Continuing my multi-part series on the recent Automated Vehicles Symposium, my most recent article provided an overview and key take-aways for the broader industry. Established in 2014, this non-profit conference is co-sponsored by the Association for Unmanned Vehicles Systems International and the National Academy of Sciences Transportation Research Board. Being non-profit and carefully curated to maximize content and minimize hype, I find each year’s event to be a touchstone on the state of automated driving. Here’s Part Two digging deep into truck automation discussions for “solo driverless” operations. Part Three will address AVS truck platooning content. Disclosure: I serve as a volunteer on the AVS Executive Committee and among the many companies discussed here, I am an Advisor to Plus.ai.
In just a few days, a substantial portion of the truck AV community appeared at AVS. As noted in Part One, many gave voice to an evolving consensus that long-haul automated trucks will be the first segment in the Automated Driving System (ADS) space to scale up. This was explored by moderator Selika Josiah Talbott of American University at the plenary session “Fact or Fiction? Why Automated Truck Fleets Will Lead the Way for AV Deployment.” Panelists were Çetin Meriçli, CEO of Locomation; Shawn Kerrigan, COO of Plus.ai; Chuck Price, Chief Product Officer at TuSimple, and Bala Ganesh, VP Advanced Technology Group at UPS, which is working with TuSimple. The presence of UPS was an AVS milestone—the first time a large trucking fleet active in ADS testing was represented—indicating how “the tech” is maturing towards providing services truly valuable to society.
The panelists spoke to the benefits of automated trucking, noting the potentially massive positive impacts on safety and efficiency. “Efficiency” encompasses many factors in addition to labor savings, including fuel savings from having the equivalent of an expert driver for every mile traveled, not affected by human foibles. Also, as Chuck Price noted, “Given no need to stop for rest breaks, a truck can cross the country in 36 hours, completely changing shipping choices. A time-sensitive load with that kind of delivery speed might be shifted from air to truck, which is likely to be more climate friendly.” In the past, others have spoken of automated trucks becoming more competitive with rail transport based on lowered costs, but Chuck raised an interesting point for specialized time-sensitive freight.
Selika Talbott asked about the timing of product launch of driverless highway operations. Most developers decline to speak about timing, but there are exceptions. Across the solo driverless companies, only TuSimple states a definitive date, which was reiterated by Chuck Price during the panel: driverless operations in 2021. This has caused some head-scratching from others who are concerned it could be risky to deploy a driverless truck prior to an OEM-produced SAE Level 4 tractor being available, which is expected in 2024. How will TuSimple safely field driverless in 2021 without a factory-built fully-redundant Level 4 ready tractor? Asking Chuck Price about this after the conference, he said “With the support of our ecosystem partners, we will be doing a retrofit with the necessary mitigations, redundancy, and fail operational measures, all at a very high level of safety integrity.”
The related question of how much on-road testing is enough was addressed in light of the 2016 “Driving to Safety” RAND study, which concluded billions of testing miles could be needed. The general consensus was that the RAND approach did not take into account important factors. For example, Çetin Meriçli said that “Most human crashes are ‘one off’ and near-term AV issues are ‘systemic’ problems based on an underlying deficiency in the system. So the system will make similar kinds of mistakes in similar kinds of scenarios, which can then be addressed during development.” Shawn Kerrigan emphasized the Plus.ai philosophy of independent validation as a key step towards deployment.
Startups and The Old Guard
Breakout Sessions were organized and moderated by myself and Andrew Krum, Group Leader, Human Factors & Advanced Systems Testing, at Virginia Tech Transportation Institute. Here, I moderated a panel titled “Solo Driverless — Development, Testing, and Deployment.” Truck sessions at prior AVS events focused on development and testing; this year Andrew and I saw the need to also focus on deployment of the technology. Most of the key startups in this space were represented. Panelists were:
· Robert Brown, Head of Government and Public Affairs at TuSimple
· Charlie Jatt, Head of Commercialization for Truckings at Waymo
Digging deeper on the deployment angle, Andrew Krum then segued into a panel titled “Implementation Strategies — Spanning Deployment to Operation and Beyond.” Panelists here were:
This impressive group of highly informed players had a lot to say. Over the last two decades, the traditional vehicle industry has extensively deployed Advanced Driver Assistance Systems (ADAS), resulting in practical learning about how to integrate sophisticated electronics into the harsh vehicle environment. The paradigm shift of automated driving has come primarily from the tech industry startups. This year’s AVS discussions made it clear that the integration between these two camps has accelerated dramatically. In the early days, startups “hacked in” to control a truck tractor so they could start testing their software. Now we’re seeing a broad symbiosis between startups, Tier 1 suppliers, and vehicle manufacturers. Efforts are turning to the challenges of integrating an ADS into commercial product.
Key indicators are a) partnering across OEMs-suppliers-startups and b) integration of ADS by OEMs. This is evidenced by the TuSimple-Navistar partnership, the first announcement that a truck OEM will build a Level 4 ready truck and also the first OEM announcement of Level 4 co-development with a truck startup. To put this in context though, I expect other OEMs are quite far along with their own similar activities and just haven’t seen a need to publicize it yet.
Can driverless trucks be retrofitted for automated driving? There was no argument on this point amongst our experts; it is do-able but not viable for scale. Close collaboration with OEMs is essential due to drive-by-wire dependence. It was noted for example that when a driver is present, controls must be fail-safe; with no driver, controls must be fail-operational. Discussing scale-ability, Stephan Olsen stressed practical factors as well, saying that “on a moving assembly line, installation of sensors for automation must be straightforward,” i.e. production friendly.
Maintenance issues are coming more into focus. Reliability is paramount in trucking, but there are limits. Bill Combs noted that, “Utilization could rise, since today drivers are limited by driving time. Automated vehicles will be utilized at a much higher level. A truck could potentially never stop driving, if engineered for this kind of use.” Penske’s maintenance business covers over 327,000 trucks across the freight industry. Through this lens, “maybe maintenance issues arise that we can’t now foresee” and the criticality of a repair takes on new meaning: “With ADS, unlike ADAS, replacing a windshield with a camera in it becomes a critical repair operation,” Combs said.
The concept of Remote Support, explored in an AVS plenary session, has taken root as core to initial AV deployment, at least in some form. While some view it as a bridge to full autonomy, others feel it will be needed for the long term as a core aspect of automated vehicle services. I find Pär Degerman’s assertion convincing: “Remote Support is here to stay, not just here for deploying quickly, it will be here for the remainder of our lifetimes. We’ll always have human beings in the loop to complement autonomy. This is a new way of operating.” Interestingly, each ADS truck developer is approaching remote support in their own way. Waymo, Ike, and Aurora have described their approaches online. This will be an important aspect to watch as early deployments occur.
We moderators posed what I call “the perennial question”: How do you know when your system is safe enough to transition to driver-out operations? Part One of this AVS series quoted Matt Schwall of Waymo speaking about their extensive process. As expected, our truck AV panelists made similar comments about safety process – referring to ISO26262 Functional Safety, ISO21448 Safety of the Intended Functionality, and UL4600 Standard for Safety for the Evaluation of Autonomous Vehicles, etc. Typically, AV companies can’t speak to the underlying details of their process; this is highly proprietary and can have liability implications. However, Pär Degerman noted that Einride is now operating driver-out at very low speeds on a limited route in Sweden, using a heavy dose of remote driving/assist.
AV Truck Regs Effectively Punted Into The Next Presidential Administration
FMCSA’s planned Notice of Proposed Rule Making (NPRM) to enable automated trucking will be vitally important. After an Advanced NPRM was released early last year resulting in significant comments from industry last summer, the NPRM was expected to be released by this summer. This did not occur. If released today, it is very unlikely to become a formal regulation prior to the swearing in of a new President, and things could change immensely with a new Presidential administration. The Advanced NPRM lays the groundwork for a much more activist USDOT if we transition to a Democratic Biden administration.
Is a big shakeup coming for the truck AV startups? This was the question posed to tech journalists in the session “Hard Truths from Journalism’s Best,” which consisted of Kirsten Korosec, Senior Transportation Reporter, TechCrunch; Joann Muller, Editor, Axios Navigate; and Alan Ohnsman, Senior Editor, Forbes. The moderator was Grayson Brulte of Brulte & Company. Several panelists saw a set of failures or acquisitions coming by the end of the year if truck AV startups don't have a manufacturing or significant technology partner. As Alan Ohnsman put it, “TuSimple has a production deal with Navistar. Undoubtedly Waymo has things in the works. There are a handful of specialist AV truck startups out there. It seems highly unlikely at year end that they will all still be with us.” As I noted in Part One, while year-end strikes me as too soon for major shifts, this basic expectation is widely held for some near-term timeframe. In followup interviews with startups, views on the tech journalists’ assertions differed. Shawn Kerrigan of Plus.ai said “Consolidation is expected as self-driving technology matures. Winning this race for self-driving trucks is not just about partnerships, but instead about commercialization and getting a real product to market. That’s been our focus and we firmly believe this is the winning formula.” Dan Goff of Kodiak Robotics weighed in to say, “It’s difficult to see a path forward for a self-driving truck developer these days without a great set of technology partners. That’s why building up our technology ecosystem has been a core focus for Kodiak since our founding.”
Much More Is Yet To Be Revealed
The truck AV industry is maturing. We are moving past the point of AV startups working alone to prove themselves. It is striking that the maturation discussed at this year’s AVS was not even in the discussion at the 2019 AVS.
Getting to early driver-out trucking appears to be driven by two models: (1) remote driving in very constrained situations and (2) retrofit for highway over-the-road operations. These retrofit approaches get the market started and allow fleets to learn and adapt. But it doesn't scale. In three years or so, the truck OEMs will be coming along with designed-for-Level-4 systems, enabling scale. And, while some companies have made definitive statements on launch timing, the internal non-public roadmaps of other startups (especially Waymo) may be equally significant as to what the future holds.
What about truck AV content at next year’s AVS? I expect we’ll have more truck OEMs present, possibly presenting along with their startup partners. And with offerings in Level 2 systems accelerating, I’m curious to see how Level 2 evolves and interacts with Level 4 capabilities in trucking. Potentially, we’ll see modest but significant increments in startups running freight driverlessly in benign settings. And we might even have some stories about driverless operations at full speed on our highways.
I fully agree with those who see the structure of this industry evolving rapidly. We may be past the days of seeing new startups popping up, but you can expect significant fund-raising, partnering, and product announcements in the coming months.
Amidst all of this upbeat chatter, the ADS developers must continue to be careful. With so many players running trucks, the risks of an autonomy-involved crash are real. In the trucking discussions at AVS, the company reps affirmed their commitment to best safety practices and diligence in taking each step towards driver-out product launch. As it should be; one testing day at a time—the road is bumpy enough as it is.