Zagreb now has cars driving its streets with nobody behind the wheel. Croatian mobility company Verne confirmed this week that it has moved from supervised robotaxi rides to fully driverless passenger trips on public roads, using self-driving hardware built by Chinese autonomous-vehicle company Pony.ai. The shift matters less for the ride itself and more for what sits inside the car: a seventh-generation sensor and compute stack that Pony.ai has now cleared for use without a human safety operator anywhere in the vehicle.
The announcement lands at a moment when robotaxi hardware, not software alone, has become the thing separating companies that can scale from companies stuck running pilot programs. Verne CEO Marko Pejković said the no-safety-driver trials began in August, with Pony.ai and Verne jointly confirming the program publicly on September 10. The test route runs 22 kilometers (14 miles) between Verne’s headquarters and Zagreb’s Franjo Tuđman Airport, a corridor the companies plan to widen over the coming months.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What Changed: From Safety Driver to Empty Front Seat
Verne’s robotaxi service in Zagreb is not brand new. The company launched a commercial ride-hailing product in April, bookable through the Verne app or Uber, priced at €1.99 per ride. That version kept a human safety operator in the car as a backstop, a standard step for autonomous-vehicle operators easing into a new market and regulatory environment. The trials confirmed this month strip that backstop out entirely on the designated test route, making Zagreb one of a small number of European cities where a robotaxi can legally carry a fare-paying or invited passenger with zero human supervision inside the vehicle.
That distinction, driver-assisted versus driverless, is the entire story here. Plenty of companies run supervised robotaxi pilots. Far fewer have cleared the regulatory and technical bar to remove the safety operator altogether. Reports from TechTarget describe the Zagreb trials as having secured the approvals required under Croatian regulations covering autonomous vehicles carrying passengers on public roads, a regulatory sign-off that few European jurisdictions have granted to date.
Several outlets covering the story, including ADAS & Autonomous Vehicle International and Gasgoo, frame this as a milestone for Europe’s robotaxi market broadly. Whether it is definitively the first fully driverless commercial robotaxi operation on the continent is a claim made by the companies themselves and repeated across coverage, but it has not been independently verified against every rival program running in Europe. What is independently confirmed is that the trial is live, it is driverless, and it is carrying passengers on a fixed route right now.
The Hardware Behind the Trial: Pony.ai’s Seventh-Generation Robotaxi
Pony.ai supplies the autonomous-driving technology that Verne’s fleet runs on, and the vehicles used in the driverless Zagreb trials are built on Pony.ai’s seventh-generation Robotaxi platform. That generational detail is the real hardware story. Each new generation of Pony.ai’s platform has historically brought down the cost and footprint of the sensor suite, shrinking the LiDAR units, radar arrays, and onboard compute that make up the physical difference between a car that needs a human backup and one that does not.
Pony.ai has not published the specific component-level specs for the seventh-generation Robotaxi hardware package through verified channels at the time of this report, so figures like sensor count, compute TOPS, or exact battery range should be treated as unconfirmed until the company publishes them directly. What is established is that this hardware generation is the one Pony.ai and Verne jointly selected to carry the no-safety-driver designation, meaning it cleared whatever internal reliability and redundancy bar the companies set for removing the human backstop. Reports have also referenced a possible underlying vehicle platform tied to BAIC’s Arcfox Alpha T5, but that detail has not been confirmed in the primary reporting on this story and should be read as unverified.
For readers tracking the hardware side of autonomous vehicles the way they’d track a GPU generation or a new SSD controller, the signal to watch is redundancy. Fully driverless operation at any scale requires duplicated compute paths, duplicated power delivery, and duplicated braking and steering actuation, so that a single component failure cannot leave the vehicle without a way to stop safely. Pony.ai’s move to remove the safety driver in Zagreb implies its seventh-generation stack has cleared that bar for at least this specific route and operating condition set.
Why Zagreb, and Why This Route
The 22-kilometer corridor between Verne’s headquarters and Franjo Tuđman Airport is a deliberate choice. Airport routes are common first deployments for autonomous-vehicle operators because they combine predictable traffic patterns, consistent demand, and a controlled set of pickup and drop-off points, all of which reduce the number of unpredictable scenarios the driving system has to handle. It is the same logic Waymo used when it began expanding airport access in US markets, and the same reasoning behind many robotaxi pilot corridors globally.
Verne and Pony.ai have said the driverless route will expand over the following months toward Verne’s wider Zagreb operating area, the same zone already served by the supervised commercial product launched in April. That phased approach, prove it on one fixed corridor, then widen the operating zone, mirrors how nearly every robotaxi operator globally has approached removing the safety driver. Nobody has gone from zero to citywide driverless coverage in one step, and Verne’s plan does not appear to be an exception.
Verne’s Timeline: April Launch to August Trials to September Announcement
The sequence of events matters for understanding how fast this moved. Verne’s commercial service, with a human safety operator aboard, opened to the public in April at €1.99 per ride, bookable through the Verne app or through Uber’s platform. Four months later, according to CEO Marko Pejković, the company began the first fully driverless passenger trials in August. Pony.ai and Verne then jointly confirmed those trials publicly on September 10, and coverage of the milestone has continued through mid-October as outlets outside the companies’ own channels picked up the story.
That is roughly five months from a supervised commercial launch to a confirmed, safety-driver-free passenger trial on public roads. For context, it took Waymo years of supervised testing in Phoenix and San Francisco before it removed safety drivers from its commercial service, and even then the removal happened gradually, neighborhood by neighborhood. Verne’s compressed timeline is partly a function of starting later and inheriting a more mature hardware platform from Pony.ai, rather than developing the self-driving stack in-house.
| Milestone | Date | Safety Driver | Access |
|---|---|---|---|
| Commercial launch, Zagreb | April 2026 | Yes | Verne app, Uber app |
| Driverless trials begin | August 2026 | No | Passenger trials (public booking unconfirmed) |
| Public confirmation of driverless trials | September 10, 2026 | No | Joint Pony.ai / Verne announcement |
| Planned route expansion | Following months | No | Wider Zagreb operating area |
| Expected public driverless availability | 2027 (per CEO) | No | Zagreb, then other European cities |
Pony.ai’s Role: A Chinese Autonomous-Driving Supplier Inside a European Market
The arrangement between Verne and Pony.ai splits the business cleanly: Pony.ai supplies the autonomous-driving technology, and Verne operates the service in Croatia. That structure, a domestic mobility operator licensing self-driving hardware and software from an outside technology supplier, is becoming the default model for robotaxi expansion outside the United States and China, where the handful of companies with mature full-stack autonomy (Waymo, Pony.ai, Baidu’s Apollo Go, and a few others) license their platforms to local partners rather than building out operations city by city themselves.
Pony.ai’s decision to anchor one of its first fully driverless deployments outside China and the US in Croatia specifically is notable on its own. It places Chinese autonomous-vehicle hardware inside the European Union’s regulatory perimeter at a moment when EU policymakers have been actively scrutinizing Chinese-made hardware and software across other categories, a pattern already visible in how the EU has treated Chinese AI chip suppliers gaining share against Nvidia. Verne operating as the public-facing brand, with Pony.ai’s technology running underneath, is a structure that may face more regulatory attention as the service scales, even though the current trial has reportedly cleared the Croatian approvals required for autonomous passenger operations.
How This Compares to Robotaxi Hardware Elsewhere
Zagreb’s driverless trial does not exist in isolation. Robotaxi hardware platforms are now running, in various supervised and unsupervised configurations, across multiple continents, and the gap between “supervised pilot” and “safety-driver-free commercial service” is where most of the competitive separation is happening. Waymo remains the clearest leader on raw driverless fleet size and track record in the US, while Tesla has been scaling a much newer robotaxi operation built on its own camera-first hardware stack, and regulators in the US have already opened a probe into Tesla’s robotaxi fleet over behavior near school zones. Baidu’s Apollo Go continues to expand driverless coverage inside China, and even passenger-van-style autonomous hardware is entering the mix, as seen in the recent comparison between PIX’s RoboBus and Tesla’s Robovan.
| Operator | Autonomy Supplier | Region | Safety Driver Status |
|---|---|---|---|
| Verne (Zagreb trial) | Pony.ai (seventh-gen Robotaxi) | Croatia, EU | Removed on 22km trial route |
| Waymo | In-house (Waymo Driver) | US, planned Canada expansion | Removed, multi-city |
| Tesla Robotaxi | In-house (FSD-derived stack) | US, multi-city rollout | Mixed, supervision varies by market |
| Baidu Apollo Go | In-house (Apollo platform) | China, multi-city | Removed, multi-city |
| Pony.ai (direct operations) | In-house | China, parts of US testing | Removed, limited cities |
For readers following Tesla’s own robotaxi expansion, the contrast is instructive. Tesla has been scaling its fleet in Austin and other US cities, with coverage on sites like this one tracking the current robotaxi tally against Waymo and the underlying hardware inside vehicles like the Cybercab fleet in Austin. Tesla’s approach leans on camera-only perception rather than LiDAR, a hardware philosophy that differs meaningfully from Pony.ai’s sensor suite. Waymo, meanwhile, has been pushing its own driverless hardware north, with plans reported for a first Canadian robotaxi deployment in Toronto targeted for 2028. Verne’s Zagreb trial effectively adds a fourth distinct hardware and regulatory path to that global map.
Regulatory Path: How Croatia Cleared a Driverless Robotaxi
Autonomous-vehicle regulation across the EU has generally moved slower than in the US or China, with individual member states setting their own frameworks for testing and deploying self-driving vehicles on public roads. Croatia’s approval of Verne’s driverless trial suggests the country built out, or adapted, a regulatory pathway specifically to accommodate a no-safety-driver robotaxi operation, something few EU member states have done publicly to date.
The approval appears scoped narrowly, to the specific 22-kilometer test route rather than a blanket citywide clearance, which tracks with how most jurisdictions have handled early driverless approvals elsewhere. It is a geofenced permission, not an open license, and the plan to expand gradually toward Verne’s wider Zagreb zone suggests future route additions will likely require their own incremental sign-off from Croatian authorities rather than one sweeping approval covering the whole city at once.
What Pejković and Pony.ai Have Said
Verne’s leadership has been public about its ambitions since the April commercial launch. CEO Marko Pejković said at the time, “For the first time in Europe, there is a real commercial robotaxi service. People can use it and take real autonomous rides,” according to a statement published on Pony.ai’s blog. He added at the same launch, “We said we would launch in Zagreb in 2026. Today, we did,” underscoring that the company had been working toward this specific timeline for some time before the April debut.
Pony.ai founder and CEO Dr. James Peng framed the Zagreb partnership in global terms, saying, “Partnering with Uber and Verne represents an important milestone as we continue to expand autonomous mobility globally,” in a statement carried on Uber’s investor relations site. Pejković separately described the broader goal behind Verne’s approach: “Europe needs autonomous mobility that can move from testing to a real service,” and added that Verne is “bringing together the technology, platform, and operational capabilities required to make this a reality, starting in Zagreb before expanding to new markets,” per the same Uber release. Those quotes predate the September confirmation of the fully driverless trials specifically, but they establish the stated strategic intent, scaling past Zagreb, that the current driverless milestone is a step toward. Pejković has separately said he expects the fully driverless service to become publicly available next year, with plans to eventually expand to other European cities.
Market Impact: What This Means for European Mobility
A single 22-kilometer corridor in one Croatian city will not reshape European transportation on its own, but it establishes something competitors will now have to respond to: a working precedent for safety-driver-free robotaxi operation inside the EU regulatory framework. Any other European operator pursuing a similar milestone, whether that is a Waymo expansion, a European automaker’s own autonomy program, or another Pony.ai-style licensing deal, now has a reference point for what Croatian-style approval looks like and how long it took Verne to get there.
It also puts pressure on the “Europe’s first” framing that has circulated in coverage of this story. If the claim holds, Verne and Pony.ai gain a meaningful marketing and regulatory-credibility advantage heading into future city-by-city expansion talks across the continent. If a rival surfaces with an earlier or broader driverless claim, that advantage shrinks fast. Either way, the Zagreb trial sets a visible bar that other operators will be measured against through the rest of 2026 and into 2027.
Historical Context: How Robotaxi Hardware Got Here
Removing a human safety driver from a commercial vehicle has taken the industry roughly a decade of hardware iteration to reach a point where regulators will sign off on it anywhere. Waymo’s earliest fully driverless rides in Arizona happened years after its first supervised pilots, and even then the company expanded coverage block by block rather than all at once. Baidu’s Apollo Go followed a similar multi-year path inside China before reaching its current multi-city driverless footprint.
Pony.ai’s own hardware lineage reflects that same pattern of generational iteration, with each successive Robotaxi generation aimed at shrinking sensor and compute costs while improving reliability enough to justify removing the human backstop. The seventh-generation platform now running in Zagreb represents the current endpoint of that iteration, at least for the markets Pony.ai has chosen to deploy it in. Verne’s compressed five-month path from supervised commercial launch to driverless trial only looks fast because it is standing on top of hardware and software maturity that Pony.ai spent years building before Zagreb entered the picture at all.
What’s Still Unconfirmed
Several details circulating around this story remain unverified and should be treated with caution. It is not independently confirmed whether the fully driverless trials are open for general public booking or restricted to invited passengers, with at least one report specifying the latter. The exact vehicle make and model underlying Pony.ai’s seventh-generation Robotaxi platform, including a reported link to BAIC’s Arcfox Alpha T5, has not been established in primary reporting. And the “Europe’s first” claim attached to this milestone comes from the companies themselves and from outlets repeating that framing, not from independent verification against every other driverless program operating in Europe.
Predictions: Where the Zagreb Trial Goes Next
Based on the companies’ stated timeline and the broader pattern of how driverless rollouts have unfolded elsewhere, a few outcomes look likely over the next 12 to 18 months.
- Verne will likely widen the driverless route incrementally within Zagreb rather than opening the whole city to no-safety-driver rides at once, matching the pattern CEO Marko Pejković has already described publicly.
- Pejković’s stated target of public driverless availability in 2027 is plausible given the current trial’s progress, though regulatory timelines in the EU have a track record of slipping past initial company estimates.
- Pony.ai will likely use the Zagreb deployment as a reference case to pitch additional European mobility operators on licensing its Robotaxi hardware and software stack, especially in EU cities without a strong domestic autonomy program of their own.
- Scrutiny of Chinese-supplied autonomous-vehicle hardware operating inside EU infrastructure is likely to increase as the service scales past a single test corridor, mirroring the regulatory attention already directed at Chinese hardware in other sectors.
- Rival operators in Europe, including any projects tied to Waymo or European automakers, will face pressure to produce their own driverless (not just supervised) milestones to counter the “Europe’s first” narrative Verne and Pony.ai have established.
The Bigger Hardware Story
Strip away the headlines about a first-of-its-kind European milestone, and what is actually new in Zagreb is a hardware generation clearing a trust threshold it had not cleared before, at least not in this market. Pony.ai’s seventh-generation Robotaxi platform is now operating without a human fallback on a fixed route, which is a different and higher bar than running supervised pilots anywhere in the world. That is the detail worth tracking as this story develops: not whether Zagreb gets more driverless miles added to its map over the coming months, but whether Pony.ai’s hardware, and whatever comes after the seventh generation, keeps clearing that bar in new regulatory environments without a safety-driver fallback built back in.
How this plays against other ongoing autonomy stories matters too. Tesla’s robotaxi program has faced its own hardware and liability questions in the US, including scrutiny after incidents involving its driver-assistance systems, while Waymo continues to expand its own driverless hardware footprint internationally. Zagreb adds a third major hardware lineage, Pony.ai’s, to a global conversation that until recently was mostly a two-way comparison between Waymo and Tesla.
Frequently Asked Questions
What is Verne, and what does it have to do with Pony.ai?
Verne is a Croatian autonomous-mobility company that operates a robotaxi service in Zagreb. Pony.ai, a Chinese self-driving technology company, supplies the autonomous-driving hardware and software that Verne’s vehicles run on. Verne operates the service; Pony.ai provides the underlying self-driving technology.
Is the Zagreb robotaxi really driving with no human inside?
Yes, on the specific 22-kilometer trial route between Verne’s headquarters and Franjo Tuđman Airport, the vehicles operate without a safety driver in the car. This is a step beyond Verne’s commercial service launched in April, which used a human safety operator as backup.
Can anyone book a ride in the fully driverless trial right now?
That is not independently confirmed. The trials are described in available reporting as passenger trials, with at least one report specifying invited passengers rather than open public booking. Verne’s existing commercial service, which still includes a safety operator, remains bookable through the Verne app or Uber.
What hardware is inside these robotaxis?
The vehicles run on Pony.ai’s seventh-generation Robotaxi platform. Pony.ai has not published detailed component-level specs for this generation through verified channels, so specific sensor counts or compute figures remain unconfirmed. A reported link to BAIC’s Arcfox Alpha T5 as the underlying vehicle platform has also not been independently verified.
Is this really Europe’s first fully driverless robotaxi service?
Verne and Pony.ai have made that claim, and several outlets have repeated it, but it has not been independently verified against every autonomous-vehicle program operating in Europe. It is accurate to describe the Zagreb trial as a confirmed, currently operating fully driverless passenger service.
When will the driverless service be available to the general public?
Verne CEO Marko Pejković has said he expects the fully driverless service to become publicly available next year, with plans to eventually expand to other European cities. That timeline has not been independently confirmed by Croatian regulators and should be treated as a company projection rather than a guaranteed date.
How does this compare to Tesla’s and Waymo’s robotaxi hardware?
Waymo and Tesla both build their autonomous-driving hardware in-house, while Verne licenses its technology from Pony.ai. Waymo and Tesla also operate primarily in the US, with Waymo additionally planning expansion into Canada, while Verne’s driverless trial is currently confined to a single route in Zagreb, Croatia.
What happens if the Zagreb trial runs into a safety incident?
No incident has been reported in connection with the driverless trials as of this writing. Croatian regulators would be expected to review any such incident under the approval framework that cleared the trial, though the specifics of that process have not been detailed in available reporting.
