Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseInsight · On RadarArs Technica→ original

Waymo pulls robotaxis off highways after new entries into work zones

Waymo has pulled its robotaxis off highways. California's DMV demanded a halt to highway trips because the cars again entered road-work zones at speed. Similar cases occurred in the spring in Phoenix and California. In June the company recalled nearly 3,800 vehicles, and in July it returned to highways with updated software. New incidents on US 101 showed the patch did not help. The takeaway: temporary work zones remain a weak point for autonomous vehicles. The service continues to operate on regular streets: more than 500,000 rides per week. Highway service is suspended in Los Angeles, the Bay Area and Arizona. In April in Phoenix, robotaxis failed to recognize exit-closure signs six times.

AI-processed from Ars Technica; edited by Hamidun News
Listen to article

Waymo's driverless taxis, considered the most mature on the market, are once again off highways. The company has temporarily limited its fleet to regular city streets after California authorities noticed a dangerous habit in its cars: entering road-work zones at speed. For anyone building autonomous systems, this is a lesson in how expensive the "long tail" of rare situations is.

What happened

According to the San Francisco Business Times, which Ars Technica cites, California's DMV in early November ordered Waymo to immediately stop rides on the state's highways. The condition for resuming: the company must convince the regulator that its cars no longer pose an "immediate threat to public safety."

The trigger was recent cases. In early November, a video spread widely showing a robotaxi on I-280 driving past two California Highway Patrol cars. The patrol cars were running a "rolling roadblock," holding back traffic behind slow construction equipment. In addition, according to authorities, one Waymo vehicle entered a work zone on US 101 near Burlingame. The company stresses that there were no injuries and no collisions. Waymo later temporarily suspended highway service in its other cities as well, in order to check its own performance metrics.

This is not the first time

The problem has recurred since spring. In April in Phoenix, robotaxis failed to recognize exit-closure signs six times and drove into work zones. In May in California, seven vehicles entered lanes with active repairs. In June, Waymo recalled nearly 3,800 vehicles capable of highway driving. Here, a recall means a software update, not vehicles being returned to the factory. Until the end of July, the company operated in California only on city streets.

In July, Waymo told the NHTSA regulator that it had updated its software. The new version better recognizes highway work zones, prioritizes them more strongly over other road hazards, and reduces speed when it suspects it is in such a zone. The response to emergency services was also improved. The company added "operational measures" and expanded remote assistance, in which a human operator advises the vehicle on what to do. The new incidents show that this was not enough.

Why it matters

A repair zone looks simple, but for an algorithm it is a set of unstable signals. Cones are not placed according to a standard pattern, signs are often temporary, lanes are closed in nonstandard ways, and sometimes a patrol car serves as the traffic director. High highway speeds leave little time to react. Such situations are rare, so there are few of them in training data. Fixing one specific scenario, such as exit-closure signs, does not guarantee the system will handle a neighboring, slightly different case.

The business cost is already visible. Waymo completes more than half a million rides a week, and the service continues to run on regular streets. But highways provide speed and revenue on long routes, and until the regulator's trust is restored, they are unavailable. Before the suspension, Waymo drove on highways only in Los Angeles, the San Francisco Bay Area and Arizona, so the loss affects all three markets.

What follows

First, a software update does not equal solving the problem: regulators look at real on-road behavior, not statements about a patch. Second, any autonomous system, whether a robotaxi or a production agent, requires measurable safety criteria and a clear rollback procedure. Third, companies planning highway launches will have to show regulators statistics specifically on work zones.

The restriction is temporary for now, and there is no timeline for returning to highways. Waymo says it is working with the DMV and other state agencies. It is also known that the incidents involved no injuries, and Waymo emphasizes this point. But the recurrence of incidents after the recall and patch shows that a single software fix does not settle the matter.

Story

Key Questions Answered

What happened?

According to the San Francisco Business Times, which Ars Technica cites, California's DMV in early November ordered Waymo to immediately stop rides on the state's highways. The condition for resuming: the company must convince the regulator that its cars no longer pose an "immediate threat to public safety." The trigger was recent cases. In early November, a video spread widely showing a robotaxi on I-280 driving past two California Highway Patrol cars.

What are the key details on This is not the first time?

The problem has recurred since spring. In April in Phoenix, robotaxis failed to recognize exit-closure signs six times and drove into work zones. In May in California, seven vehicles entered lanes with active repairs. In June, Waymo recalled nearly 3,800 vehicles capable of highway driving. Here, a recall means a software update, not vehicles being returned to the factory. Until the end of July, the company operated in California only on city streets.

Why it matters?

A repair zone looks simple, but for an algorithm it is a set of unstable signals. Cones are not placed according to a standard pattern, signs are often temporary, lanes are closed in nonstandard ways, and sometimes a patrol car serves as the traffic director. High highway speeds leave little time to react. Such situations are rare, so there are few of them in training data. Fixing one specific scenario, such as exit-closure signs, does not guarantee the system will handle a neighboring, slightly different case.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Unified AI Workspace · Nomad

Test Frontier LLMs & Autonomous Agents in One Workspace

Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.

Try Nomad AI →Launch Nomad Agent
What do you think?
Loading comments…