IE 11 Not Supported

For optimal browsing, we recommend Chrome, Firefox or Safari browsers.

The Same Tech Used in Driverless Cars Is Helping Public Works

Cities are turning to the same type of technology that allows robotaxis to navigate roadways to identify potholes and errant drivers. The technology is helping to remove humans from potentially dangerous tasks.

Workers spreading asphalt to repair a pothole
Workers repair a pothole in San Francisco.
(TNS) — By now, stories of driverless taxis disrupting traffic and interfering with crime and fire scenes on San Francisco's streets are legion. Less well known is that the same technologies robot cars use to navigate around town, including cameras and artificial intelligence software, are also being put to use in myriad other, less risky, urban jobs where they save governments time and money.

Towns including Irvine and Oregon City, Ore., for instance, are using smart cameras mounted on street sweepers to spot potholes and notify road crews to fix them. San Francisco startup Hayden AI is helping transportation companies around the country, including the Bay Area's AC Transit, keep bus lanes clear of ne'er-do-well drivers with cameras that spot and report errant vehicles. And researchers at Stanford have developed ways to use Google Street View to spot illegal dumping for speedier cleanups and even to track gentrification.

Though it's ahead of the pack in self-driving cars, San Francisco has been slow to adopt these emerging applications of visual sensing and AI. But Public Works department spokesperson Rachel Gordon said the agency has been discussing ways to use AI to operate more efficiently. It already uses computer mapping to locate and track public works jobs.

The pothole spotter comes from a Canadian startup called CityRover, which sells a camera about the size of a smartphone that can be attached to a street sweeper or other decidedly non-autonomous vehicles already roving city streets.

The camera is trained with an algorithm to recognize potholes in all their varied forms, using machine learning and comparing what it sees to the many examples of the annoying street divots it has already been taught. The software then generates a geotagged report of their locations, making dispatching street crews more efficient than relying on 311 or other sources.

Oregon City, a town of about 40,000 people southeast of Portland, signed a two-year contract with CityRover that cost less than $10,000 and now has cameras mounted on sweepers that cover the city, said Jayson Thornberg, the city's transportation maintenance manager.

"It's way more efficient," said Thornberg, who used to work on road crews patching potholes. Previously, he said, residents would report a pothole and his team would go hunting for the damage to patch it, often wasting time when wrong addresses were included in the file.

Some holes would go unreported for years, he said, worsening as the freeze-and-thaw cycle in cold and rainy Pacific Northwest winters took its toll on the asphalt.

CityRover isn't perfect, Thornberg concedes. He still has to go through the reports the software generates to weed out false positives incorrectly labeled by the bot, such as reflected rainwater or duplicate images.

But that work usually takes him only about an hour, he said, and he still saves his crews time because the software maps out the street damage and suggests the fastest route between repair sites.

Meanwhile, AC Transit is using AI and smart cameras very differently, mounting them on the front of buses to detect when a car is illegally blocking a bus lane or stop. The pilot program was made possible using technology from Hayden AI.

Agency spokesman Robert Lyles said AC Transit conducted a proof-of-concept study using cameras on two of its Tempo buses during a 48-day period and found the technology successfully identified vehicles violating the bus-only lane with an accuracy of more than 99%.

A previous type of camera the agency tested required a driver to press a button to take a photo of a car's license plate and for city staff to send out a ticket. The Hayden system does all of that automatically.

The program was also enabled by the passage of California's AB917, which extended earlier permission for AC Transit to use video imaging to issue traffic citations to vehicles parked in bus lanes.

Lyles said those results went before AC Transit's policy board over the summer and that the agency is now negotiating a contract with Hayden.

A report from New York City's Metropolitan Transit Authority, which also uses Hayden AI's cameras, said that as of March it had issued 224,000 violation notices since its camera enforcement program began, generating $11 million in revenue. The report noted that commercial vehicles were the most frequent culprits.

Hayden AI spokesperson Jenna Fortunati said in an email that the system is meant to be "revenue neutral" because its monthly cost is almost always offset by revenue from traffic citations.

Fortunati said the company has "introduced our technology with agencies throughout the state of California including ( San Francisco's) SFMTA and look forward to continuing the conversation."

San Francisco Municipal Transit Agency buses don't currently use Hayden AI cameras, according to the company. But some Muni buses are fixed with more conventional forward-facing cameras that take photos of cars parked in transit lanes for citations to be issued. Agency spokesman Stephen Chun did not provide responses to emailed questions about Muni's technology in time for publication.

San Francisco is also accelerating efforts to incorporate technology into other aspects of city governance. Mayor London Breed recently announced plans to put a measure on the ballot allowing the Police Department to install and use publicly owned surveillance cameras as well as drones to aid in catching suspects. If approved, the measure also would let the department test other kinds of technology without Board of Supervisors review.

As AI technology gets better, its uses will continue to evolve.

Researchers at Stanford have been developing algorithms that chew over Google Street View images and detect everything from illegal trash dumping to street conditions to signs of new construction that could signal gentrification.

CityRover is also working on other applications of its visual sensing technology.

The company's technology chief, Roy Tal, said CityRover aims to give its devices the ability to detect illegal dumping sites so that city workers can be dispatched to clean up junk.

"Once you have a full picture of the city, you can make better work planning (decisions)," Tal said. "You know which areas to send your crews to to fix problems proactively."

©2023 the San Francisco Chronicle, Distributed by Tribune Content Agency, LLC.