain: the set of roads, weather conditions, speeds, traffic situations and other constraints for which the system has been validated. The route itself becomes part of the product. High-definition mapping, localization, perception of vehicles and road users, motion planning, redundant vehicle systems and fleet operations all have to work together over repeated trips.
The release does not disclose the vehicle model, sensor configuration, local operating limits, remote-assistance procedures, test-passenger eligibility, regulator requirements or the metrics that will determine whether the trial expands. Those details will be necessary to judge how readily the route can progress from a supervised program into a dependable, scaled transport service.
Pony.ai and Verne’s roles in a cross-border deployment
Pony.ai is supplying the autonomous-driving capability for the Zagreb effort, while Verne is the local partner named in the test program. The pairing is strategically relevant because robotaxi deployment depends on more than a driving stack. A service needs local fleet operations, passenger experience design, vehicle logistics, public-sector coordination and a route that fits a real transport demand pattern.
Pony.ai says its autonomous vehicles have traveled more than 100 million kilometers worldwide, including more than 40 million kilometers in fully driverless operation.[1] Those are company-reported cumulative figures, not independently audited safety results. Distance can indicate operational exposure, but it does not by itself reveal performance on a particular Zagreb route, the frequency of human intervention, incident rates, service availability or how the system behaves in rare edge cases.
Still, the move illustrates a broader industry shift. Autonomous-driving developers that have accumulated experience in their original markets are increasingly seeking partners that can make their systems deployable elsewhere. The challenge is not merely translating software across borders. Local road rules, street design, driving behavior, weather, insurance structures, data practices, labor considerations and municipal acceptance can all alter the economics and feasibility of a service.
Market implications: infrastructure, not just a fleet pilot
A driverless airport route could become a useful template if it can sustain safe, reliable operations. Airport trips are operationally valuable because demand is visible, riders often accept a premium for convenience, and each trip can be long enough to improve vehicle utilization. For cities, an autonomous service could eventually complement rail, buses and conventional taxis during off-peak periods or in areas where adding frequent fixed-route transit is difficult.
The near-term business case remains unproven. Autonomous vehicles carry substantial costs for hardware, validation, maintenance, insurance, teleoperations infrastructure and local fleet staff. Removing the onboard safety operator is significant because it is necessary for a robotaxi model to improve its unit economics, but it is not sufficient. The service must also achieve high vehicle uptime, effective incident handling and enough passenger demand to spread fixed costs across a large number of paid trips.
European deployment also invites scrutiny over regulatory accountability. Policymakers and the public will reasonably ask who is responsible after a collision, what information is available to investigators, how emergency services interact with the vehicles, and whether the system performs consistently for pedestrians, cyclists and people with disabilities. The companies’ announcement is evidence of permission to test; it should not be treated as a definitive answer to those broader governance questions.
What to watch as the Zagreb tests continue
The most meaningful next indicators will be operational rather than promotional. The companies will need to show whether the route runs across a widening range of conditions, how passengers are selected and supported, whether the operating area or hours expand, and what path exists toward commercial authorization. Public reporting of fleet availability, cancellations, safety events and accessibility provisions would make it easier to assess real-world readiness.
Zagreb is also a test of whether the robotaxi sector can establish repeatable deployment partnerships beyond its earliest markets. If the program advances, other airport corridors may become logical candidates: they offer bounded service geography without being isolated from the complexity of public roads. If it stalls, the reasons will be equally informative, whether they concern regulation, local operations, economics or the technical limits of a tightly defined autonomous-driving domain.
Editor’s Take
I see the Zagreb airport link as more important than another announcement about an autonomous vehicle completing rides in a limited area. A city-to-airport trip is recognizable transport work: passengers value reliability, luggage handling, predictable pickup and an actual connection to a major destination. Running it without an onboard safety driver is the operational threshold that makes the economics worth testing seriously.
The hype would be getting ahead of the facts if this were described as a broadly available Zagreb robotaxi service. It is not; it is a public-road passenger test. What I would watch is whether Pony.ai and Verne publish evidence of dependable daily operations and broaden the conditions under which the service runs. If they can do that, the real product is not only the vehicle software—it is a replicable airport-route deployment playbook for European cities.
