In wake of Argo AI collapse, self-driving tech trickles down to make cars safer

In February 2017, Ford's billion-dollar investment into Argo AI, a little-known startup run by former Google and Uber executives, seemed to herald the era of self-driving cars. 

"We think that automation is going to define the vehicle in the next decade," Mark Fields, Ford's CEO at the time, said, adding that it was Ford's intention to launch a Level 4 (fully self-driving) car in 2021. "The investment positions Ford for its autonomous future."

"When we look back on this day, it's going to be a very important milestone on our road to an autonomous future for Ford," Fields said.

In July 2019, Volkswagen Group jumped in, investing a total of $2.6 billion into Argo AI, including $1 billion in cash, in a transaction that valued Argo at more than $7 billion. Argo AI would take over all of VW's self-driving operations and open a research center in Europe.

But this October, Ford poured cold water on those dreams, announcing it and VW would wind down Argo AI at a loss, with the 2021 deadline having come and gone. 

CEO Jim Farley said the reason was simple: "It's estimated that more than $100 billion has been invested in the promise of Level 4 autonomy," he told investors on Ford's third-quarter earnings call. "And yet no one has defined a profitable business at scale."

The shuttering of Argo AI was portrayed as the latest disaster for autonomous-vehicle development – Reuters called it "the death knell for driverless car hype" – but those billions of dollars invested in tech for autonomous robo-taxis are now trickling down to improving driver-assistance features on cars that can be bought today, experts, automakers and suppliers say. 

ADAS, or advanced driver-assistance systems, include popular safety and convenience features such as adaptive cruise control, lane-keeping and automatic emergency braking.

"A lot of the strategies have been to take all the amazing work that's going into AV [autonomous vehicles] but bring it to market sooner, and that means taking a more methodical approach," Danny Shapiro, vice president Automotive at Nvidia, said in an interview.

Centralized computing, sensors designed for robo-taxis, AI and machine learning will make cars safer, more comfortable and easier to operate, Shapiro and others say. "You still put the sensors on the car, still put an AI brain inside the car," he said.

"This is a logical way for automakers to see early returns on that investment, and still be able to bring lifesaving technology into vehicles sooner rather than later," he added. "They don't have to wait until it's 100 percent ‘done' to bring it to driverless." 

Volvo Cars CEO Jim Rowan said Volvo is "fully focused" on improving its Pilot Assist ADAS systems with upgrades such as adding lidar to the coming EX90 large SUV.

"I think on AD [autonomous driving] the narrative maybe three or four years ago was that you needed to have it to be relevant in the market. That narrative is tapped out now," Rowan said. "Everybody understands that the journey toward full AD is going through the valley of ADAS."

The mantra is continuous improvement of ADAS systems, Rowan said. "It will still be ‘hands on,' but with a lot more capability such as operating at faster speeds, offering faster lane changes and providing more accurate visibility," he said of Pilot Assist.

And as ADAS systems grow in sophistication, the SAE levels used to define them are becoming sub-segmented into Level 2, Level 2 + and Level 2 ++, for example (see chart, below). 

The global market for ADAS technology is surging. Valeo, a leader in sensors, notably lidar, expects faster growth than even electrification, with 2021 sales of 15 billion euros growing eight-fold to 120 billion euros in 2035. Valeo's own ADAS sales are expected to grow by 19 percent annually through 2025, from 1.9 billion euros in 2021 to 4 billion euros on 2025.

Valeo's lidar systems are powering Mercedes' Level 3 Drive Pilot (and a Japan-only Honda Level 3 effort), but CEO Christophe Perillat says the increasing sophistication in Level 2 has been good for his business.

"There are a lot of cars that are Level 2 Plus, which means the responsibility remains with the driver, but the car is stuffed with more and more sensors -- cameras, ultrasonic sensors, radar, lidar around the car, inside the car," Perillat said in an interview at the end of June.

ISI Evercore, in a report last month, expects the ADAS market to grow 25 percent annually until 2025, with content per vehicle for Tier 1 suppliers such as Valeo to grow from $400 to more than $600. Tier 2 suppliers such as Nvidia or Mobileye could see even greater benefits, with content per vehicle rising to more than $500 from $50 now. 

Adoption rates are "at a mass market inflection point," ISI says, with more than 50 percent of new cars equipped with at least basic ADAS features including adaptive cruise control, automatic emergency braking and forward collision warning. By 2030, 80 percent of cars will be so equipped – with about half having sophisticated Level 2 plus/Level 3 features offering some conditional automation.

Suppliers and automakers are investing heavily into ADAS research. Stellantis has announced plans to open a software hub in Bengalaru, India, that will employ 500 people to develop cockpit and driver assistance technologies through the STLA AutoDrive platform, which it is developing with BMW. The Canadian mega-supplier Magna is expanding its ADAS center in Romania with 300 new positions.

And when Harald Kroger, the former head of Bosch's Mobility Solutions business sector, joined the high-profile startup SiMa.ai, it was to work on energy efficient machine learning system on chip (SOC) platform for ADAS rather than full self-driving cars. 

"SiMa is not betting on autonomous cars," Kroger said in an interview. "We're betting on Level 3. I think that is absolutely feasible and doable and makes a ton of sense, and you get it at a cost position that can generate volume."

A "sensor suite" for a true self-driving car includes cameras, radar and lidar, to gather information in every situation – long- or short-range, different light and weather conditions, and object classification and detection.

Those sensors, and the powerful central computers that fuse their images together, are now making ADAS systems more effective and redundant, Shapiro of Nvidia said.

"The amount of information the car has and the ability to detect things dramatically increase," he said. "As a result of the massive computing power in the car, it means it's a lot safer." 

ADAS functions that have been enabled or improved include cross-traffic detection, blind-spot monitoring and pedestrian-detection systems, he said. 

"Some of these things have been around for a while, but they were little gray boxes with a chip that were doing a specific function, but with the move toward centralized computing they're all integrated," Shapiro said, and functionality can be enhanced with software. 

For example, software that merges driver-monitoring sensors on the interior and exterior camera could prevent an accident at a stop sign in which a driver might be looking to the left but doesn't see a child stepping off the curb to the right, he said. In another example, sensors could prevent a driver from opening a door if a cyclist is coming from behind. 

ISI Evercore in its report estimated that ADAS features could prevent 20 to 40 percent of the 1 million annual traffic fatalities worldwide -- with specific functions such as lane-keeping assistance each saving thousands of lives.

"ADAS is the next air bag, seatbelt or ABS," ISI said in describing the impact advanced driving assistance could have on traffic fatalities.

Automakers are taking another look at SAE Level 3, meaning the driver is free to do other things, but must be ready to take over control within seconds. Worries about liability and a lack of regulations had derailed several early efforts such as Audi's.

In 2017, Volvo announced it would skip Level 3 and go straight to Level 4, with Hakan Samuelsson, the CEO at the time, calling it "unsafe."  And in 2020, BMW pushed back a plan to offer it on its iNext EV, saying the technology wasn't ready.

But with technological advances and a new U.N. directive allowing automated driving up to 130 kph on the books, Level 3 may have regained its momentum. 

Mercedes-Benz was the first brand in Europe to make the leap with Drive Pilot, which is certified in Germany for limited use – 60 kph or less, in good weather, on highways, and only in heavy traffic. It is available as an option on certain S-Class and EQS models.

Ford will focus on advances to its BlueCruise Level 2 system to reach this level, Farley said, and "give them back the single most important commodity in our modern lives, time." BMW and Stellantis are also aiming for Level 3 in the next year or two. Even Volvo said last January that it would introduce what it calls "unsupervised autonomous driving" via a system it calls Ride Pilot, although Rowan -- who took over as CEO in March -- demurred when asked about a time line for its introduction.

Advances within Level 3 are likely to be methodical, gradually increasing the operational design domains (ODD) to include higher speeds or the ability to operate on damp roads, a Mercedes spokesman said. "It's a new and complex technology, and with new technologies you can't take too-big steps," he said. 

Even as Mercedes pushes ahead with Level 3, it has already commercialized a Level 4 feature. Just last week it won certification to offer a driverless parking service in which the car will park itself remotely, at a single garage in the Stuttgart airport.

"Level 4 is in the future, but with Drive Pilot we've set the foundation for upcoming features, especially with redundancies and the sensor sets," the spokesman said. "We won't flip the switch and there will magically be a Level 4 or Level 5 car. We'll take it slow and safe."

The shakeout in the self-driving industry is creating a "two path" approach, LMC Automotive said in November. "On one development path, we see companies like Waymo striving to create Level 4 AV vehicles for shared fleet usage; and on the other path, OEMs (and their suppliers) carefully developing ADAS technologies up through the SAE Levels but aimed at owned vehicles."

Kroeger, the former Bosch executive, agreed that the path to true self driving would involve many small steps.  "Maybe you don't want to start with a robo-taxi in the middle of a pedestrian zone, which is maybe the hardest thing to do. Maybe it's a hub-to-hub trucking service that makes more sense," he said.

"It's important that the industry doesn't panic and say it's the end of the autonomous dream," he added. "I think that's wrong as well."

Douglas A. Bolduc contributed to this report