When it comes to autonomous vehicles, you may be wondering “Are we there yet?” And by and large, the answer is actually yes.
So why have autonomous vehicles not proliferated our roads with robo-taxis and made car ownership all but unnecessary?
The answer to this question requires looking at autonomous vehicle technology from both an investment and technological angle.
Autonomous Vehicles and the VC Investment Race
In 2020 and over the ten years prior over $250bn was invested globally into mobility technology startups, according to a recent Pitchbook report on Mobility Tech. This includes investments into full-stack autonomous vehicle startups (such as Waymo) and software and hardware components such as HD maps and LiDAR, as well as related services such as:
- Auto commerce
- Marketplaces
- Fleet management and Connectivity
- Ridesharing
- Micro-mobility
- Last-mile technology
- Electric vehicles
- Urban air mobility
Although investments peaked in 2018 with over $60bn invested globally, 2020 still saw a healthy $41bn invested into mobility tech. However, only late-stage companies saw an uptick in investments while early-stage companies saw investments drop nearly 40% YoY.
As investments have shifted to late-stage companies, valuations for these companies have nearly doubled to a median value of $300mn—whereas valuations for early-stage startups fell 20-30% YoY.
From a startup investment’s point of view, this means that mobility tech, including autonomous vehicle technology, is nearing maturity with many competitors in each AV category and winners and losers becoming apparent.
Over the past few years, and especially in the last 12 months, we have seen many AV software and hardware companies exiting through acquisitions or SPACs (such as Aurora, Cruise, Aeye, Innoviz, Deepmap, etc.). Competition is particularly fierce in the LiDAR space, where over 75 companies currently exist.
Automakers and big technology companies are racing to dominate the space. As a result, a wave of consolidation is also occurring amongst self-driving technology companies as automakers and big technology companies race to dominate the space. This gives large global companies or mature start-ups with deep pockets a significant advantage in this space and just in the past year, several large M&As were announced including Intel’s acquisition of Moovit for $900mn, Amazon’s acquisition of Zoox for $1.3bn, and so on.
Technologically, Autonomous Vehicles Fall Short By a Hair
It’s fairly clear that from an investment point of view, AV technology is considered to be fairly mature, moving into the late-stage and pre-IPO phases. What about from a technological point of view?
It seems that we’re about 99% there, but we need to get to 99.9999% before AV can fully attain public acceptance.
Consider autonomous vision, for example. Autonomous vision is a combination of LiDAR, radar, and cameras that can already outperform human vision. It’s able to “see” much further ahead (up to 200m) and in the dark or in poor weather conditions. Although there are still some flaws and new unproven technologies (such as FMCW) still grab investors’ interest, the biggest issue with LiDAR is the cost (a few thousand dollars per unit).
Going forward, companies such as Sense, are focusing on bringing costs down to around $100 per unit, enabling AVs to adopt LiDAR en masse.
Further proof of the superiority of AV’s driving vs the average human driver? In its first-ever report, Waymo announced that during 2019 and the first nine months of 2020, over 300 of their AVs with safety drivers drove 6.1mn miles and its fully driverless vehicles drove 65k miles in simulated and real-life environments. During this time, their vehicles were involved in only 18 minor crashes and 29 near-miss collisions (including times when safety drivers had to take control).There were no major injuries and nearly all of the collisions were the fault of the other driver.
As a comparison, the US Department of Transportation as well as the car insurance industry estimates the average US driver drives an approximate 13,500 miles per year and will be involved in 4 car crashes in their lifetime. This would mean if they drove the equivalent number of miles as Waymo’s AVs, they would be involved in over 1800 accidents.
Granted, Waymo’s AVs were driving relatively slowly (approx. 40 miles or below) and their real-life testing grounds were limited to 100 sq miles service areas located in several towns in Phoenix. However, it does prove that AVs do have the technical capabilities to be on the roads today.
So what is holding back AV’s from being on the roads?
Public Trust and Understanding
AV companies are not transparent; they tend to be fairly siloed and are not inclined to share AV-related information. While this benefits them by keeping costs low, protecting proprietary data, and maintaining control of their narratives, it also breeds distrust in the general public.
In fact, Waymo’s report was the first time an autonomous vehicle company released a detailed overview of its safety methodologies, including vehicle crash data. This challenges other AV companies to do the same, hopefully kick-starting a transition to increased transparency from the AV industry.
Traditionally, AV companies have kept a tight lid on measurement metrics and only shared information to the public in very controlled settings ie press releases, or blog posts. This created a lack of understanding that the public has regarding the technology. For example, several fatal Tesla crashes while on autopilot have made splashy headlines, bringing into doubt the maturity of AV technology. However, documents from California’s Department of Motor Vehicles revealed that Tesla’s autopilot was only Level 2 autonomy, a step up from cruise control vs true autonomous driving which is a Level 4 or 5.
For ADAS, Telsa’s autopilot technology is outstanding but it’s not quite self-driving and shouldn’t be treated as such.
In addition, self-driving technologies improve continuously as they log in hundreds of thousands of hours on the road or in simulated environments. What was a major problem a couple of years ago—for example, the unprotected left-hand turn—may no longer be a big issue today. Therefore, along with increased transparency, more education, and regular updates about autonomous driving technology, guidance on what can be expected from these technologies and how they should be handled is required before AVs can gain true public acceptance.
Lack of Regulations
In the US, there are currently no federal laws governing AVs. That includes any requirement for AV companies to submit information about their testing activities. Instead, the government relies on individual states to regulate AVs and most states require next to nothing in the way of testing data.
Since 2012, only 29 states have enacted measures related to AV. In addition, there is no federal or state system in place for independent certification of AV technology, and regulations regarding liability, data privacy and security have also not been defined.
In a recent report on “Driving the Future of AV Regulations: Barriers to Large-Scale Development,” the Center for Strategic and International Studies (CSIS) listed the lack of regulations as a roadblock to AV adoptions. They stated:
“While there has been recent bipartisan interest in introducing legislation regarding self-driving cars, most recently as part of the Endless Frontier Act, Congress has not passed AV legislation thus far. Similarly, federal regulators have issued guidance—but no binding regulations—on AV operations. States have filled this vacuum, creating a patchwork of regulations that could hinder deployment of this technology at scale. As a result, the AV industry faces an uneven and uncertain regulatory environment and lacks a clear path to large-scale deployment.”
Still, this situation is improving as the National Highway Traffic Safety Administration is in the process of formulating a framework for automated driving systems which will likely include non-binding guidance as well as binding regulations on AV. This framework, once formulated, will likely be implemented in phases in what looks to be a multi-year process.
Nonetheless, federal regulations will be key to unlocking the benefits of AV, allowing AV companies to scale production across states and providing comfort to consumers that AVs can safely follow clear, consistent, and defined standards.
Outside of the US, KPMG publishes an annual report on which countries are the most prepared to adopt AVs. From a regulatory perspective, Singapore, the UK, Netherlands, Finland, and New Zealand round out the top 5.
Lack of Industry Standardization
Within the AV industry itself, there are no agreed-upon standardized methods of measuring and assessing AV safety or ensuring there are no overlaps or gaps in safety standards. This is in addition to the lack of federal regulation we mentioned before. And while this process is not straightforward and continual technological modification adds complication, there have been some improvements.
In 2020 a wave of new industry standardization initiatives started to roll out, with a common mission to develop safety standards for AI-driven systems used in AVs and robotics. IEEE alone is kickstarting several working groups aimed at AV safety: IEEE P2846, IEEE P2851, and IEEE P1228. In addition, Underwriters Laboratories’ UL 4600 draft looks to be a guide in building the safety case for an AV design.
Prior to this, there were a few industry consortia led by major auto OEMs, 1st tier auto suppliers and global tech conglomerates such as SaFAD and Automated Vehicle Safety Consortium that were formed to provide higher-level guidance (vs detailed instructions on safety for autonomous vehicles).
This sudden rush to set standards for safety and reliability in AI-based systems is a marked departure from the recent past where, as mentioned previously, AV developers have kept their methods close to the vest, viewing AV safety as a competitive edge vs a required collaborative effort.
Overall, the AV industry is still at the beginning stages of collaborating and developing standards for autonomous vehicles. There is no agreement across the industry or across regulatory bodies on how machine learning should be trained, tested, or validated—or that any new learning is safe, and that the system doesn’t forget previously safe behaviors.
Conclusion
As far as AV technology goes, autonomous vehicles essentially have the capability to hit the roads tomorrow, though continuous improvements are still required. However, it will likely be many years before AV becomes ubiquitous as regulators and the industry have yet to set standards to measure and ensure the safety of AVs, as well as on the data collected by AVs.
AV companies will remain reluctant to send their cars on the roads until regulators set clear guidelines on liability. Without these guidelines and more concerted efforts to educate and update the public on AV technology and safety efforts, we cannot expect large-scale deployment of AVs across the US. Other countries such as Singapore and Finland where there is both high public acceptance and strong regulatory efforts might see universal deployment of robo-taxis, driverless trucks, and other AVs before anyone else.
