July 8, 2021
The rollout timelines, the scale of the engineering challenge and the data required, and the right market fit are sources of misconceptions, AV experts say.
Head of Field Safety at Waymo
The general public is still often confused about the difference between driver assistance and fully autonomous driving technology, and even those who understand the difference rarely appreciate how big the technological gap is between these two levels of capabilities. SAE defines six levels of autonomy, ranging from none to a system that can drive anywhere in any condition, but at Waymo, we think that the most important distinction boils down to just two — is a licensed driver required for operating the vehicle or not?
The confusion we see in consumers and the general public between these two distinct states can create false confidence in the capabilities of driver-assist (not fully autonomous) systems. Advanced driver assist features (like the technologies that are sold in vehicles today) require a fully attentive driver and most therefore instruct the driver to keep their hands on the steering wheel at all times.
There is a massive gap between the technology in driver assistance systems versus the technologies that are required for a fully autonomous system (e.g., sensor suite, redundant systems and perception, behavior prediction and path planning software). The Waymo Driver, which operates the first public fully autonomous ride-hailing service in the world, has these technologies and does not require a licensed driver in order to operate. In fact, we remind our riders to "please keep [their] hands off the wheel," since our autonomous driving technology is in control at all times so they can sit back, relax and just enjoy their ride.
Chief Legal and Policy Officer at Nuro
We commonly hear misconceptions about job loss from the deployment of autonomous vehicles and the accompanying safety of those vehicles. The truth about employment is this: Autonomous delivery boosts local economies and creates jobs. New research from Steer Group estimates delivery AVs will create more than 3 million new jobs over a decade.
Restaurants that introduce delivery often experience a 15% increase in sales, leading to more hiring and improving the local tax base. AV technology lowers the cost of home delivery, making it more affordable and equitable. This can significantly improve access to healthy groceries in America's low income food deserts.
Finally, AV delivery can significantly improve road safety. A recent VTTI study found that for every mile of manual driving replaced by autonomous delivery vehicles like Nuro's R2, road injury risk is reduced by 60%. Fatal accidents on our roads increased by 7% last year—an estimated 38,000-plus deaths despite Americans driving 13% fewer miles compared to 2019. Zero of these deaths were caused by fully autonomous vehicles like Nuro's. 94% were caused by human error. While no driver or driving system can create a risk-free environment on the roads, delivery AVs will reduce car crashes, save lives and make our streets safer for all.
Co-founder and Chief Product Officer at Aurora
Autonomous vehicles have to safely handle the myriad situations that human drivers encounter every day, whether common, like construction, or rare, like a darting animal. At the rate these rare events occur, our industry can't continue to depend on massive on-road vehicle fleets to gather data. Brute-force data gathering is common in first-generation AV programs, as it's simpler to execute and tells a good story. But it's too slow to get you to a scalable product.
We've built a better way to experience the world far more rapidly, efficiently and safely: Aurora's Virtual Testing Suite, a collection of sophisticated simulation and data-replay tools that leverages the over 4 million miles our test vehicles have experienced in the real world through the years, and the interesting events encountered by one of our commercial partner's fleet in their over 7.5 billion yearly miles. We further permute these scenarios, varying things like weather, traffic density and pedestrian behavior to expand their coverage. With this, we test the on-road equivalent of over 20 million miles every day, with far fewer on-road vehicles. To put this in perspective, our virtual fleet drives the daily equivalent of over 46,000 round trips between Houston and Dallas, experiencing a diversity of scenarios and driving conditions that would take even a large on-road fleet decades to experience.
As the industry reaches the long tail of development, we believe that companies that invest in virtual testing will achieve higher levels of safety and quicker paths to a scaled product. While it won't sound as impressive as large on-road fleets, it's the only way to scale safely, quickly and broadly.
Founder & CEO at Waabi
In all my years working in self-driving technology, the piece of the puzzle that is most widely misunderstood is AI — how it works, what it is capable of and what role it should play in the development of self-driving technology. This misunderstanding exists in both the public sphere as well as among developers of the technology.
For the public, years of hype around the potential of "robot vehicles" collided dramatically with the realities of slow and limited progress in the industry. This has resulted in a wariness around self-driving technology and, as an extension, AI.
The reality, though, is that AI has never been fully leveraged in self-driving technology. There are two main existing approaches to autonomy used by the industry today: (1) the traditional, unscalable approach which does not exploit the full power of AI, but instead uses multiple AI systems to solve small subproblems; and (2) the "black box" AI approach, which is not explainable and/or verifiable and cannot generalize to unseen scenarios. Neither is suitable for commercial deployment of self-driving technology.
At Waabi, AI is at the center of the solution. What we mean by this is that a single AI system is able to take the output of the sensors and produce the decision making for the self-driving vehicle. This breakthrough approach leverages deep learning, probabilistic inference and complex optimization to create software that is end-to-end trainable, interpretable and capable of very complex reasoning.
By reimagining the role of AI in self-driving technology and developing a new generation of algorithms to enable safe and effective driving, Waabi can help to transform the public's perception around AI as well as forge a path for future world-changing innovations.
Co-founder and CTO at Zoox
It has been eye-opening to watch the public's reaction to self-driving cars over time. It's quite easy to get enchanted with one or the other of these observations: either that the problem is practically impossible because of all the things that still aren't perfect, or that it must be almost solved because of some super cool demo or milestone that seems incredibly impressive. The reality is in between, and for whatever reason, it's surprisingly hard for people to maintain a nuanced appreciation of that balance. Achieving a world with ubiquitous autonomous vehicles will be an incremental process that advances every year — and remember, the alternative is the bar of human performance that stays nearly stagnant. It's the opportunity of a lifetime to participate in the journey of making autonomous driving technology relentlessly better. Soon, it will reach a crossover point where the public begins to adopt it at scale, which will be a transformative win for society at large.
I have always tried my best to be upfront, honest and realistic about where the technology is — and while I've certainly not nailed all of my predictions, I do think I've managed to be fairly balanced overall. As technologists, when we are overly optimistic or pessimistic, we do a disservice to ourselves, the industry and our technology.
Chair, Autonomous Transportation and Shared Mobility Practice at Fenwick & West, LLP
One misunderstanding from the public about autonomous vehicles is that within the next 10 years they will be able to purchase an autonomous vehicle that will eliminate their need to drive anywhere. The technical difficulties in having a vehicle fully eliminate the need to drive, e.g., leave from their home in California and travel to anywhere in the country at any time, is significant and is many years away from reality. Examples of the technical issues include updating maps, driving in bad weather and driving on roads with poor lane markings. In addition, the cost of AVs will be significantly higher than today's vehicles because of the additional sensors and other required hardware. This high cost will place AVs out of the purchasing range of the vast majority of people, at least initially, until the cost of the additional hardware drops significantly.
The first AVs will likely be used as fleet vehicles for transportation-as-a-service companies because the AVs can be used nearly all the time, which will enable the companies to realize more revenue per vehicle than what is possible today using drivers. These companies can solve some of the technical issues by limiting the geographic scope of the service which bounds the problem of creating and updating highly accurate maps to the service area and eliminating (or accounting for) poorly marked roads. The service area may also lend itself to having the AVs travel at slower than highway speeds which enables the vehicle more time to analyze and react to its environment.
CEO at Embark Trucks
When people think about autonomous vehicles, they instantly imagine being chauffeured around in a self-driving passenger car. It's understandable — we want to envision how this technology will benefit our daily lives, and much of the early AV industry was focused on cars. However, what people don't realize is that freight trucks will likely be the first wide-scale commercialization of autonomous vehicle technology. First, there is a clear business case. Car decisions are wrapped up in emotion, status, and require each driver to make a personal choice to "go AV." In contrast, when it comes to freight trucking, a solution that is proven to be cheaper, safer and more efficient will be quickly adopted at scale by major carriers in the $700 billion freight trucking industry.
Second, trucking is a better fit for the first scale deployments of AV technology. The relative simplicity of long stretches of interstate lend themselves better to automation while being more difficult for human drivers, who can struggle with boredom, fatigue and distraction. And third, there is real demand for autonomous trucks. While consumers may not be sure they want an autonomous car, the trucking industry is facing a 60,000 driver shortage that is projected to reach 160,000 by 2028. Over the last few years I've seen both new crops of AV startups focused on trucks, and longstanding AV programs pivot to trucking. So the next time people think about autonomous vehicles, maybe trucks should be the first thing they imagine!
SVP, Engineering at Cruise
Self-driving is an all-encompassing AI and engineering challenge. It's easy to see an AV on the streets and think only about the AI models that power them or the compute and sensor suites built as part of it, but there is a virtual software assembly line built alongside the car itself that enables us to meet the unique scale and safety imperative at play here.
To enable AVs to drive superiorly in any given scenario, and continuously evolve and adapt new paradigms, it requires an ecosystem capable of ingesting petabytes of data and hundreds of years worth of compute every day, training and testing models on a continuous loop for multiple times a week software updates that improve performance and ensure safety. The complex network of new tools, testing infrastructure and development platforms that are behind every seamless handling of a construction zone or double-parked car are themselves significant engineering achievements that stand to have an outsized impact beyond AV as they push the boundaries of ML, robotics and more.
Founder & CEO at SafeAI
The biggest misconceptions I see about autonomous vehicles today revolve around the timeline for rollout — both when exactly self-driving cars will be available, and whether it ultimately matters if that moment is one, or five, or 10 years away.
One of the biggest questions in the industry has always been when the everyday person will get to experience a self-driving car. Self-driving car companies have often gotten caught up in the excitement, too, committing to timelines that turned out to be unrealistic. It was widely asserted, for example, that autonomous vehicles would be available by 2020 — a deadline that has come and gone with only pilot programs. And still, before self-driving cars can take off at scale, there remain important hurdles for the industry to clear — from sharpening the technology to account for myriad edge cases in city centers, to implementing proper regulation so self-driving cars can integrate with existing transportation systems.
But I would also argue that the matter of when self-driving cars will become a reality is not quite as relevant as most people think. The question of whether they are two or 10 years away is a narrow one, and one that won't ultimately matter a tremendous amount in the grand scheme of things. Autonomous vehicles are poised to fundamentally transform one of the core pillars of human society, ushering in a new era for transportation. The specifics of timing pale in comparison to the revolutionary impact this technology will have when it's ready.
Director of Autonomous Vehicle and Future Technology at Ford Motor Company
Autonomous vehicles are extremely complex. To achieve a safe, automated driving experience, we need to mimic or exceed human capability to respond and make decisions. Overall, the technical challenges including the resources required, have been underestimated by many. In the past, there was speculation and hype that we would already be at an inflection point leading to the scaling of numerous autonomous vehicles operating within cities. We know this is not our reality, but why?
Automated driving algorithms rely on complex sensing and compute capability with extensive machine learning development, including refinement and validation to address potential and rare scenarios the vehicle might encounter. For example, the vehicle must quickly respond if a pedestrian steps out into an intersection, or a vehicle unexpectedly hits its brakes. Achieving human-level perception is a tremendous challenge and requires state-of-the-art machine learning techniques.
While we know the industry is testing autonomous vehicles daily on public roads this is however, in a relatively controlled situation. The human driver often has the ability to take control of the vehicle to address the localization, perception, prediction and control tasks. The driver also takes into account the complexity of road users, visibility, road conditions, and reacts to other vehicles' maneuvers.
Even as some prognosticators have acknowledged that fully automated driving is more difficult than previously thought, we have been cautious to ensure that there is sufficient robustness in our driverless operation to build trust and operate safely with our customers and partners, including in cities where we are testing and planning to launch.
See who's who in the Protocol Braintrust and browse every previous edition by category here (Updated July 7, 2021).
Kevin McAllister ( @k__mcallister) is a Research Editor at Protocol, leading the development of Braintrust. Prior to joining the team, he was a rankings data reporter at The Wall Street Journal, where he oversaw structured data projects for the Journal's strategy team.
More from Braintrust