DiDi R2 Robotaxi Starts Driverless Trials in Beijing, Guangzhou

What happened

DiDi Autonomous Driving has put its next-generation R2 robotaxi into fully driverless service trials. Riders can book the car through the main DiDi app in designated demonstration zones in Beijing and Guangzhou.1 Coverage dates the start of the trials to August 31, 2026, and the announcement was issued from Dubai.2 Weeks later, the story is still circulating in video explainers and supplier press releases, which suggests it is being read as a milestone rather than a one-day product note.623

The R2 was co-developed with carmaker GAC Aion. DiDi calls it purpose-built for robotaxi work and designed for deployment outside China.1 The two companies set up a joint venture in April 2024. When the R2 was delivered on January 23, 2026, it was described as the first robotaxi model in China launched through a joint venture between an autonomous-driving company and an automaker.2 The early fleet received intelligent connected vehicle road-test licenses in Guangzhou. Since delivery, the car has been tested on public roads in Guangzhou, Beijing, Shenzhen and other cities.2

On hardware, the reports agree. The R2 has 33 sensors: lidar, cameras, 4D millimeter-wave radar, infrared cameras and audio sensors. They feed a triple-domain fusion central computer with more than 2,000 TOPS of compute.2 Braking, steering, power, communications and software all have redundant backups. DiDi also says the vehicle meets five-star C-NCAP and Euro NCAP standards.12 Thermal-camera supplier Raytron says its module can pick out vulnerable road users up to 300 meters away. DiDi Autonomous Driving CEO Zhang Bo described a test in Yizhuang where the infrared sensor spotted a person lying by a dimly lit road at 100 meters, after cameras missed him and lidar returned only sparse points.23

Inside, the car has a 17.3-inch ceiling screen and a voice assistant. Riders can use it to confirm their PIN, start the trip and control the climate, windows and lighting. More features, such as an exterior display and horn-assisted vehicle finding, will be added later.1

Where the reporting diverges

The accounts differ on one figure: the 74 percent cost cut. One explainer says the central computer costs 74 percent less than the previous generation.6 A more detailed account ties the cut to the platform as a whole and lists it alongside lower latency and doubled sensor capacity.2 The difference matters because cheaper compute is not the same as a cheaper vehicle. Until DiDi publishes per-unit costs, the safer reading is that the saving applies to the computing and sensor-integration stack, not to the whole car.

The second gap is about what "service" means. DiDi's own wording is "fully driverless service trials" in "designated demonstration areas."1 None of the coverage mentions paid fares, fleet size or ride volume. One industry tracker lists DiDi's deployment as limited and mixed, covers two public cities, and has no figures for kilometers, fleet or geofence.22 DiDi already runs a 24/7 driverless pilot in Guangzhou's Huangpu district, launched in December 2025, where riders can request a robotaxi or a mixed order that dispatch fills based on traffic and supply.2 Read together, the R2 launch puts new hardware into a dispatch model DiDi already uses. It is not yet a commercial rollout at scale.

Why DiDi's position is different

DiDi's advantage is demand, not miles driven. The company handles about 40 million rides a day in China, and almost all of them still have a human driver.6 A robotaxi bookable inside that app reaches a customer base that Baidu's Apollo Go, Pony.ai and WeRide have to build or rent. That makes the R2 more strategically important than its small, geofenced footprint suggests.

DiDi is still behind on deployment, though. Apollo Go says it operates in 28 cities.6 Pony.ai reported about 1,975 commercial vehicles as of mid-2026, and WeRide about 1,800.22 Even so, one composite index of robotaxi players shows DiDi among the biggest recent gainers, second only to XPeng in its latest movement.29 That fits a company whose operational numbers are still undisclosed but whose platform could turn hardware into rides quickly.

The ride-hailing companies are converging on robotaxis from the platform side. DiDi, Caocao Mobility and Ruqi Travel already handle dispatch and fleet management, and robotaxis and driverless minibuses do the physical transport.24 That layer may itself be squeezed if AI assistants on phones and wearables become the first place travel requests land.24 Taken together, this suggests DiDi is racing to own the vehicle layer before the order layer slips away from it.

The regulatory backdrop: a freeze, then a thaw

The timing of the R2 trials is best explained by regulation. In spring 2026, dozens of Baidu Apollo Go robotaxis stopped abruptly in Wuhan traffic. Beijing then suspended new autonomous-vehicle licenses nationwide, which blocked fleet additions, new pilots and expansion into new cities.3839 Most accounts place the Wuhan outage in March, and one puts it in April.3337 Neither Baidu nor regulators have published a root cause.33

Permits started flowing again gradually after an industry-wide review finished in late June. Momenta received one of the first new permits, for Shenzhen.3335 Some reports say the restart came with tougher requirements on data reporting and system redundancy.37 On that reading, the redundancy DiDi stresses in its R2 materials is a compliance feature as much as an engineering one.2 Beijing granted Baidu and Pony.ai commercial licenses to charge fares in Yizhuang from August 21. Those cars carry an employee in the front passenger seat, ready to step in.34 DiDi's driverless trials started ten days later. They look like part of a wave of approvals rather than a lone breakthrough.

The national framework is also tightening. In February, MIIT published a draft mandatory safety standard for Level 3 and Level 4 systems, due to take effect July 1, 2027. It requires L4 vehicles to drive without human intervention while allowing remote assistance. It also requires them to yield to emergency services and to carry a black-box data recorder.36 China still has no single national autonomous-driving law, only a patchwork of city rules, test permits and pilot zones.32 This was at least the second time Beijing froze approvals: an earlier pause in late 2024 followed complaints from taxi drivers about losing work. For DiDi, whose business depends on millions of human drivers, that labor issue is especially sensitive.

The Waymo comparison

In the United States, Waymo is scaling at a pace no Chinese operator has publicly matched in one market. On September 1, it opened service in Denver, San Diego and Tampa, bringing it to 14 U.S. cities with a fleet of more than 4,000 vehicles.11 By early October, coverage put it at 15 markets, backed by a $5 billion loan, its first debt financing.12 Weekly ride figures vary by source, from roughly 400,000 to 500,000 paid rides, against a target of 1 million a week by year-end.1315 California regulators approved its largest expansion so far in August, covering 18 counties, freeways and night driving in rain and fog.16

The two systems show the same pattern. Waymo's Ojai minivan, built by Zeekr, is cheaper to build and run, and it is the only vehicle offered in Denver and San Diego.11 DiDi's R2 makes the same bet: a vehicle designed for the job and built on a production line will beat retrofitted cars on cost. U.S. regulators are not standing still either. NHTSA and the NTSB are investigating Waymo cars that illegally passed stopped school buses.12 California passed a law allowing fines for robotaxis that obstruct emergency responders.15 Minneapolis's council voted to require human drivers in robotaxis, and the mayor vetoed it.15

The bottom line

The R2 launch matters less for what it is today, an unpriced trial in a few demonstration zones, than for what it shows. China's licensing pipeline is open again after the Wuhan scare. Hardware costs are falling sharply. And the country's largest ride-hailing platform now has a factory-built driverless car it can send through its own app. The test is whether DiDi starts publishing fleet sizes and paid-ride numbers the way Waymo does, and whether it can grow those numbers under a mandatory national safety standard arriving in 2027. Until then, DiDi holds the biggest potential demand in Chinese robotaxis and some of the least disclosed operations.