Driver's ed for robots? What Cambridge's self-driving car gurus are learning in street tests | Crain's Boston

Driver's ed for robots? What Cambridge's self-driving car gurus are learning in street tests

Even a practiced driver can be a bit bewildered by Boston streets. As Cambridge-based autonomous vehicle company nuTonomy is finding out, the same goes for driverless cars.

After a brief rain delay, nuTonomy began the state’s first autonomous vehicle test drives earlier this month. So far their electric Renault ZOE has logged almost 50 miles of travel, all within the Raymond L. Flynn Marine Park in the Seaport. Through a partnership with the city, nuTonomy is hoping to make Boston the proving ground for the first commercially viable autonomous vehicle technology in the world.

The company has already beaten competitors including Google and Uber to market with a test fleet of self-driving taxis now on the road in Singapore—an ideal place to start given the Asian city’s mild weather and orderly traffic.

“And then you have Boston,” said Karl Iagnemma, nuTonomy’s founder, “which can have some very rough weather, some difficult traffic conditions, and some of the road users are, let’s say, liberal in their interpretations of the rules of the road.”

Those roadblocks are exactly why nuTonomy wants to give their technology a spin in Boston. If it can work here, the logic goes, it can work anywhere.

During nuTonomy’s maiden voyage in Boston, however, it wasn’t sleet or surly drivers that tripped up the first city-approved non-human driver, but something else entirely.

“It was one of those bendy buses, the long articulated buses,” Iagnemma said. “Our system didn't know what the heck it was. It didn't know if it was a house, a truck, a bus-and-a-half or what. It's those kind of things that are really hard to predict.”

The test drives are exposing just how challenging it is to build a fully autonomous vehicle. Rather than simply an evolution of automotive technology, Iagnemma said designing a driverless car is “much closer to the technical development that goes on in an aircraft or spacecraft.”

Despite the fact that no one has a finished product ready for widespread adoption, Iagnemma said he expects fully autonomous vehicles to be on the open road in the U.S. by the end of 2019 at the latest.

To get there, Iagnemma said he plans to double the size of nuTonomy over the course of 2017, from 60 employees to 120. Most of those hires will be engineering and robotics PhDs, like Iagnemma and his co-founder Emilio Frazzoli.

Iagnemma grew up in the Detroit area and studied engineering at the University of Michigan, interning one summer at General Motors. Despite that background, he says it was never his intention to end up in the automotive world. He still describes nuTonomy as a software company.

As principal investigator of the Massachusetts Institute of Technology’s Robotic Mobility Group, Iagnemma helped forge the first generation of “mobile robots” that would become the burgeoning autonomous vehicle industry—a market projected to top $40 billion by 2025. In the early 2000s, Iagnemma’s MIT research was largely funded by the military, which was interested in vehicles that could reduce casualties in war zones.

Since then the cost of the technology has tanked, from millions of dollars per vehicle to tens of thousands. Iagnemma expects that to drop to by another order of magnitude soon.

Paying for critical components like on-board cameras may have gotten easier in the intervening years, but programming the software—building the driverless car’s brain from various subsystems of perception, mapping, and decision making—remains difficult. Iagnemma describes having to “train” the algorithms that help the vehicles make decisions by showing them examples of what they should and shouldn't do on the road. Other times developers rewrite the code line by line.

NuTonomy wants to solve the problem before competitors in California and elsewhere, and Boston officials are cheering them on.

"Boston is ready to lead the charge on self-driving vehicles," said Boston Mayor Martin Walsh in a statement. "This is an exciting step forward.”

In October Walsh and Massachusetts Governor Charlie Baker simultaneously issued executive orders permitting on-street testing of autonomous vehicles. For now, the plan is to see how they fare on Boston streets in controlled conditions. Driverless vehicles are also part of Boston's transportation plan for the next 13 years, Go Boston 2030.

If that comes to pass, Boston drivers will have to learn to negotiate traffic with cars being driven by computers, in addition to the existing dangers and distractions of the road.

“At first these cars are going to be a little more cautious,” said Iagnemma. “That sociological experiment of how your everyday human drivers are going to interact with these cars is a bit of an open question.”

Before taking on city streets, however, nuTonomy will have to navigate an uncertain legal environment. Last week Massachusetts legislators introduced the state's first bill setting statewide standards for autonomous vehicles. In a letter to the Boston Globe, nuTonomy slammed the proposed rules, which include a new 2.5 cent-per-mile tax on self-driving vehicles and a prohibition on cars traveling more than one mile without a passenger. 

“In the short term, the . . . tax would spur an exodus of companies currently developing and testing AVs in the Commonwealth, erasing an opportunity for economic development and diminishing our reputation as a leader in innovation,” wrote nuTonomy's chief counsel, Matthew Wansley.

One of the bill's sponsors, Democratic Sen. Jason Lewis from Winchester, said the ban on empty vehicles going more than a mile would prevent "zombie cars" from roaming streets and worsening traffic.

Even if they resolve that dispute, nuTonomy and other autonomous vehicle companies will soon have to grapple with novel traffic safety challenges. Motorists have already run afoul of self-driving cars. In Mountain View, Calif., a van crashed into one of Google's driverless cars, but the human driver appeared to be at fault. The driverless car waited “at least six seconds” at a green light before rolling into the intersection, according to Google, where it was hit.

Nonetheless safety is a major selling point for those banking on a future full of self-driving cars. While U.S. auto deaths have decreased dramatically in the last 50 years , traffic deaths still claim tens of thousands of lives per year. Iagnemma, who has lost family members to car crashes, said the proliferation of autonomous vehicles could help decrease that number even more.

National Highway Traffic Safety Administrator Mark Rosekind has also said  self-driving cars could help “potentially prevent or mitigate 19 of every 20 crashes on the road.”

But that claim is somewhat hard to back up, because there is  little data about autonomous vehicles available. One study found Google’s self-driving cars were less accident=prone than people—once the human tendency to underreport crash data was accounted for. Even then, though, the comparatively low rate for driverless cars was based on a sample of just 11 crashes recorded as part of a federal study. Would it hold up if driverless vehicles became common on streets across America?

The data being gathered by nuTonomy and the City of Boston could help illuminate that question. 

“When we get to the day when I can convince myself we can put a car on the road safely, the day these accidents become largely preventable,” said Iagnemma, “it will be a success.”

January 18, 2017 - 5:57pm