Vehicle Autonomy and Connectivity: Developments and Broader Implications
Reinventing the Wheel - Special Issue on AI and Digital Transformation
Dan Ammann – then president of GM – predicted in 2016 that the auto industry would change more by 2021 than it had in the previous 50 years, referring to driverless vehicles, ride-sharing, connectivity and electrification. Elon Musk said in 2020 that Tesla “will have the basic functionality for Level 5 autonomy [complete driverless capability in all conditions] this year.” Both of those timelines were wildly optimistic, despite billions in investments by many companies.
Where do things stand? What went wrong? The Center for Automotive Research1 provides an overview of the current development and deployment of advanced driver assistance systems (ADAS), autonomous vehicles (AVs) and vehicle connectivity. Advances in, and obstacles to, these technologies are relevant not just to the auto industry, but to the broader future of AI and digital transformation.
Advanced Driver Assistance Systems
Though human error is responsible for the vast majority of serious crashes, ADAS have their problems too, including technology limitations and driver misuse. ADAS features that aid, warn or assist drivers have rolled out gradually over 24 years, with some – like backup cameras and collision warning – now common. AVs that operate only in certain conditions and still require a driver’s attention have been launched, but have been involved in high-profile accidents. Although Elon Musk predicts that revenues for Tesla’s system (deceptively named “Full Self Driving”) will be enormous, we doubt that drivers will perceive sufficient value to subscribe. Lower costs, technological improvements and regulatory pressure will likely be necessary for ADAS to become truly widespread.
Automated Driving
ADAS assist the driver, but in true AVs (SAE Level 4 and 5), the vehicle is in complete control. Five or ten years ago, many predicted that AVs would be widespread by now. Currently, few project that AVs will arrive by 2030 outside of narrow, “geo-fenced” fleet or freight applications.
A range of issues proved to be tougher to solve than proponents thought:
consumer acceptance and risk tolerance
hardware and software capabilities and development costs
adverse weather
lack of supporting infrastructure
regulation and policy
cyber/data risk
liability
maintenance costs.
None of this was a surprise! All of them were known 5-10 years ago. AV proponents assumed that remarkable improvements in software and hardware would continue indefinitely, that costs would plummet with scale and time, and that the benefits would be so obvious that consumer and governmental resistance would melt away.
Expectations have become more realistic. Developing the right technology platform remains highly uncertain, partly because the original deep-learning models aren’t sufficiently flexible, robust or transparent. This is an important lesson for broader AI applications. Machines do not (yet?) have general intelligence – the ability to learn and adapt. Applications that face a wide range of environments – like vehicles on city roads – and which have scant room for error –like vehicles with passengers – are poor near-term candidates for AI. Based on discussions with experts inside and outside the industry, the Center for Automotive Research now foresees widespread deployment of automated robo-taxis and freight services beginning only about 2035, with a similar timeline for affordable personal AVs.
Vehicle Connectivity
V2X is the shorthand used to encompass vehicle-to-vehicle, vehicle-to-infrastructure, and vehicle-to-network connectivity. Such connectivity has long been viewed as a critical enabler of AVs and improved safety.
The US government saw great potential in V2X, and in 1998 allocated a significant portion of the spectrum to Wi-Fi based dedicated short-range communications (DSRC) technology relying on roadside units. Over the past 20 years, DSRC has been installed in many vehicles but has never been used, mainly because most vehicles still don’t have the technology. In 2016, the Obama administration proposed a V2X mandate using DSRC, but it was never implemented.
In 2017, a cellular-based approach (C-V2X) received support from telecom and tech companies, partly because the infrastructure is in place, and partly because they wanted to access some of the spectrum allocated to DSRC. Over the objections of the Department of Transportation, in 2020 the FCC reallocated part of DSRC’s spectrum to Wi-Fi and C-V2X. A federal grant program was set up to retrofit vehicles to C-V2X, and survived a legal challenge. Meanwhile, China has mandated C-V2X and the EU has been on the fence between DSRC and C-V2X.
For V2X to improve safety and reduce congestion, extensive infrastructure will be required; the technology must be much more widely adopted; and the solutions must also be reliable and profitable, while respecting users’ privacy. It’s unlikely that all these conditions will soon be met in the US, where there's far more interest in the (unproven) potential for cloud-based V2X to produce saleable data; but V2X has a strong chance of adoption in China.
Conclusions
The automotive industry – except Elon Musk – has become more realistic about ADAS, AVs and V2X. All these markets now seem more evolutionary than revolutionary. The industry’s experience has lessons for us all:
Consumers are understandably reluctant to depend on unfamiliar technology for essential tasks like driving.
The number and reach of assumptions that underly a scenario have an inverse relationship to its plausibility and timing.
Applications that depend on network benefits (like V2X) need to have private benefits to users even at small scale or be mandated by the government. Facebook was fun even if only people in the same dorm were using it.
Early successes in AI are more likely in tightly controlled environments, not in applications that require sensory input, flexibility, or true machine learning.
1Souweidane, N., Smith, B. (2023). State of ADAS, Automation, and Connectivity. Center for Automotive Research, Ann Arbor, MI.