Washington, DC CNN  — 

Jaime Waydo spent years building self-driving cars at Waymo, and was quick to praise their abilities. The cars already had the experience of 300 years of average driving experience, she said in a 2017 interview. Waymo’s self-driving cars could see 360 degrees, she noted, while humans could see about 120 degrees.

“[The cars] can also see up to two football fields away, which is a significantly better vision than what you or I probably have,” Waydo told an interviewer in 2017.

But in a recent interview with CNN Business, Waydo spoke more of the challenges and limits of self-driving cars, like identifying puddles, or seeing clearly when the sun is low on the horizon. She said her years working on self-driving showed her all of the unusual things can happen on the roads, which make it difficult for even advanced artificial intelligence to react appropriately to the myriad new scenarios it might encounter. A Waymo self-driving car once saw a baby crawling on a road, she recalled. (The baby was fine, according to Waydo.)

Waydo, who left her role as lead systems engineer at Waymo in 2018 to work on autonomous systems at Apple, is now joining Cavnue, a startup making digital infrastructure intended to help self-driving cars understand roads, so they can actually drive better than a human.

A US Department of Transportation illustration shows how vehicle to infrastructure technology is designed to function and improve road safety.

Roads, signs and traffic lights were designed when governments expected that only humans would drive cars. A bright green light may be the easiest way to tell a human it’s safe to drive ahead, but a self-driving car may be better off if the information is messaged to them over a radio or cellular network. As the task of developing and rolling out self-driving cars has started to look slower and more difficult than expected, autonomous driving executives like Waydo are increasingly warming up to the idea of retrofitting infrastructure to get them on roads faster. But critics say the technology is unproven and the costs could be expensive for governments already struggling with the cost of simple road upkeep, like filling potholes.

For years, self-driving car companies operated under the belief that their technology and current road infrastructure would be enough, and their cars wouldn’t need any dedicated vehicle-to-infrastructure technology, such as traffic lights that broadcast their color directly to a vehicle’s computer, or cameras mounted at blind turns or on light posts to get crucial information for the cars, like if a cyclist is approaching. But self-driving has proved harder to get off the ground than expected. Companies have missed deadlines for deployments. The industry has consolidated, with even a company as big as Uber selling its self-driving vehicle program. Other startups have shut down or sold themselves to big tech companies.

Jaime Waydo, a longtime self-driving car engineer, has joined Cavnue to develop smart technology to help autonomous vehicles drive safe.

Waydo wants self-driving technology to spread broadly so that people like her father, whom she said lives in a rural area and has Parkinson’s Disease, can benefit from the technology. (Waymo, which is owned by Google parent company Alphabet, operates a self-driving robotaxi service in part of Phoenix, Arizona, but hasn’t expanded it across the country.)

She’s not alone in coming around on the merits of smart infrastructure.

Michael Fleming, CEO of the self-driving truck company Torc Robotics, which Daimler owns a majority stake in, said his company expects to announce smart infrastructure partnerships this year. He’s been working on autonomous vehicles since competing in a 2004 government-sponsored race that’s widely credited with kicking off the self-driving car industry.

“There was a lot of uncertainty as far as which technical solutions made the most amount of sense,” Fleming told CNN Business. “When we first started we really didn’t know. Today, I’m convinced.”

He says smart infrastructure, like a traffic signal that broadcasts its color, has an advantage of being simpler than the artificial intelligence algorithms companies use to perceive things like red or green lights.

Waydo recalled that when Waymo expanded to Texas, it had to adjust to traffic lights that are horizontal, requiring more work from engineers. With smart infrastructure, she said, traffic lights can all seem the same to self-driving cars, making it easier to launch the cars in new places.

Eryk Nice, a vice president of technology strategy at Hyundai’s self-driving venture Motional, is a believer that smart infrastructure can speed how fast self-driving cars arrive to the public. For example, self-driving vehicles struggle to navigate parking garages, but sensors mounted in a garage could tell the cars exactly where they are.

Vehicle-to infrastructure technology is an ambitious goal. Cavnue and its competitors will have to develop cost-effective sensors that are hung along roadsides and capable of communicating with nearby cars. Maintenance costs will need to be manageable. The company said it’s planning for the technology to be cheaper than the cost of lighting a roadway. Waydo declined to say specifically what type of sensors may be used.

But some critics are holding their ground.

Kyle Vogt, the founder of Cruise, GM’s self-driving operation, said in 2014 that, “if you have a job in [vehicle-to-infrastructure] right now, you should probably look for a new one.”

Vogt described the hurdles of the technology as proving it works before installation, getting local communities to support installing it and convincing local governments to pay for it.

Cruise spokeswoman Tiffany Testo told CNN Business that Vogt’s views haven’t changed. She pointed to the progress of self-driving cars when compared to smart infrastructure as an indication that it’s not needed.

Vogt isn’t alone in his concerns. Paul Godsmark, co-founder of the Canadian Automated Vehicle Centre of Excellence, recalled when the US federal government pushed for connecting cars to infrastructure via short-range radios, but the auto industry didn’t embrace the technology.

“The [Department of Transportation] had backed the wrong horse,” Godsmark said. “We know tech moves faster than the public sector.”

Chandler, Arizona, home to Waymo’s self-driving ridehail service, says it hasn’t installed any infrastructure especially for self-driving cars.

Micah Miranda, Chandler’s economic development director, told CNN Business that currently the technology seems like an inefficient use of capital as self-driving vehicles aren’t being used broadly, and it’s unclear what infrastructure technology would be best as it isn’t standardized.

“Tearing up roads, light poles, that kind of stuff, the cost of that infrastructure is very expensive,” Miranda said.”We have to make sure our capital dollars are going into serving the broadest portion of the population in their transportation needs.”

The largest players in self-driving — Waymo, Cruise, and Aurora, which acquired Uber’s self-driving arm — all say their vehicles don’t rely on smart infrastructure today. They also declined to say if they believe smart infrastructure would speed their ability to deploy self-driving cars to cities.

But Waymo’s parent company Alphabet may be hedging its bets. Cavnue was founded by Sidewalk Infrastructure Partners, which is funded by Alphabet.

In February, Cavnue was selected to help alleviate congestion on a Maryland highway outside Washington DC, Interstate 270. Cavnue will work on the project with the Australian infrastructure company Transurban, which was selected by the Maryland Department of Transportation and then chose Cavnue.

Rob Deans, Transurban’s vice president of technology, said his company was impressed with Cavnue’s experience in connected and autonomous vehicles.

“Autonomous vehicles are probably the single biggest change coming to transport,” he told CNN Business. “I think they have an opportunity to leverage Alphabet, the leader with their subsidiary or spinout, Waymo.”

Waymo is part of Cavnue’s advisory team on a smart road project outside Detroit. Cavnue CEO Tyler Duvall said its business model may rely on tolls for road access, or payments if a road isn’t congested.

Cavnue’s team is small today, with about 14 people, according to Duvall. But it plans to grow, and he said it’s finding significant interest, and is having conversations with about eight governments.

“There’s insatiable interest for innovation,” he said of transportation officials. “It’s can we build the team fast enough to bring the idea to them.”