Self-Driving Vehicle Technology: Progress and Promises

Automakers have already spent at least $16 billion developing self-driving technology, with the promise of someday creating fully autonomous vehicles.² What has been the result? Although it seems that we have more promises than actual progress, some encouraging experiments are now under way, and there have been intermediate benefits in the form of driver-assist safety features.

Engineers started on this quest to automate driving several decades ago, when passenger vehicles first began deploying cameras, radar, and limited software controls. In the 1990s, automakers introduced radar-based adaptive cruise control and dynamic traction control for braking. In the 2000s, they introduced lane-departure warning and driver-assist parking technology. Since 2017, Waymo, Uber, Daimler, the U.S. Postal Service, and several other automakers all have launched experiments with robo-taxis or robo-trucks, targeting Level 4 Autonomy (see the sidebar on the last page of this column).^4,13 If and when this technology will make its way into your average passenger vehicle is uncertain, but there is no doubt that companies have been moving closer toward their goal.

The basic technologies and engineering skills needed to make self-driving vehicles more widely available already exist. The most popular camera packages from Mobileye (purchased by Intel in 2017) and OmniVision are relatively inexpensive. However, some self-driving systems deploy much more expensive lasers (usually referred to as "lidar" for Light Detection and Ranging) as well as radar and ultrasound sensors, provided by firms such as Ibeo, Velodyne, and Autoliv. Major auto parts and technology suppliers, led by Bosch, Denso, Aptiv (formerly Delphi Automotive, which also purchased the AI and robotics software company NuTonomy in 2017), TRW, and Continental, assemble components into various driver-assist systems and use microprocessors from Intel-Mobileye, Nvidia, and ARM. Blackberry, formerly a pioneer in secure email and smartphones, has become a player in the automotive IoT software market with its QNX operating system, which runs on some 150 million vehicles.¹¹ Green Hills Software competes in this business as well, along with Google. Automakers, auto parts vendors, and robotics and AI startups, all have been learning how to design the AI and machine-learning applications needed to process data and warn drivers or provide instructions to the vehicle subsystems.

Electric vehicles rely heavily on computers to control their functions, and this characteristic makes them especially suitable for self-driving technology. Not surprisingly, Tesla vehicles deploying driver-assist technology have logged nearly two billion miles and the company probably leads the industry in data collection.¹⁴ Tesla was able to commercialize its Autopilot system because it used cameras, radar, and ultrasound, rather than the more expensive lidar. However, Tesla vehicles are still somewhere between Levels 2 and 3—far from the goal of autonomous driving. They also have been involved in several high-profile accidents when drivers stopped paying attention, so the company now insists that drivers keep their hands on the steering wheel and eyes on the road.

Another key player is Waymo, founded as a Google R&D project in 2009 and spun off as a fully owned subsidiary in 2016. Waymo does not manufacture cars but has partnered with Fiat-Chrysler, Audi, Toyota, and Jaguar to retrofit their vehicles. It also makes a lot of its own hardware and software to reduce costs. Waymo's technology is more advanced than Tesla, and is presumed to operate at Level 4—but with caveats. The vehicles drive mainly on predefined routes and rely on an expensive combination of lidar, cameras, and radar, as well as human drivers for backup. Since March 2019, Waymo has been operating 600 autonomous vehicles and claims the lead in Level 4 data, with approximately 20 million miles logged on public roads.¹⁶ Since mid-2019, Waymo also has been operating a robo-taxi pilot in California, offering thousands of rides each month.⁷ Although it has lost billions of dollars, Waymo's ultimate goals are to offer ride-sharing services with tens of thousands of vehicles (perhaps with Uber and Lyft as partners) and to license technology to automakers and service providers. To expand its ride-sharing business, Waymo has ordered 62,000 Chrysler Pacifica vans and another 20,000 Jaguar I-Pace cars.¹⁰

Even automakers with modest financial resources can now buy access to self-driving technology. Several companies provide turnkey driver-assist or semi-autonomous driving systems; others focus on data and simulation software, sensor hardware (cameras, lidar, radar, and ultrasound), mapping and location-based software, and vehicle communications systems.⁵ But there is a problem: Exactly what combination of hardware and software works best remains unclear, and there is, as yet, no single industrywide "platform" or common approach for self-driving vehicle technology and communications. Tesla could have been an industry leader by making its software platform available, but it has proceeded largely on its own.

Other automakers and ride-sharing businesses have formed partnerships that now compete with each other, though they often rely on the same suppliers.⁵ For example, some 80% of vehicles with Advanced Driver Assistance Systems (ADAS) already use Intel-Mobileye cameras, chips, and software.⁶ Volkswagen is at the center of one group built around Argo AI technology, with Ford as a major investor. This alliance has loose or indirect ties to Mercedes-Benz (Daimler), BMW, Toyota, and GM (which bought Cruise Automation in 2016). Other allies are Lyft and Didi. In addition to Argo AI, technology providers include Bosch, Nvidia, Microsoft, Apple, Huawei, Qualcomm, Baidu/Apollo, TomTom, Waymo, and Here (mapping technology). BMW and Mercedes-Benz have a separate alliance, with loose ties to Renault-Nissan, Geely in China, and Audi (a Volkswagen subsidiary). They rely on many of the same suppliers as well as IBM. Toyota leads another group, with ties to GM, Geely, BMW, Mercedes-Benz, and Uber. Mercedes-Benz, which has been working with BMW, Audi, and Bosch, launched another partnership in June 2020 with Nvidia to develop a unique software-defined self-driving architecture by 2024.¹⁵ It is not clear how this effort will impact other Daimler/Mercedes-Benz partnerships. Various automakers and technology vendors, including Intel-Mobileye, are also testing self-driving technology in ride-sharing or ride-hailing ventures while partnering with Uber, Lyft, and Didi—and providing competition to Waymo.

Despite the intensifying competition, there are good arguments for more cooperation. First, the technology remains expensive to develop, especially as companies try to move beyond Level 2. Cameras are necessary to view road signs and traffic lights, but they perform poorly in bad weather. Radar is cheap and able to detect the range and speed of distant objects, but radar images are not as precise as the three-dimensional pictures lidar generates, albeit at considerable expense. Ultrasound or land-based sonar, used extensively in Tesla vehicles, provides a 360-degree view that compensates for camera blind spots and aids in parking, but it can only detect nearby objects and does not replace camera vision.¹²

Second, autonomous driving requires massive amounts of data to refine the vehicle control systems. The more fragmented the market and variations in sensor combinations and algorithms, the less usable data available to any one manufacturer or platform provider.

Third, it could be helpful for vehicles to communicate with each other and with some traffic control systems, as airplanes do. This type of communication does not solve the problem of a pedestrian suddenly appearing in front of a moving car, but it could reduce accidents with other vehicles, especially if we retrofit older cars and trucks with inexpensive communications devices or smartphone cameras.¹ The Networking for Autonomous Vehicles Alliance (see https://nav-alliance.org/), founded by Bosch, Continental, Marvell, NVIDIA, and Volkswagen of America, is also working "to provide a platform for the automotive industry to develop the next generation of in-vehicle network infrastructure for autonomous vehicles," though it has yet to set any global standards.⁹

Ride-sharing and ride-hailing companies are likely to buy or lease fleets of self-driving vehicles to get rid of driver costs, their main expense, and their large-scale purchases could help reduce costs for the automakers. However, the ride-sharing companies are already losing billions of dollars per year, and they will have to take on the enormous costs of owning or leasing millions of their own vehicles.^a Self-driving technology might even someday eliminate demand from companies like Uber, Lyft, and Didi.³ Most privately owned automobiles sit idle approximately 95% of the time.⁸ Tesla and other automakers are exploring how to enable owners to share their vehicles when not in use and earn some revenue from this activity, rather than relying on Uber or other intermediaries.

Back to Top

Conclusion

In sum, there currently are several experiments with robo-taxis and robo-trucks on prescribed routes, but still with human drivers as backups. Full automation at Level 4 or 5 remains a distant goal for the average consumer, and it is difficult to pinpoint a timeframe when this will become a reality. Meanwhile, all this R&D is not for naught. Even if automakers never advance much beyond Level 3 over the next decade, driver-assist technology has already made driving safer. Assisting rather than replacing drivers should perhaps be our end goal, rather than full automation.

Back to Top

Back to Top

Back to Top

Back to Top