It’s official: Gartner has finally sent automated vehicles into the “Trough of Disillusionment” in 2018’s edition of its “Connected Vehicles and Smart Mobility” hype cycle.
Of course, the market analysis firm eventually does this with every emerging technology, as marketing finally runs into reality before progressively maturing towards successful commercialization. While AVs had been inching closer to this point for several years, multiple factors likely contributed to Gartner’s recent assessment:
- Uber’s fatal self-driving vehicle crash in Tempe, Arizona in March.
- Tesla’s continued troubles with its Autopilot software and inattentive drivers.
- Consumer surveys reporting increasing uneasiness with the technology.
- A deluge of new research and media reporting suggesting various components of AV tech are not nearly mature enough to support scaled commercialization.
I’ve spoken with dozens of industry experts and insiders since the start of the year, and there’s a clear and growing split between AV software developers and OEMs, and nearly everyone else, in terms of deployment readiness. The developers tend to align with the most optimistic assessments, and argue we truly are 2-3 years from seeing multiple large-scale deployments of level 4 AV technology. Industry analysts, people in the conventional freight and logistics business, policy wonks, and academic researchers have moved decidedly in a bearish direction. According to them, at best we’re looking at 8-10 years before mass market penetration, and for some that time horizon extends at least 15-25 years out. Level 5 autonomy is completely off the table.
While my own assessment leans towards the bear case (see my piece from March on AV economics), both sides have arguments worth considering.
Nobody understands the technology better than the people developing it, through their design and iterative testing of full stack AV architecture. Moreover, they have all the data about how their AI performs in both routine and edge case scenarios. While there are reasons for skepticism, few of these firms have backed off their aggressive deployment timelines. Waymo will launch a commercial AV ridehailing service later this year, deploying the first of up to 62,000 Chrysler Pacifica minivans it intends to purchase over the next few years. GM Cruise is still planning a 2019 commercial fleet launch, and now has significant backing from SoftBank’s Vision Fund. Several other firms, including Daimler, Uber, and Ford have commercial deployments planned between 2019 and 2021.
AV startups reaped more than $3 billion of investment in 2017, and are still hauling in large amounts of cash through the first half of 2018. Zoox is raising more than $600 million, and China’s Pony AI just hauled $100 million. AV trucking startup Embark just closed a $30 million round, following in the footsteps of competitor Starsky Robotics. This doesn’t include hundreds of millions invested in startups developing sensors and other components of the AV technology stack. SoftBank’s doubling down on the AV market through its $2.25 billion GM Cruise investment shows that some of the smartest money in the world is still betting big on AV momentum.
That all said, many of these firms have significant incentives to overstate the readiness of their technology. Some automakers have invested billions through internal R&D and acquisitions, and must demonstrate (a path to) value for shareholders. Startups developing AV tech need to either commercialize it through licensing, or sell off for the highest returns possible to investors. It’s also possible they have unbridled faith in their own abilities, and truly believe no obstacle ahead of them is insurmountable.
One recent, heavily-discussed piece highlighted shortcomings in current AI capabilities, citing seemingly-infinite edge case challenges and excessive expectations. Research is finally emerging on the energy consumption requirements for AV technology, and early findings suggest these information architectures still require significant refinement. Federal legislation also hit a standstill after building serious momentum throughout 2017.
Even Waymo CEO John Krafcik is urging caution about overly optimistic expectations for commercial services, and few serious analysts doubt Waymo’s lead over the field in deployment-readiness. At the Automated Vehicles Symposium in San Francisco, Lyft’s AV head Nadeem Sheikh painted a picture of very gradual progression towards scalable fleets of robotaxis. He explained that customer expectations and operating economics are likely to keep early AV fleets heavily geofenced, with coverage areas only expanding as the technology matures. Comma.ai CEO George Hotz goes a step further, arguing AV ridehailing will fall flat completely against superior (human-driven) incumbent competitors.
On the trucking side, my own conversations with industry experts have illuminated increasing skepticism for the near-term prospects of automation. Despite ever-pressing concerns over long-haul truck labor shortages and industry economics that are generally begging for automation, significant hurdles may significantly slow deployment. Peloton, Embark, and Starksy Robotics offer impressive solutions for automating trucking, but freight operators haven’t yet attempted to integrate these solutions into their standard operations.
Platooning offers the promise of 6%-10% energy use reductions (estimates vary) but only in ideal conditions. Congestion, cut-ins from other vehicles, navigating on and off highways, and drive time on local roads all reduce the efficacy of the technology. Moreover, drafting can create air flow challenges for radiators and tires, impacting both asset management and energy consumption. No startup working on more extensive automation is targeting elimination of labor altogether. Embark is working on a solution that combines highway automation with short-haul drivers to take over in denser metro areas. Starsky sees teleoperation as a solution for these situations, which would cut driver costs, but requires an entirely new set of capabilities in high-liability contexts.
Gartner’s assessment is fairly accurate, and we could be 3-5 years from reaching the “Slope of Enlightenment.” There’s too much money pouring into this space to see progress completely hit a standstill, but despite enthusiastic headlines, there will be little transformative impact. Industry firms will continue to forge R&D partnerships with AV developers, similar to what we’ve recently seen with Nuro/Kroger and Ford/Dominos/Postmates, but no commercialized partnerships will emerge in the short-term. AV fleets from Waymo, GM Cruise, and others may launch at impressive scale, and possibly even with paying customers, but will fail to dent the ridehailing markets in deployment cities.
For trucking specifically, long stretches of highway are the lowest hanging fruit, but volume/scale considerations make these dubious without large freight operators reengineering their supply chains. Segregating AV truck traffic through dedicated roadway facilities (managed lanes) would make automation, and specifically platooning, a more attractive investment for operators, but that’s a far-off prospect. Platooning will struggle to demonstrate any real value until operators agree to telecommunications interoperability, currently a point of resistance due to data sharing concerns. Otherwise, the trucking industry is simply too fragmented among owner-operators and other small fish to recognize serious highway travel efficiencies.
And that’s fine. These firms still have a lot to learn about urban geometry and traffic flow dynamics. Software developers still seem to think routing vehicles through a dense urban street grid is like routing packets on a network. Mixed vehicle traffic and multimodal interactions (bicycle, pedestrian, etc.), create many challenges that are very difficult to model in AI system logic. Though Waymo has compellingly argued against it, partial (or parallel) automation may offer additional safety or driver comfort without degrading quality of service. As hardware inputs grow cheaper, trucking firms may decide investment in AV tech is worthwhile even without labor reduction.
If the current phase of AV development has seen tremendous strides in creating functional technology that firms can safely introduce into uncontrolled environments, then the next will refine that technology for services that customers actually value.