This year’s North American summer conference scene has been dominated by two major conferences. In late June, the 2026 Autonomous Trucking Conference, in its third year, brought together truck AV developers and a broad spectrum of stakeholders from the upper Midwest, as well as Canada, to Fargo, North Dakota.
In late July, the Automated Transportation Symposium, sponsored by SAE, convened in San Diego. The elder statesperson of the AV space, the ATS event in various forms has been meeting annually since 2012.
Part 1 provided insights about the Autonomous Trucking Conference.
In this Part 2, I cover the Automated Transportation Symposium.
To my knowledge, there is nothing like ATS anywhere else on the planet. The event brings together cutting-edge experts from autonomy deployers with public sector stakeholders (city, state, national) trying to get their heads around robots on the streets. It brings together peers from Asia, Europe, and North America. Academics abound, studying and conceptualizing what needs to be addressed and solved. And of course, no shortage of consultants like myself. At this year’s event, attendance was about 600 of these kinds of folks.
By my reckoning, AV deployers represented at ATS were Aurora, Bot Auto, Einride, Gatik, General Motors, Kodiak, MOIA-America, PlusAI, Stack, T2, Tesla, Torc Robotics, Uber, Waabi, Waymo, and Zoox.
Representatives from states included Arizona, California, Ohio, Maryland, Michigan, Minnesota, and Texas, as well as the District of Columbia.
Government representatives from other countries included Canada, Germany, Japan, and the UK. Additionally, the European Commission was represented.
(I would welcome a note apprising me of any omissions to these lists.)
A few “big dogs” provided their viewpoint and prospects for autonomy. The Administrator of the USDOT National Highway Traffic Safety Administration, Jonathan Morrison, provided a meaty update as to what’s on their plate. This was a quick in-and-out but NHTSA’s Deputy Administrator, Lee White, also gave a talk and was on hand for several days. Not only that, Michael Graham, Vice Chairman, National Transportation Safety Board, spoke. On the deployer side, Uber’s Global Head of Autonomous Mobility & Delivery, Sarfraz Maredia, was in the house, as well as Zoox Co-Founder and CTO Jesse Levinson.
A nice addition to the ATS experience was having one of the first MOIA vehicles in the USA on display.
After many years with various sponsors, the tie-up with the Society of Automotive Engineers feels like a perfect match – because SAE is all about vehicle engineering when it comes to cars, trucks, and industrial vehicles in the commercial space.
The event was an odd transition for the San Diego waterfront! Just prior to ATS, Comi-Con overtook the city, with hordes of fun and freaky-dressed acolytes of the comic book, pop culture, movies, television, and gaming scene. Then, poof! Overnight Comic-Con was gone and the waterfront transitioned to normal looking people speaking about what was once fantasy as well – self-driving vehicles. … except for a strange overlap as Comic-Con hangers-on mingled with ATS’ers on the sidewalks in front of the Hilton San Diego Bayfront Hotel.
In pulling together all the information from the event to provide this summary, I have lifted quite a bit of text from the program and the breakout summaries. Much gratitude to the organizers and the breakout leaders for providing such good material.
Get a refreshing drink and a comfy chair for your late August reading – there’s a mountain of content here from this unique and energetic event.
This tome will first cover the plenary talks, then perspectives from ATS participants, followed by a review of the breakout sessions. The article concludes with an acknowledgement of the ATS Executive Committee.
Reader comments are most welcome – how useful is this long form review?
SUMMARIES AND PERSPECTIVES ON PLENARY TALKS
Executive Chats
The discussion with Uber’s lead on autonomy, Sarfraz Maredia, was moderated by Jane Lappin, with Blue Door Strategy and Research LLC. The discussion with Jesse Levinson, Zoox’s Co-Founder and CTO, was moderated by Aarian Marshall, a WIRED journalist who has covered the AV space for many years. Both companies play a key role in the AV ride-hailing space and it is a feather-in-the-cap that their executives appeared at ATS.
As an audience member well versed in the AV space, I didn’t expect new content, but I was alert for interesting nuance. The moderators did a fine job in putting forth questions and probing for further elaboration. For both discussions, I came away with no takeaways… only because executives of this sort limit their comments to what’s already been said publicly. No criticism, just reality. Many of my colleagues in the room might have had a similar experience, but I’m sure those newer to the AV scene found this highly interesting.
Regional Updates
From Pilots to Scale: What’s Next For Autonomous Mobility in Europe?
Europe has spent more than a decade building a pan‑European ecosystem for research and innovation in connected and automated mobility (CCAM). Now, the tide is turning from testing technology to large‑scale demonstrations and the operation of real-world automated services. Dr. Stephane Dreher, Head of CCAM, for ERTICO - ITS provided an update on EU policy and regulation, including how recent European actions to strengthen automotive competitiveness are accelerating innovation, shared software and hardware building blocks, and cross‑border deployment pathways. The activities link to the current portfolio of EC‑funded Horizon Europe initiatives under the European CCAM Partnership. Dr. Dreher also covered deployment momentum and practical implementation challenges in Europe, i.e. what it will take to move from pilots to sustainable, large‑scale operations across regions, countries, and regulatory frameworks.
The European Commission and individual member states have in the past years provided most of the funding for CCAM activities. In the last couple of years, we’re seeing robo-taxi and robo-shuttle deployments which are purely commercial. The tide is shifting and scale no doubt is coming.
China!
Two intrepid adventurers happened to be in China not long ago, with a mission to experience rides in as many robotaxi brands they could wrangle. Bryant Walker Smith of the University of South Carolina and Jon Qian of Valgo narrated a fun and surprising set of video’s of what worked, what didn’t, and what perplexed. All in all, my takeaway was that the offerors in China had their act together, even though glitches occurred (just like in the U.S.).
Cities and States
D.C. Department of Transportation Preparations for AV Deployment
The District of Columbia government has had several robotaxi companies knocking on their door seeking to allow AV operations there. So far, no dice. But Stephanie Dock in the District Department of Transportation has played a lead role in determining what AV deployment could mean at an operational level. In this talk, she shared the unique challenges the District is facing.
Panel Discussion: Ready, Set, Go! City and State Readiness and Resilience for ADS-Operated Vehicle Deployment
This panel of state and city leaders explored bridging operational, infrastructure, and policy gaps, moving beyond general readiness toward concrete practices that build resilience across public and industry partners. Chaired by Danielle Chou of FHWA, perspectives were provided by states: Michele Mueller (Michigan Department of Transportation) and Tara Olds (Minnesota Department of Transportation), and cities: Jarvis Murray (Los Angeles Department of Transportation) and Rachel Castignoli (City of Austin).
Panel Discussion: Disaster Response: Are We Ready?
Building on last year’s breakout discussions, this session asked: what happens in a disaster when a significant portion of our vehicle population is AVs? Experts considered disaster scenarios from different perspectives during the Plenary. And these guys have jobs that no doubt keep them up at night!
The panel consisted of Mary Ellen Carroll, Executive Director Department of Emergency Management, San Francisco; Piet van Os, Head of Enterprise Resilience, Waymo; and Pamela Murray-Tuite, Professor at Clemson University.
The session was enhanced by a scenario board at one of the receptions which allowed attendees to contribute to the global conversation in this area.
Public Sentiment
Public Attitudes and Perceptions of the “Careful and Competent” Driver Benchmark: Findings from Survey Research in the UK
The UK Automated Vehicles Act of 2024 established a benchmark requiring AVs “to achieve a level of safety equivalent to, or higher than, that of careful and competent human drivers.” But without a commonly recognized measure, the UK Department for Transport commissioned a public survey to better understand public perception of the “careful and competent” benchmark. Dr. Nick Reed of Reed Mobility presented findings from this study which are no doubt interesting to road agencies in other countries.
The study included workshops with the public, asking “What behaviours positively characterise a careful and competent driver?” The participants ranged across drivers, non-drivers, as well as setting (rural / semi-urban / urban). The participants were queried initially, then learned more about the topic from experts, followed by a second survey.
Independent certification or a regulator’s audit was by far was the choice of participants. They focused on the behavioral expectations of AVs.
Recommendations from the study were:
• More concrete definition of “Careful and Competent” is needed
• Performance must be much higher than an average driver
• Outcome metrics / behaviors differ by ODD
• “Companies cannot ‘mark their own homework’”
• Independent oversight / transparent communication is expected
• Trust likely to grow over time
Looking towards future research, Dr. Reed asked, “what behaviors and messaging are most persuasive for members of the public?”
AV Operations
Panel Discussion: Opening the Garage Door to See What’s Happening Inside AV Depots
AV depots provide service and energy to ensure that all systems are operating safely as designed. Moderated by Miguel Acosta, California DMV, panelists Ming Maa (Moov), Brenda Mejia (Gatik), and Thomas Ashley (Voltera) talked about the operations, and explored the myriad trade-offs that companies must make when seeking the optimal location for their depots.
Fireside Chat: Remote Operations: Headlines, Reality, and What the Difference Means for Public Trust
Moderated by Ed Straub of Velouria Consulting, speaker Erin MCurry of the SAE’s Automated Vehicle Safety Consortium provided context for the non-expert stakeholder addressing high-level issues, needs, and approaches associated with the many, various aspects of ADS remote operations. A very important session for the city and state players in the room who want to understand this behind-the-scenes world that nevertheless affects their daily jobs.
AV Trucking
The Long and Short (Haul) of Automated Trucking
Moderator Andrew Krum of VTTI pulled together some of the leading companies in autonomous trucking. Rich Steiner (Gatik), Jordan Coleman (Kodiak), Chan Lieu (Aurora), and Baptiste Le Poittevin (Einride) quickly jumped into a lively conversation exploring the different use cases represented and how these interact with various business models. In addition, there was discussion of ICE versus EV powertrains, giving that Einride vehicles are exclusively EV’s.
Japan’s First Level 4 Transport Network
Autonomous trucking is coming to Japan via a new company called T2. Hiroshi Iwai, Director of Solution Planning at T2, provided highlights from their experience operating long-haul transport on the Tokyo-Osaka expressway where the highway agency is constructing AV Priority Lanes.
T2 started last summer with L2 operations and are now running 21 vehicles in commercial service. As Mr. Iwai put it, “we have driven the equivalent of ‘six earths’ during testing, demonstration, and commercial operations.” will transition to L4 over time.
L4 testing is already underway. In March of this year their vehicle ran autonomously from Ayase to Kobe, a journey of 550 km. T2’s autonomy is sourced from PlusAI.
Their timeline shows L4 operations beginning next year, with 500 vehicles operating by 2030. T2 will purchase L4-redundant chassis from truck manufacturers and upfit their AV tech to operate as Transportation-as-a-Service.
Federal Government
National Transportation Safety Board
Michael Graham, Vice Chairman of the NTSB reviewed the NTSB’s approach to the findings and recommendations related to the Ford Blue Cruise investigation, and how the ADS industry can learn from this investigation. This was a good talk. It’s a shame NTSB hasn’t posted the comments on their website.
USDOT NHTSA Update: Lee White, Deputy Administrator
(summarized from original comments)
Mr. White noted that he joined NHTSA about a month ago, coming over from the USDOT Office of the Assistant Secretary for Research and Technology (OST-R). He opened by sharing “some of the incredible forward-looking innovation happening at the departmental level across USDOT.”
Modernizing National Infrastructure: The Digital Corridors Initiative
The Digital Corridors Initiative (DCI) is a federally assisted, state-led effort to modernize the nation’s interstate highway system through large-scale deployment of Transportation Digital Infrastructure.
Jointly led by OST-R and the Federal Highway Administration, DCI rests on a true three-way partnership between USDOT, the states, and private industry, facilitated by the American Association of State Highway and Transportation Officials (AASHTO), the Transportation Research Board (TRB), and ITS America.
The initial use case is “simple yet critical:” achieving real-time work zone and lane closure data sharing and alerting, end-to-end across interstate corridors. Ten state DOT Chief Executive Officers volunteered as “anchor states” to champion these corridors: California, Delaware, Florida, Georgia, Iowa, Minnesota, Nevada, Oklahoma, Kansas, and Texas. The architecture uses a concurrent top-down and bottom-up strategy—combining federated data architectures nationally with active local projects on the ground.
The goal is to demonstrate end-to-end data exchange on a primary corridor by Spring 2027, scaling to nationwide deployment across multiple interstates by Spring 2028. Going forward, DCI will expand to cover 911 dispatch alerts, roadside incidents, rail, freight, and port operations.
Automated Shuttle Pilots in National Parks
Mr. White noted that USDOT recently awarded a project in partnership with the National Park Service and the Department of the Interior to launch automated shuttle pilot programs at Yosemite and Acadia national parks, beginning in summer 2027. These pilots will test “American-made” autonomous transit in controlled park environments. These will be time-limited evaluations designed to tackle congestion, improve accessibility, protect natural resources, and gather real-world performance data.
SMART Grants
At the community level, USDOT established the Strengthening Mobility and Revolutionizing Transportation (SMART) grant program. This is a two-stage program, starting with demonstration projects and then onwards to scaling grants. To date, SMART has awarded $270 million to 122 projects across 45 states, DC, and Puerto Rico.
Examples are:
· Shreveport, Louisiana (SporTran EMPOWER Project): Partnering with labor to deploy AV shuttles and modernize public transit.
· Canton & Youngstown, Ohio (SARTA): Integrating Advanced Driving Assistance Systems (ADAS) directly into public transit buses.
· Fort Worth, Texas: Deploying micro-weather sensors to stream safety-critical data directly to automated vehicle and drone fleet operators.
Direct Industry Opportunities & Practitioner Exchange
Mr. White noted two immediate opportunities for industry and public-sector engagement that he wanted to highlight.
Cooperative Driving Automation (CDA) Prototype Broad Agency Announcement: this BAA focuses on a Cooperative Driving Automation Small-Scale Industry Prototype, offering up to $5 million across awardees. Proposals are due in mid-September. USDOT are seeking OEMs, Tier 1 suppliers, and Infrastructure Owners/Operators to submit proposals to test connected AVs in real-world partnerships.
Inaugural Automated Vehicle Peer Exchange: in partnership with the San Francisco County Transportation Authority and the University of San Francisco, USDOT is hosting the inaugural Automated Vehicle Peer Exchange in San Francisco on September 9–11, 2026. The event will convene over 50 practitioners from leading states and cities to share hands-on deployment experiences, synthesize best practices, and set the baseline for future federal peer exchanges through 2027.
Turning to his new role at NHTSA, Mr. White focused on safety, transparency, and public trust. “It’s true that being in the AV space right now can feel like living in a fishbowl. The public and media are scrutinizing every deployment. Every viral video showing an AV misinterpreting a road rule or making a mistake chips away at public trust. You only get one chance to make a first impression, and once public trust is eroded, it is almost impossible to regain.” Yet, “Automated driving systems hold incredible promise. They stand to unlock unprecedented freedom and mobility for senior citizens, people with disabilities, and anyone unable or unwilling to drive. Access to transportation is, fundamentally, access to freedom. Furthermore, AVs can directly address our national roadway fatality crisis by eliminating risky human behaviors like speeding, distraction, and impairment.”
“Building that trust requires direct cooperation with local communities,” he said. “NHTSA regularly hears from city leaders who feel rollouts are moving too fast without enough operational transparency. If you want your technology to be embraced, bring local officials in early, communicate openly, and demonstrate that your systems add value to the community rather than creating a hazard.”
But he noted that the most urgent issue is AV interactions with first responders. “NHTSA recently issued a public call to action after documenting multiple concerning incidents where AVs entered active emergency scenes, blocked fire engines and ambulances, or failed to recognize basic hazard signals like flares, smoke, cones, and flashing lights. An automated vehicle that cannot safely interact with emergency personnel is a danger to the public—full stop.”
He noted that NHTSA was “pleased to see one of the L4 providers step up after the call to action and announce a fleet recall to update their software to address a concern with detection of heavy smoke, which can be present in an active emergency scene. That is how the oversight process is supposed to work.”
Mr. White emphasized that “NHTSA will not hesitate to actively exercise its enforcement authority for any developer that fails to proactively address safety concerns. Safety must be the foundational component of all of this innovation.”
Speaking to the ATS audience, he said “It is on the people in this room - OEMs, technology developers, and government leaders - to keep collaborating together to succeed. Our emergency responders are key stakeholders in this process, and we all must work together to ensure that trust is developed and improved at all levels. We have an exciting future!
This truly is the golden age of transportation.”
USDOT NHTSA Update: Jonathan Morrison, Administrator
(summarized from original comments)
Administrator Morrison’s comments were packed with key perspectives on AV’s as well as new announcements. Here I focus on a key areas. The full text is at NHTSA’s Speeches page.
Mr. Morrison made it clear that the Agency is supportive of AV’s: “A well-developed, safely deployed automated driving system offers tremendous potential benefits for our nation and our people. It is not an exaggeration to say that the potential changes to personal mobility enabled by automated driving systems are greater than any technological innovation since we’ve moved from the horse and buggy to the automobile.”
Nevertheless, he spoke sternly to the AV cognoscenti: “Whether we reach this promising future depends on each of you. This technology needs to be developed appropriately, safely, and responsibly. Whether AVs succeed or fail, whether they become the transportation of the future or a passing fad, that’s up to you.”
NHTSA’s approach is one of balance, he said, combining policy choices that enable innovation with an unapologetic exercise of our oversight and enforcement authorities. “We believe this approach will make our vehicles both safer and more innovative than ever. And we are moving quickly on both fronts. I call it deliberate urgency – moving ahead with purpose but at a pace that never puts safety second.”
He outlined the Federal Automated Vehicle Framework, which is based upon three pillars:
· Prioritizing the safety of ongoing AV operations on public roads.
· Enabling the commercial deployment of AVs to enhance safety and mobility for the American public.
· Unleashing innovation by removing unnecessary regulatory barriers to the safe development of the technology.
First Pillar: Prioritizing Safety
Regarding the first pillar, safety, his stance was clear: “If we see unreasonable safety risks, you’re going to hear from us.” He zeroed in on particular concern, echoing Mr. White’s comments: “Over the past few months, we’ve identified a clear pattern of AVs interfering with law enforcement and other first responders, and so I issued a call to action to the AV community to develop solutions to these far-too-common issues. We expect you to prioritize first responder interactions, and we’ve already seen very positive responses from many developers—including issuing a recall.” Several discussions are underway between NHTSA and AV developers.
“We won’t hesitate to take further action as we see issues in the field—be that opening defect investigations for manufacturer-unique safety issues, or calls to action if we see general industry concerns—which may precede defect investigations into individual developers. Our balanced approach means we can conduct oversight while we remove barriers to innovation without compromising safety.”
“But it’s not fair or good public policy to play whack-a-mole and taking enforcement action while remaining silent on expectations. This is why we’ve been so active in engaging with industry, our co-regulators in State and local government, safety experts, and first responders, and why we’re focused on developing guidance.”
AV Guidance
Us old-timers were pleased and gratified to see NHTSA issue a series of “AV Guidance” missives during the mid-2010’s. This added important momentum to the space.
Noting the community looks to NHTSA and the federal government for leadership and guidance on AVs, Mr. Morrison said, “Our current voluntary guidance document, AV 2.0, dates back to 2017, and the technology and our understanding of it has changed dramatically since then. As risks have evolved, so has the need for updated Federal guidance. We’re moving ahead with developing new guidance, which we expect will help shape the industry’s approach to safety for years to come. This updated guidance aims to support the industry’s strategic transition from test-centric operations to mid-scale driverless deployments. The guidance focuses on critical safety elements, such as emergency responder interactions, safety management systems, remote assistance, crashworthiness, and post-crash behavior.”
NHTSA is soliciting feedback on proposed new draft technical chapters for the Guidance. “I signed a Federal Register notice announcing the creation of a docket to be open for one year. We’ll be issuing individual chapters of draft guidance, which will be compiled into one document after reviewing and reacting to the comments we receive. We want to hear from you on all aspects of this proposed guidance.”
Second Pillar
Within the AV Framework, NHTSA’s second is “enabling the commercial deployment of AVs to enhance safety and mobility for the American people.”
Federal law authorizes NHTSA to grant exemptions that allow manufacturers to produce vehicles for a limited period that do not comply with certain Federal Motor Vehicle Safety Standards, or FMVSS. “Today,” he said, “we announced that NHTSA is granting the country’s first-ever commercial deployment exemption for a novel-designed automated vehicle for passenger service. NHTSA is granting an exemption to Zoox for its purpose-built robotaxi, which is designed without manual controls or mirrors and has a “campfire” seating arrangement instead of a traditional forward-facing cabin. We are granting this exemption because Zoox demonstrated that their vehicle design has an equivalent level of safety as a vehicle that meets all FMVSS. Given the unique nature of the vehicle and the fact that it is operated by an automated driving system, the grant is subject to conditions that provide increased operational transparency and provide the agency with enhanced oversight.”
With Zoox’s exemption, he said NHTSA is establishing an “operational authorization.” He noted that the exempted vehicles will be subject to continued operational oversight, while also prescribing specific operating conditions that can be tailored to different operating environments. “These authorizations offer much more flexibility, allowing the agency to modify approvals as appropriate,” he added. The Zoox exemption is NHTSA’s first “operational authorization,” which is “tailored to the company, its robotaxi, and the ADS’ existing capabilities.”
Pillar Three: Removing Unnecessary Regulatory Barriers
While the Vehicle Safety Act requires that manufacturers certify vehicles to comply with all applicable Federal Motor Vehicle Safety Standards, “these are really hard to draft in the objective manner demanded by Federal law, which doesn’t allow for the types of subjective standards you see in Europe. The Vehicle Safety Act included an exemption process in large part in anticipation that Federal Motor Vehicle Safety Standards cannot, and should not, be amended so quickly to keep pace with every potential vehicle design. An exemption allows market access to innovative technologies, and the agency to learn from deployments. And we are all aware the existing FMVSS were not written with AVs in mind.”
Thus, NHTSA final pillar aims “to remove unnecessary regulatory barriers to the safe development of automated vehicle technologies, and thereby unleashing innovation that the United States creates better than anywhere else,” he said. “We are working diligently on much-needed FMVSS updates to account for AVs without compromising safety. While I tend to believe FMVSS operate best when establishing performance requirements, our current safety standards require certain pieces of equipment.”
“You’ve likely heard about our recent proposal to amend FMVSS 135 to remove the requirement for a brake pedal in a vehicle designed never to be operated by a human.
If you have a vehicle that’s designed never to be operated by a human, it simply doesn’t make sense to require a brake pedal. What’s important is that the vehicle meets reasonable stopping distance requirements. A brake pedal isn’t just unnecessary in a fully driverless AV; it could actually be a safety hindrance if misused by a passenger.”
Continuing, “Our approach isn’t to eliminate rules for AVs or say they’re exempt from braking requirements—that would be using a hatchet, when what we really need is a scalpel. We propose that purpose-built vehicles must still meet braking requirements. By updating the rules to prioritize the safety performance requirements over design specifications, we restore logic to the FMVSS in line with design of the Vehicle Safety Act. This is common sense. When the predecessor to FMVSS 135 was created, NHTSA didn’t envision a day when an automated driving system could control all aspects of the driving task without a human. And we shouldn’t have to—we don’t have a crystal ball to predict how automotive technology may change in the future—and by orienting our standards toward performance requirements, we won’t need to.”
In face, NHTSA has published Notices of Proposed Rulemaking for five FMVSS, including the braking requirement. “With no possibility for a human driver, there would be no safety need for a transmission shift display, windshield defrosting and defogging systems, or wiping and washing systems engineered to support a human driver. We have also proposed removing the requirement for a tire and loading informational placard at the driver’s door. After all, there’s no driver, and developers should be allowed to place the placard where it makes the most sense for that specific vehicle. These are the sorts of iterative changes to FMVSS that reflects NHTSA’s approach today, moving with a deliberative sense of urgency, but looking to get things right.”
He noted that NHTSA also plans to amend FMVSS 111, the requirements for mirrors and rearview image displays, saying that “Mirrors are there to help drivers see objects and people around vehicles. An ADS, however, has no need for these mirrors.”
“In the same spirit, we’re planning to update FMVSS 108, which addresses vehicle lighting. Clearly, every vehicle needs lights for visibility and conspicuity. That’s a clear safety regulation and benefit. However, with no human driver, an ADS wouldn’t need manual controls specified in the standards—such as turn signal stalks. But the way the FMVSS is written right now, a vehicle must have manual controls to meet the requirements. So that’s something we are looking to modernize to remove barriers that don’t have a safety need.”
But sometimes it gets tricky. “One standard we’re working on is FMVSS 101—controls and displays, which specifies requirements over telltales and indicators. While some may argue that telltales are designed to inform the human driver of underlying conditions and therefore aren’t relevant for vehicles designed never to be operated with a human in mind, others can observe that certain telltales provide information necessary to provide human occupants with agency over deciding whether to ride in the vehicle—think airbag malfunction telltales. We look forward to delivering a thoughtful proposal and taking comments from the public.”
ADS Competency: ASCEND
While all of these FMVSS updates involve specific performance tests to ensure compliance with applicable safety standards, the Administrator noted that “there’s a larger question people are naturally asking as ADS vehicles deploy – how do we know if an AV is a safe driver?”
“NHTSA plans to seek public comment on behavioral competencies for AVs and how to assess the safety performance of these vehicles. Any competency requirements would need to be objective, practicable, and repeatable, just like all of our FMVSS. We also plan to establish metrics associated with each behavioral competency, and methods and means of assessment.”
Since announcing this initiative in March, NHTSA has initiated work to develop AV competency tests, including reaching out to AV companies inviting them to NHTSA’s Vehicle Research and Test Center in Ohio to participate in the effort.
“But developing competency requirements in a vacuum or with limited external input would almost be assured to yield imperfect results—we need to expand the conversation. So, I am very pleased to announce that we’re entering into a cooperative agreement with SAE to launch the Advancing America’s Safety and Competitiveness for the Evolution of National Deployments, or ASCEND, consortium,” he announced. This is a joint government and industry initiative to expedite the development of data-driven AV performance standards and directly support NHTSA’s AV Framework. “Our top priority for the consortium is ADS competency, which we hope will inform and accelerate a corresponding FMVSS,” said Mr. Morrison. The goal is to complete this work within the next three years, including development of competency standards within the first year that will support near-term rulemaking.
NHTSA has launched a new one-stop shop online for AV resources.
Safety Regulations
Panel Discussion: Alternative Approaches for Regulating the Safety of Automated Driving Systems
This panel discussion built on the keynote presentation from NHTSA Administrator Morrison by adding brief presentations summarizing the regulatory approaches developed by the UNECE WP.29 (World Forum for Harmonization of Vehicle Regulations) and the State of California. The in’s and out’s of these important documents – seen as a historic milestone – were discussed by government representatives and the automated driving industry aimed at identifying opportunities for harmonization.
Moderated by Matt Wansley, Cardozo School of Law, panelists were Stacy Balk, NHTSA; Richard Damm, Federal Motor Transport Authority Germany; Jeff Farrah, AVIA; and Emily Warren, CalSTA.
Among other topics, behavioral competency was discussed. Germany has a requirement for storing incident data as a minimum requirement for the ADS safety case to prove an ADS meets the requirements in the ODD.
As to first responder issues, Ms. Warren noted that emergency geo-fencing can be issued by the city as “an ounce of prevention.” Mr. Farrah noted that AVIA has engaged with first responders from the early days.
Safety Assessment
How Can Third Party ADS Safety Assessors Enhance Public Understanding of ADS Safety?
This panel brought together three independent auditing and safety assessment organizations to discuss their use of data to support different types of evaluations of the safety of ADS systems they perform to help earn public confidence in ADS safety. Dr. Steve Schladover, retired, UC Berkeley, moderated a chat with Eric Teoh, IIHS; Bijoy Bhaskaran, TÜV SÜD Americas; Peter Janevik, AstaZero.
Artificial Intelligence and Security
Panel Discussion: Trustworthy AI for ADS and ADAS: From Safety Benefits to Safety Assurance
This session explored how AI enhances automated driving safety and robustness while examining the testing and evaluation frameworks necessary for credible safety assurance and trustworthy deployment. Moderator Jiaqi Ma of UCLA held a discussion with experts Manmohan Chandraker (UC San Diego), and Dr. Jan Reich, Fraunhofer Institute for Experimental Software Engineering.
The State of Cybersecurity
Summer Fowler of Exponent provided a review of the current state of vehicle cybersecurity, including the challenges in hardening ADS against remote hijacking, sensor spoofing, and adversarial attacks.
This is a vital technical field which is only understood by a relatively small cadre of experts. Even fewer can address automated driving. Yet folks in the broad AV community want to have a sense of what’s-what. Its tough to strike the balance between being “too high-level” (and boring) and “too detailed” where the audience gets lost.
Mr. Fowler did a yeoman’s job of finding the right balance.
The Business of AV
Panel Discussion: Scaling the AV Industry: Perspective from Investors and Operators
John Casesa of Guggenheim Partners moderated a panel consisting of Joseph Heyer (Stack AV), Dagan Mishoulam (Rivian), and Mark Norman (FM Capital).
Every year, ATS brings in the perspectives of investors. This year’s panel had three investors and one recipient of investment, namely truck AV developer Stack AV. What really happens among these players is never seen in the light of day, until an investment is made or an IPO is announced. So,the energetic back and forth was useful for audience members who don’t hang out in this space.
The Final Word: Business Journalist Perspectives on the Road Ahead for Automated Vehicles
Jane Lappin of Blue Door moderated this panel of journalists, namely Alan Ohnsman (Forbes), David Welch (Bloomberg), Junko Yoshida (Independent), and Harry Campbell (Driverless Digest).
ATS also brings in the voices of key journalists each year. While the investors have deep in-house knowledge with some AV startups, for the journalists it’s the other way around. They sort through press releases and interview material to gauge what’s really happening behind the scenes. They’re also taking a big picture view, examining all that’s happening with the gatekeepers, i.e. the public authorities. Based on years of experience in covering this space, the panelists had plenty to say. It was a late July snapshot of where the industry is going. But this is a fast-moving field. If they were to convene today, several weeks later, it would all be different.
VOICES: PARTICIPANT’S IMPRESSIONS
Jeremy Nassau, AVantage Insight, Australia
“ATS brought together the leading companies and government organizations to highlight the latest progress in AV rollout and work through the challenges they are running into as AVs are now rapidly scaling up. What struck me was the collaborative approach from so many incredibly smart people openly acknowledging the key issues and working together to figure out the best solutions. This year I had a particular interest in first responder interaction with AVs, which was covered in some depth both from the AV, government, and first responder perspective, and outlined both the current approaches as well as the inherent difficulties. Coming from a country that is further behind in adoption on-road, this was one of the best ways to understand the latest state of play in AV.”
Paul Schmitt, Reynolds & Moore, USA
“Having attended four ATS conferences, I continue to be struck by the constructive, cross-disciplinary environment it creates. Safety experts, policymakers, academics, standards developers, and industry leaders can set aside their usual roles and work through the hardest questions in autonomous mobility together. Two moments captured that spirit for me: a standing-room-only discussion on safe AI that pushed beyond technical capability into assurance and public trust, and an interactive workshop that explored how autonomous-vehicle behavior must be guided by a broader social philosophy, not only the rules of the road. ATS brings diverse voices together around a shared goal: realizing the promise of autonomous driving for society.”
Henriette Cornet, Urban Innovate, USA
“ATS is highly valuable to me because it brings together my entire network within the automated transportation sector. It is a fantastic opportunity to connect in person with remote colleagues and build new professional relationships. This year’s highlight was the FTA-Volpe session on automated transit, which is my primary area of interest. I particularly appreciated how they structured the session to include diverse perspectives from industry, academia, and government agencies. Moving forward, I would love to see more sessions facilitating this kind of dialogue between cities and mobility providers.”
Tom Alkim, MAPtm, Netherlands
“ATS is a great symposium where you can find policy makers, industry and research, just like the European (EUCAD) and Japanese (SIP-adus, MIWJ) equivalents, but ATS is the OG. I like the fact that it’s in California, so relatively close to Silicon Valley which allows me to combine it with side visits and rides.”
“Flying from Europe, I initially landed in San Francisco. This year, when walking the streets in SF there were undeniably more AVs on the road than the previous years where I was always scanning to see if I could spot a Waymo, Cruise, Zoox or other vehicle. It seems that there was one around each corner, proof that deployment and commercial use is real. In just two days I had an about seven rides with the new Ojai, the “conventional” i-pace and a Zoox. A highlight of the symposium was the announcement that Zoox got the permit to start commercial operations. With each year that I’m coming to the US, the smoothness and assertivity of robotaxis increases, although I had a pretty interesting situation in LA where our Waymo could make an unprotected left turn because there was so much traffic (and the time between getting the red light and the cross traffic that got the green light, actually white… was very short). After three failed attempts remote supervision kicked in and the support person tried to help us make the turn bit also failed a couple of times before deciding to go straight. But honestly, this is a situation where humans drivers are also struggling and it seems more of a traffic engineering problem than an automation problem, because in reality you are relying on bending/breaking the rules to get going.”
“This brings me to one of the standout topics that was much discussed this year, remote monitoring and support, a necessity in my opinion. Being able to deal with first responders and overcoming challenging situations calls for this kind of fallback mechanism. It’s encouraging to see how Waymo has a whole program dealing with this, including outreach and training.”
Jane Lappin, Blue Door Strategy, member of the ATS Executive Committee
“ATS isn’t an industry association meeting, not a commercial meeting, not a meeting produced by a couple/few people hoping to build a reputation. We’re independent of the industry, independent of government, unaligned with any particular world view. We simply aim to convene across sectors and borders to address critical shared issues affecting the future of automated transportation.”
“I was very happy with ATS26. We spend six or seven months developing the agenda — considering current issues, identifying speakers, staying focused on our audience, the industry, evolving societal issues, and safety. We enlist creators, critics, and regulators for their input to the program, and try to create the right balance -- highlighting industry advances without being too commercial and bringing light to societal impacts while remaining nonpartisan. We strive for gender balance on the program, and largely succeed, but it can still look lopsided. And then there are the last-minute changes — a sick child, a death in the family, cancelled flight, cancelled visa. It’s a big lift, and we’re never entirely sure how good it’s going to be until we see it on stage. This was the second year that we partnered with SAE, and we’re definitely on track for a solid future together.”
“This year we began looking “more like us again,” with increased levels of participation from Europe and Asia, which has been steadily growing post-COVID. I’m thankful to people like Henriette Cornet, Urban Innovators, Tom Alkim, MAPtm, Stephane Dreher, ERTICO, Ryota Shirato, ITS Japan, Manabu Umeda, Tokyo University, Nick Reed, Reed Mobility, and to the Japanese OEMs, who have been great supporters. We’re developing new relationships with Chinese innovators who came this year for the first time.”
“I can’t say enough good things about the support we get from our financial backers. For example, Aurora, Waymo, Southwest Research Institute, and Virginia Tech Transportation Institute back us year after year and tolerate it when the program content is critical, as it must be. USDOT has been centrally engaged from the start in 2012, working with us to develop the agenda, organize sessions, and sending teams of people to participate in the program.”
“We got very lucky when the SAE board hired new CEO, Reuben Sarkar. He comes from the American Center for Mobility, so he’s knowledgeable about AV challenges, and he’s spoken at the symposium pre-COVID on behalf of the Department of Energy, so he knows us. It’s great to have an advocate at the head of SAE who sees ATS as a strategic contributor to the organization.”
“I’m looking forward to a bigger meeting in July 2027. We’ll be back in San Diego, which everyone enjoys, and we’ve got a seasoned planning group with very strong breakout session organizers. I want ATS27 to be bigger, but not too big! I’m aiming for 800 in 2027 — big enough to be self-supporting, but small enough to feel like you’ve met mostly everyone.”
ATS BREAKOUT SUMMARIES
This section liberally pulls from text in the ATS breakout reports, which were written by the army of breakout organizers.
POLICY AND PLANNING
Capital A(V): AVs Have Their Day in the State Legislature
This intriguing breakout session was cast as a mock state legislative hearing in the fictional “State of Jefferson,” putting industry representatives, safety advocates, labor groups, and community voices in the witness chair. With AV developers seeking to deploy robotaxis and autonomous trucks on the state’s roads, the Jefferson State Senate’s Transportation and Infrastructure Committee has called a hearing to hear from experts and the public, with witness testimony is grounded in real-world expertise and drawn from actual congressional and state AV hearing records. There was also interrogation from the breakout attendees so as to stress-test typical AV policy messaging from all sides.
The session thus dissected and critiqued current AV policy messaging from industry, safety advocates, community members, and more. There was also discussion of how AV policy narratives are shift as deployments expand.
Key take-ways included:
· Educating lawmakers on AV issues needs to be a gradual process. One hearing is not enough to build the knowledge needed to pass effective AV legislation.
· Rather than generalized messaging, approaches tailored to the realities on the ground for a given state can help build support for AV legislation.
· Trusted messengers, including state agency staff, transportation industry experts from beyond the AV space, and advocacy groups are important.
A concluding point was that research on types of messaging targeted toward legislators and the resulting success or failure of AV-related legislation would be valuable.
Comment: The folks on the hot seat, i.e. AV deployers, take all these steps as a matter of course, coordinating with one another through the Autonomous Vehicles Industry Association. Others within the breakout likely got a front-row seat to what can seem a mysterious process, taking away a perspective they could bring back to their organizations.
Spies in the Machine? Autonomous Vehicles and National Security
As autonomous vehicles become increasingly connected, software-defined, and data-driven, national security considerations arise in transportation policy.
This session explored how governments are using legal, regulatory, and policy tools to address potential security risks associated with autonomous vehicle technologies, supply chains, data flows, and remote operations.
Panelists examined the intersection of autonomy, cybersecurity, geopolitics, trade controls, and critical infrastructure protection, including the U.S. Department of Commerce’s connected vehicle rule. The discussion addressed emerging risk vectors, the effectiveness of current regulatory approaches, and what the next generation of AV-related national security policy may look like.
FMVSS600: Designing Regulations Specific for Automated Driving Systems
Given that NHTSA is moving beyond voluntary frameworks and beginning to embark on formal rulemakings related to Automated Driving Systems (ADS), the industry faces a pivotal question: How do we move from “translating” old rules to “codifying” new ones?
This session explored the creation of a hypothetical NHTSA Federal Motor Vehicle Safety Standard (FMVSS) 600 Series, a dedicated suite of standards focusing on ADS vehicles, as well as the Code of Federal Regulations.
Speakers included experts from Aurora, TRC, VTTI, and Zoox, as well as NHTSA and ex-USDOT officials.
Quite a big lift for a breakout session! After the talks, the crowd bravely embarked on a hands-on exercise to develop hypothetical FMVSS for ADS across three ADS capability scenarios: Perception, Minimal Risk Condition, and Emergency Vehicle Detection.
Key points arising were:
ADS regulation is uniquely hard because the Federal government (via FMVSS) historically governed the vehicle, while states governed the human driving task. Now ADS regulations blur that delineation.
Tension over whether to regulate the “body” (vehicle hardware) separately from the “brain” (ADS decision-making).
Self-certification vs. third-party independent testing.
· Unlike static vehicle standards, ADS capabilities evolve over time, complicating a point-in-time compliance model.
The group emphasized that research is the foundation upon which FMVSS and other regulations can be built. Without sufficient research and data, the task of developing FMVSS becomes ever more difficult. Also, that not all regulatory actions need to take the form of an FMVSS; NHTSA has other authorities it can taken advantage of to address ADS safety.
All Politics is Local: Potlucks, Potholes, and Priorities in Cities
This session explored how the politics of AVs are shaped by each region’s political landscape, economic priorities, and geographic nuances. It featured interactive elements including mock press conferences and small-group discussions.
In Part 1, a panel of local policymakers and stakeholders shared perspectives from different stages of AV operations. In Part 2, attendees had the opportunity to observe a mock press conference, then break into small groups for “potluck conversations” about how neighbors back home might perceive AVs in their communities. In Part 3, experts shared their efforts to measure and incorporate public sentiment. A final Part 4 opened with another mock press conference before engaging attendees in a discussion about how to bring the findings home.
Participants included speakers from the Los Angeles Department of Transportation, the California Public Utilities Commission, the UC Davis MoSAIC within the Institute of Transportation Studies, Reed Mobility, Partners for Automated Vehicle Education,
The Innovation Majority Institute, RAND Corporation, Urban Innovate, and Waymo.
Panelists from a large U.S. city DOT, a state DOT automated transit program, and the
U.K./London market described what local leaders count as a successful AV rollout, as well as where early optimism is turning to friction. They also discussed how public sentiment is measured, how to weigh competing voices, and how AV companies can partner with cities to advance shared objectives.
Attendees discussed how their neighbors would (or do) perceive AVs operating in
their neighborhood and how they’d think about engagement and education in their
own community. Intriguingly, at one point, “neighborhood” tables were approached by an AV company that wants to operate in their fictional city, with ensuing back and forth as to issues at stake such as safety, multimodal integration, public education, workforce transitions, local investment, and the built environment. Each table then shared results with the full group.
It was clear that transparency and standardization are cities’ top asks – with a preference for standardized first responder and incident protocols across operators and interoperable data feeds rather than company-by-company playbooks.
AV operators noted that they have a wealth of opportunities to support local objectives such as upskilling the workforce and sharing actionable data regarding potholes, missing signage, and other infrastructure challenges.
Nevertheless, regulators stressed that they need defensible frameworks for how sentiment, narrative, and organizational readiness are weighted against ridership, disengagement, and crash data in decisions that must survive appeal.
INFRASTRUCTURE
Digital Road Rules
The SAE ITC Digital Road Rules Consortium (DRRC) is a collaborative industry group focused on establishing a comprehensive end-to-end process for interpreting and translating rules of the road into engineering-ready requirements for driving automation systems. Through cross-industry collaboration, the DRRC aims to enable safer, more predictable, and scalable deployment of vehicle automation technologies. The DRRC, working with Audi, organized this session.
The DRRC presented its recently published white paper, “Framing Rules of the Road Compliance for Driving Automation Systems from an Engineering Standpoint,” followed by a panel discussion where the proposed solution was explored in further detail.
The panel included a range of stakeholders, from regulators to enforcement agencies, industry, and academia. In particular, the Michigan Department of Transportation,
the California DMV, and Waymo participated.
TRANSIT
Shoot for the STARs: Bringing Automated Driving to Transit Service
With driving automation technologies having the potential to transform the transit industry, during this session participants had the opportunity to hear from deployers and engage with researchers focused on transit bus automation.
As a key part of this breakout, Federal Transit Administration (FTA) staff shared progress on current transit bus automation research activities guided by their current five-year research roadmap, Strategic Transit Automation Research (STAR) Plan 2.0, presenting previews of current research findings.
In reviewing current transit bus automation research and the overall state of the industry, it was clear that while it’s time to move beyond pilots and demos, it is by definition a slow process due to the heavy government role in U.S. transit systems. Expert panelists noted that development and commercialization is still at an early stage and will take time. In addition, funding is difficult and will require federal research support. Nevertheless, launching transit services takes a long time.
Panelists highlighted the importance of progressing past single-vehicle proof-of-concept pilots/demonstrations and noted that transit bus automation is transitioning to longer-duration services with more vehicles and that are designed to address real service gaps.
Key points were:
· Implementing driving automation systems in transit buses will be a substantial investment, and funding it may be a challenge, especially with limited revenue streams and resource constraints. Federal investment in transit bus automation research is critical to creating outcomes that meet public policy goals.
· Several transit automation projects have completed evaluations, providing insights into current technology maturity and perceptions of these technologies by both the public and agency staff.
Formal channels are needed so agencies don’t repeat the same pilot lessons in isolation.
· Evaluations shouldn’t be static post-mortems; they must be living documents applied before, during, and after deployments.
Automation is reimagining, not replacing jobs. There is a need for workforce transition planning and training.
TRUCKING
Connecting the Country: Preparing for Interstate AV Trucking and Freight
The trucking breakout has been a mainstay of ATS and its predecessors for well over a decade.
Public sector participants included California (DMV), Texas (DOT and DMV), while at the Federal level, USDOT’s Federal Motor Carrier Safety Administration
U.S. House of Representatives Transportation and Infrastructure Committee.
AV developers and deployers were represented by Aurora, Bot Auto, Einride, Gatik, Kodiak, PlusAI, Stack, Torc, and Waabi, while the Texas Transportation Institute and the Virginia Tech Transportation Institute held down the research side. The Commercial Vehicle Safety Alliance was also represented, as well as expert consultants.
A panel titled “The Here and Now (and Later) of AV Freight” discussed the realities of AV trucking – both present state and near future deployment of automated trucking and freight operations – in both urban and rural areas on state highways, city streets, and across multiple jurisdictions. Topics included driving forces (and barriers) for deployment and scaling, best use cases for a range of locations, current economics of AV trucking for immediate customer needs and opportunities, potential effects use cases may have on community economic development and socio-economic problems, and how current trucking practices might affect AV deployment.
A following panel focused on “Pathways for Getting to Scale,” discussed what is needed to address challenges identified in the first panel and how the automated freight future will unfold, particularly regarding regulatory developments for AV at the state and Federal levels. The latter was framed as “you are free to move about the country.” Clever!
Panelists also discussed the opportunity presented by mistakes as robotaxi’s have ramped up, with an admonition to “not rush” the scaling process.
And yet the trucking space must go its own way, as current federal CMV regulations (like enhanced inspections) will not easily scale to AV trucking.
There was emphasis on educating the public, current Commercial Drivers Licenses holders, inspectors, developers, regulators, and state/federal officials.
AV DEPLOYMENT MODELS
International Review of AV Deployments for People & Goods: Lessons from Europe, the U.S., and Asia
This big picture breakout session was designed by Urban Innovate, MAPtm, and ERTICO-ITS Europe. Speakers and facilitators came from the public sector (JRC / European Commission, Minnesota Department of Transportation), the private sector (Cityfi, Einride, fka GmbH), industry associations (International Association of Transportation Regulators), and academia (University of Tama Japan, University of San Francisco, University of Wisconsin-Madison).
This workshop focused on real-world automated vehicle deployments for passengers across three continents: Europe, the United States, and Asia. It was designed for practitioners who need to move beyond high-level narratives and extract actionable learnings that can inform policy, procurement, partnerships, and operational readiness.
Short, structured continental reviews covering these regions were offered with concrete examples of deployment contexts, operating models, and outcomes. Next, interactive workshops at tables allowed many voices to be heard. The focus was to extract what can be transferred across borders, and what requires adaptation. The intent was to create a “transferability toolkit” (framework + key insights) which participants could bring back to their organizations.
This was followed by a focus on goods movement on roads across the various continents.
The group found that autonomous mobility is transitioning from isolated pilots towards
early commercial deployment, even while each region follows a distinct pathway shaped by its regulatory framework, market priorities, and societal needs.
Passenger and freight automation are evolving differently, with robotaxis and public transport progressing alongside rapidly expanding autonomous logistics applications in ports, highways, and last-mile delivery.
Regarding freight, it was interesting to hear that 2000 trucks are operating in China with no drivers. This may be the largest deployment for trucking worldwide, however it wasn’t clear how many of these trucks are operating on public roads versus ports and logistics yards.
A key point was that international collaboration provides complementary strengths. Europe emphasizes harmonized regulation and cross-border interoperability, the U.S. goes for commercial service deployment, while Asia leads in large-scale logistics implementation
and integrated technology ecosystems.
It was noted that successful deployment depends on more than vehicle technology.
Regulatory readiness, operational integration, sustainable business models, digital infrastructure, and public acceptance are equally critical. This includes cross-border interoperability, workforce implications, safety governance, and community acceptance.
Going forward, the breakout participants saw the need to strengthen international knowledge exchange to develop common approaches for safety assurance, regulatory interoperability, performance evaluation, and operational best practices across regions.
In the research domain, there was the need to expand beyond technical performance to address public acceptance, human behavior, governance, workforce transition, and the integration of automated services into existing transportation systems.
A priority was to focus on longitudinal evaluation of commercial deployments to identify scalable business models, operational performance, societal impacts, and transferable lessons that can accelerate responsible deployment worldwide.
OPERATIONS AND APPLICATIONS
The Next Level for Emergency Responders: From Standards to Streets
With SAE Recommended Practice J0911 nearing formal adoption, state legislation continues to place importance on emergency responder readiness in enabling commercial operations, this session built directly on outcomes from emergency responder-focused sessions at ARTS 2024 and ATS 2025.
Participants engaged in an interactive session that focused on two critical topics related to emergency responder interactions with ADS-equipped vehicles:
· communications and scene management
· training and preparedness at scale
ADS developer training content and modalities were presented by Aurora, Waymo, and Zoox, thus covering both the robotaxi and truck space.
Participants collaborated on refining the research needs that remain considering the progress made in relevant standards. There was also a focus on building consensus on the topics and formats that a common training protocol could include to cover multiple vehicle types and operational models within the constraints of first responder schedules and budgets.
Attendees participated in interactive demonstrations of the types of training modalities and topics offered by ADS developers in both the passenger and freight sectors.
The aim was to refine research needs on communication methods given recent standards development, as well as engaging in interactive demonstration of training protocols and contribute to consensus on a common framework.
Currently, trainings are still specific to a single ADS developer. A key priority is to standardize training across ADS developers/platforms.
Communications begin with geofencing/avoid-the-area messaging, but not all interactions can be avoided, so direct communication at a scene – electronic, verbal, and non-verbal – is also an important piece of the puzzle.
It was noted that first responders will figure out workarounds if not given workable solutions, such as whacking the car hood to faster remote support!
The group emphasized that AV players need to be going to emergency responder organizations and conferences, rather than expect them to be able to attend our events.
Ready for AVs? A Collaborative Workshop on AV Deployment Maturity
This workshop combined brief framing presentations with facilitated, small-group working sessions to collaboratively validate and refine a draft AV Deployment Readiness Catalog developed by the UCLA Mobility Center of Excellence.
The activity catalog aims to help state and local agencies assess their readiness for AV deployment based on current deployment experience, serving as a planning reference and conversation starter for smoother coordination between agencies and AV operators. The creators of the tool notes that it enables self-assessment of relative maturity of their policies and practices to help identify implementation priorities across deployment stages (pre-deployment to fully deployed), drawn from cities with deployment experience.
Breakout participants focused in to contribute to the refinement and validation of the catalog, which is designed to facilitate high-level collaboration between industry stakeholders, public agencies, and the broader community. It covers readiness dimensions of governance, public safety operations, physical and digital infrastructure, curb management, accessibility, and workforce capacity.
The panel and workshop focused on seven dimensions: authority and governance, operator intake, permitting and deployment management, public safety, physical infrastructure, digital infrastructure and data exchange, performance measurement and
transparency, and funding and workforce capacity.
Panelists included speakers from Arizona DOT, City of Austin, CMU, TRC, Uber, and Zoox.
This group had a quite the set of take-aways and action prompts!
· Immediate priorities: Everything. But governance, public safety, performance metrics were highlighted. Physical/digital infrastructure are a “nice to have.”
· Among public-private stakeholders, we don’t have a common definition of safety – inside the car, outside the car, the impact on the transportation environment?
· Many items requested from operators do not bring immediate benefit to operators but are important to build trust – just make it formal and standardized.
· Define a common method to assess deployment readiness levels (akin to technology readiness levels).
· Find ways for agencies to sustainably support managing, tracking, and interacting with AV operators.
· Safety and trust building as a continuous process that inevitably rests upon human relations among public-private stakeholders is needed. First responder training and feedback plays a key role for the continuous improvement of fleet(s) behavior.
· The public may not care about the details – but expects vetted (by authorities at state/local level or third party), consistent fleet operations and vehicle behavior in safety-related events across operators.
· Independent assessors provide testing grounds and analysis that can help the entire industry figure out what needs to be measured to make safe decisions.
· Need for a centralized information-sharing platform that could communicate road closures, geofenced areas, operational restrictions, and other deployment-related information.
· However… we don’t have infinite public resources. We need to make communication more efficient (emergencies, special events) and find ways for agencies to sustainably support all efforts dedicated to managing, tracking, and interacting with AV operators. “Cities can’t do this alone.”
Weather & AVs: Who Needs What?
In this breakout, organizers and panelists included Audi, Mcity, National Weather Service, NSF National Center for Atmospheric Research, University of Michigan, Zoox, as well as consultants and engineering firms.
The aim here was to better understand road weather information needs by AV developers, at the same time examining roles of the public, private, and academic weather and transportation communities to see how best to meet these needs. Additionally, to evaluate the functional differences between Infrastructure Owner Operator (IOO) data aggregation and weather provider views to identify how these data sources can be used to complement one another for AV road-weather information.
Looking back at the ATS 2025 conference, panelists at a session entitled, “Weather & AV: What Can We Do About It?” had a very productive conversation about the challenges and opportunities surrounding AV operations under a range of adverse weather conditions. These primary conclusions were drawn from that session:
· While AV Operational Design Domains (ODDs) continue to improve and are able to operate under an ever-expanding range of weather events, there was consensus that an “ODD only” solution isn’t feasible and that AV developers must take a multi-pronged approach to operate under all weather scenarios, including vehicle-based sensing abilities, getting (preferably standardized) information on road weather conditions from the public sector, and getting weather products from both the NWS and private sector weather providers.
· The AV community needs to work more closely with the weather and public transportation communities to articulate their information needs, and do so in a coordinated, not siloed, manner. However, the communication platform for these discussions doesn’t formally exist.
· The opportunities to collect and share weather observations from AVs is huge and needs to be better explored. Such observations would feed into road weather models, 511 platforms, and products that would cycle back to the AVs for better performance.
This session continued the conversation surrounding these topics, focusing on two distinct topics:
· specific data levels required by AV developers
· the roles of public/private providers and IOOs in meeting those needs.
The intent this year was to dive deeper into the ways in which AVs’ needs can be met. What weather and road condition information do AVs need to be able to navigate these adverse situations? What are the best and most reliable sources of this information? How do we take advantage of efficiencies offered by the public sector to provide this information, including the National Weather Service and State Departments of Transportation? Which roles are better filled by private sector weather providers?
Here’s a sampling of the findings and lessons learned:
· Street to street conditions vary widely (snow plowing progress, localized flooding, black ice formation, etc.) Connected cars can continuously update this during standard deployment and can help inform routing decisions for both AV fleets and private road users. Weather information must provide macroscopic through microscopic levels.
· AI Simulation may oversimplify adverse weather conditions/driving, but there is a gap that needs to be considered to ensure the tests represent real world conditions (wind vector oversimplification, effects of cold on hardware, dynamic road conditions, different terminology for snow, etc.).
· How do car-level weather observations get communicated to other connected AVs or disconnected personal vehicles? Formats and communication methods need to be standardized. AVs and individual drivers want weather data in order to know where to avoid driving; cities/states want that same data so they know where they need to deploy to mitigate hazards created by the adverse weather. Weather data is currently available from existing agencies and programs, but it is essential to collaborate with the AV industry to properly leverage those findings.
· Humans naturally identify mitigations that machine learning may not consider e.g. driving between lanes during snow to avoid the snowbanks, or driving in lead car tire tracks, may violate AV driving training. How can AI models better address identifying and responding to hazards in this way?
· AVs have a social responsibility to ensure that traffic keeps moving in their respective markets, and consumers are open to sharing weather-related data with proper safeguards. As the community begins to rely on AVs, AVs must uphold their end of the bargain by providing weather-resilient service.
· Existing programs provide standardized weather data which can be consumed by AVs, but the decisions made from that data vary by the product (personal L3, L4 trucking, L5 robotaxi) and each will have unique thresholds before mitigating actions need to be taken. To that end, AV responses to weather need to be calibrated to the specific weather conditions - stopping may not always be safest. Similarly, weather data and road conditions alone are not sufficient - impact information is needed so that operational decisions can be made from it
· Continue developing partnerships across weather data producers (NOAA, private vehicles, AVs, etc.) and weather data consumers (AV companies, DOTs, etc.) to promote data sharing and to gain a better understanding of how roadway features are impacted by weather
PERSONAL CAR AUTONOMY
Communicating Safety: the L2 (plus) and L3 Puzzle
ATS is about more than just Level 4 fully driverless operations. This session explored how consumers understand—and how stakeholders communicate—technologies with similar names but different capabilities, with a focus on L2 and L3 automated driving.
Three panels addressed consumer communication best practices (including dealer and OEM education); risk and liability implications of naming a system “L2++” vs. “L3” for insurance, regulation, design, and human understanding; and the regulatory and market future of L3 for personal ownership and rideshare.
Via interactive polling and exercises attendees were asked to apply session learnings to craft a campaign or a consumer-facing message, making participants active contributors rather than passive listeners.
The first panel was titled “What Comes Next for L2+ and L3 Technologies?” With Level 2+ and Level 3 automation typically viewed as the future of personally owned mobility, is there more to it than that? This panel explored how OEMs and suppliers are evaluating investments in L2+ and L3 technologies, the regulatory frameworks needed at the federal and state levels to support deployment, and what consumers need from industry and government to build trust in these systems. Panelists discussed the current market and how lessons learned from existing state approaches to regulating ADS technologies can inform future policy. The prospects of L3 automation supporting rideshare and other mobility services was discussed.
Panel participants came from Audi, Alliance for Automotive Innovation, and Rivian, along with expert consultants.
Panel Two addressed risk. For instance, does the naming of a vehicle technology matter for insurance, regulation, design, and human understanding? How are OEMs, insurers and regulators balancing these questions? What names have meaning to consumers or affect driver behavior? Does the name of a system (L2++ vs. L3) lead to a fundamental shift in liability from the consumer to OEM? This panel included representatives from the fields of insurance (HUB International), industry (Audi, Mazda), government (NHTSA), and human factors (Exponent).
The third panel focused on educating consumers and policymakers on L2+ and L3 systems, addressing current educational practices among OEMs and how consumers are responding. What are best practices for educating consumers about the capabilities of their L2 or L3 systems? Is there a role for dealers at the point of sale, drivers’ education programs, or online training offered by OEMs? How can education be scaled across makes and models to meet the needs of consumers (and even first responders)? Speakers here included the Governors Highway Safety Association, the California State Highway Safety Office, PAVE, and UC Davis.
A tabletop exercise focused on creating an education campaign for L3 or for a specific element of L2.
Take-Aways
The focus needs to be more on situations, scenarios, and human responsibility vs. system responsibility. Responsibility may translate to liability; one big question is how this works during takeover. Naming conventions such as SAE J3265 are important.
· Understand risk implications of naming vehicle technology in legal/regulatory, insurance and marketing contexts.
· Contribute to development of consumer education practices for safe operation of L2 & L3.
· L3 only works within it’s ODD. The system “invites” the user to go into L3 mode, and only if still paying enough attention for human to be open to takeover when requested. While consumers need to understand risk and responsibility, but “you can’t teach SAE levels,” car dealers don’t use this for selling to consumers, and consumers don’t understand AV levels.
· Lack of clarity affects insurability.
· Confusion creates gap in trust.
· Overpromising may hurt AV industry as a whole.
SAE Level 3 and 4 Features: Coming Soon to a Vehicle Owned by You
SAE Levels 0-3 require human role, whereas SAE Levels 4 and up do not. This breakout explored the evolution and overlaps of these spaces. The organizers noted that vehicles equipped with Automated Driving Systems continue to be deployed on public roads, and the use cases are proliferating as well. The personally owned AV is becoming reality, with Mercedes having deployed its Drive Pilot system in the US, Honda selling its SENSING Elite system in Japan, BMW having deployed its Personal Pilot L3 in Germany, Ford planning to introduce a Level 3 system and GM developing its next-generation, ADS Super Cruise system, both scheduled for release in the US in 2028.
However, they asserted that still active areas of research and unique aspects of personally owned AV technology that have yet to be addressed in larger AV community discussions, including:
· Are current and planned personally owned AVs SAE J3016 Level 3 or actually approaching (if not fully achieving) Level 4? Does SAE J3016 need to be updated to reflect the industry approach?
· How to measure and determine the “re-engageability” or “driving fitness” of the DDT-capable user that is engaged in a secondary task to re-engage in the dynamic driving task (DDT)?
· Assuming a DDT-capable user that can re-engage, how to structure a safe handover of the DDT from the ADS to the user including (a) determining how much transition time can be allotted depending on the scenario and re-engageability of the user, (b) what feedback/instructions should be given to the user, perhaps depending on whether the handover is planned or unplanned, and (c) whether to communicate the ADS’ level of confidence in its decision making?
· How should the engagement of an ADS be conveyed to surrounding road users, including law enforcement and other first responders such as the turquoise ADS marker lamps recommended in SAE J3134?
As the organizers put it, this session aimed to resurrect a dormant discussion on Level 3 systems and challenges now that personally owned AVs are coming to market. For 2026, what’s new is that ADS features with DDT-capable users are no longer theoretical but are now a deployed and growing technology that may have a significant impact on the automotive sector and personal mobility.
The session convened researchers, developers, human factors experts, standards body leaders, and regulators to provide different perspectives on these outstanding questions, explore different ways of answering them, and consider how they will shape future deployments.
The aim was to create greater awareness of the testing and deployment plans for AVs with DDT-capable users along with any lessons learned from AV developers focusing on “What’s Changed” from as recent as five years ago when L3 was considered “too difficult”. Additionally, to address the outstanding and unique challenges for ADS features with DDT-capable users as well as the potential mitigation strategies and techniques. Furthermore, where the research, standards, and regulations should go next to advance the technology of ADS features with DDT-capable users and help ensure safe deployment and customer satisfaction.
Panel 1 had speakers from Audi, Exponent, GM, and VTTI providing viewpoints, which was followed by a workshop discussion DDT-capable user HMI challenges and possible mitigation strategies based on the panel discussion.
Panel 2 addressed personally owned AV legal, standards, regulatory, and education challenges and mitigation strategies. Panel members came from organizations such as CA DMV, Dykema, PAVE, Touchstone Evaluations, Transport Canada, and Varnum LLP. This was followed by a workshop discussing the legal, standards, regulatory, and education challenges and mitigation strategies based on the panel discussion.
The breakout wrapped up by noting that many practicalities remain to be addressed regarding user understanding and expectations. With a new CA DMV regulation requiring an end-user education program for L3 personal cars, there will be much to talk about next year!
SAFETY
So Be Good for Goodness’s Sake: War-Gaming Trade-Offs in AV Behavior
Organizers, experts, and role players brought their game to this war-gaming event.
Organizations represented included ATSIP, Bot Auto, District of Colubmia Department of Transportation, IIHS, Maryland Transportation Authority, Reynolds-Moore, TTI, TU Hamburg, TUV SUD, UC Davis, Waymo
Part 1 was titled “Driving as your conscious dictates OR To thine own self be true,” focusing on the value-based engineering process. In the first part of the game, players were assigned to a fictional company and city, but given a real philosophical standpoint from which to work. Will you be a utilitarian, a strict deontologist, or perhaps an Aristotelian? Using a physical board game, dice, live cell phones, and animated scenarios, participants decided what to do in each scenario, according to their philosophy. Then participant’s choices and actions were judged by a key external stakeholder, allowing the participant to revise their view. This concluded with companies describing their choices and external stakeholders crowning which philosophy more aligns with their views. Sounds fun.
Part 2 was titled “To speed or not to speed, that is the question.” Starting with a talk from an automotive ethicist on the ethics of speeding, participants got a chance to live their values as members of each fictional company, coming together to translate their moral concepts into specific engineering requirements for their AV fleet. As with Part 1, each group will had a chance to discuss their choices with key external stakeholders. Finally, all voted on which engineering requirements would be used to regulate speeding for AVs in fictional cities.
The tension between traffic flow and road rules and vulnerable road users was at play.
The ultimate goal of the session was to show how these different driving values are translated into concrete, measurable “engineering requirements” that developers use to program the vehicle’s computer. To conclude the game, developers from competing companies worked together in mixed working groups to debate and draft unified engineering requirements around speeding, practicing the difficult art of programmatic compromise.
Empathy-driven AV behaviors received the highest stakeholder approval. During the simulation, the fictious company programmed with an empathetic, care-ethics framework (Kinship Systems) secured 1st place in stakeholder scoring, receiving high marks for focusing on vulnerable road users.
VirtueOS, based on the belief that good driving is about developing excellent character traits—such as patience, courtesy, and self-control—rather than strictly calculating outcomes or blindly following rules, was second place. In third place, StoicSoft, was guided by the principle that you cannot control external events (like traffic jams, erratic drivers, or arrival times), but you have complete control over how you react to them.
The highest-rated engineering requirement, drawing from multiple moral standpoints, was:
“The vehicle shall adjust speed strictly based on the factual road conditions and without exceeding 10% over the posted speed limit OR the 85th percentile of traffic flow, whichever is lower AND shall never exceed the maximum design speed of the vehicle which is determined by the mechanical limitations of the vehicle, including the ODD and the
company-tested rider comfort. The exception is emergency vehicles, work zones, when VRUs are present in or within 1ft of a travel lane.” Whew!
In taking this approach, this requirement involved consideration of VRUs (care ethics), passenger comfort and speed relative to prevailing conditions (virtue ethics with a tinge rational egosim), road rules (Kantian ethics), and the functional limitations of the vehicle.
All proposed requirements touched on the tension between traffic flow and road rules, and on VRU’s.
The organizers noted that this session did not intend to get developers and other stakeholders to choose a philosophy. Instead, it aimed to get entities to identify the philosophy they already have, examine it, think about if it really works for them, make changes, and then move forward with intention, using it to guide their actions.
Their bottom line: understanding a given locality’s cultural preferences around driving behavior (which have roots in ethics), and considering how that aligns with or conflicts with ADS behavior (which also has roots in ethics) is an essential step on the pathway to good driving behavior.
The Forensic Crash Lab: Deconstructing the “Swiss Cheese” Failure?
This immersive session transformed the traditional policy panel into a high-stakes Forensic Crash Reconstruction Lab by engaging in a root cause analysis. Moving beyond abstract theory, the organizers challenged the audience to peel back the technical, environmental, and behavioral layers of a complex Level 4 autonomous vehicle collision.
The session used James Reason’s Swiss Cheese Model to conduct a live root cause analysis, demonstrating how catastrophic failures occur not from a single error, but from the perfect alignment of “holes” across the entire safety ecosystem.
Participants navigated through a rolling series of investigative reveals to understand how safety responsibilities are distributed across a Safe System Approach.
Organizers and expert came from entities such as Aurora, AVIA, IIHS, NHTSA, and Uber.
Key takeaways were:
· Comprehensive Data Analysis is needed to uncover root causes
· Safe System Approach required to close critical safety gaps
· Bias Toward Action: support deployment with regulatory enforcement oversight capable of taking immediate action against imminent safety risks
Going forward, the group concluded that:
· Multidisciplinary crash investigations should leverage cross-functional expertise
· Policy lags revealed by crash deconstructions should be addressed
· Risks should be mitigated in short-term work zones with heightened operational caution and updated safety protocols.
· Focus on E-bike speed enhancements/ modifications, user/rider licensing requirements, AV data sharing, and training machine learning models on micromobility behaviors.
Stay in Your Lane – The Art and Science of ODD Enforcement
As Automated Vehicle (AV) fleets scale, the rigorous enforcement of the operational design domain (ODD) becomes a primary determinant of safety and regulatory compliance. The organizers of this breakout noted that the core question with an automated driving system (ADS) is no longer “Can it drive?” but rather “Can we prove exactly where, when, and under what conditions it is safe to drive, and can we enforce those boundaries on our operations in real-time?”
Furthermore, emerging regulations (UNECE, CA DMV, CPUC etc.) put special emphasis on ensuring that the ADS recognizes and prevents exceedance of the ODD boundaries, executing fallback responses upon an ODD exit. While this session addressed theoretical frameworks such as SAE J3259 (forthcoming) and ISO 34503, it primarily focused on the systematic enforcement architectures essential for proving an “absence of unreasonable risk”.
The workshop moved past static ODD definitions to explore systematic enforcement architectures essential for proving an “absence of unreasonable risk.”
Participants investigated a “Scope Control” framework designed to ensure vehicles recognize ODD boundaries, prevent excursions, and execute fallback responses upon an ODD exit.
The session featured expert insights (from Aurora, Denso, Foretelix, Waymo, and others), combined with practical, hands-on activities to reinforce the core concepts.
The learning objectives focused on the following.
In “Operationalize the Taxonomy,” the aim for attendees was to understand what an ODD is how it is defined, to apply the hierarchical taxonomy of ODD standards to decompose high-level marketing claims into engineering-grade, monitorable metrics.
In “Architect Runtime Monitoring,” participants designed conceptual architectures for In-Service Monitoring & Reporting (ISMR), focusing on the detection of dynamic ODD exits (i.e., situations where the vehicle encounters conditions it wasn’t designed to handle) and the triggering of Mitigated Risk Maneuvers (MRM).
In “Establish a Systematic Enforcement Framework,” participants built a structured roadmap for moving from theoretical ODD definitions to an enforceable architecture that prevents operating outside the ODD boundaries (e.g., weather conditions, road types etc.).
And in “Bridge Policy and Technical Monitoring,” participants developed a shared vocabulary for communicating technical weather and infrastructure thresholds to regulators and auditors.
Key take-aways were:
• Real-time ODD is a moving target: operational gaps in real-time confidence measurement, onboard vs. offboard sensing capabilities, and automated decision-making.
• Challenges:
§ ODD conditions exist on a spectrum, not binary. MRC strategies vary across operational domains –L3 & L4 have different requirements. There are sensing, inference, and dynamic environment challenges.
§ ODD boundaries are dynamic and continuous. Defaulting to fleet grounding or immediate stopping at minor environmental triggers (e.g., light fog/snow) is product-impractical; vehicles require graceful capability reduction to maintain continuous motion safely.
• Research Needed:
§ Dynamic ODD Sensing & Confidence Quantification, Adaptive & Context-Aware MRC Frameworks, and Standardized Operationalization Tooling & Attribute Deep-Dives
§ Adaptive & Context-Aware MRC Frameworks: Establish nuanced MRC/MRM policy frameworks that categorize triggers (environmental vs. spatial) and evaluate trade-offs between pulling over, slowing in-lane, or continuing degraded operations based on vehicle type and safety context.
• Operationalization Tooling & Attribute Deep-Dives:
§ Build standardized tooling and validation methods for operationalizing ODD enforcement, shifting focus from broad surface-level ODD coverage to deep-dive validation of critical ODD attributes.
§ Important to shift focus from broad surface-level ODD coverage to deep-dive validation of critical ODD attributes.
§ ODD as a Spectrum: ODD conditions (weather, visibility, road geometry) exist on a continuous spectrum rather than binary limits, presenting major product availability vs. safety trade-offs.
§ Research and develop robust onboard/offboard sensing fusion (e.g., ESC, remote assistance, local weather) to measure detection confidence for emerging environmental and road conditions in real time.
§ MRC/MRM Execution Complexity: MRC strategies vary heavily across operational domains (L3 vs. L4, Robotaxi vs. Trucking), ranging from immediate pullover to degraded continuous driving or driver takeover.
§ MRC policies are highly domain-dependent. L3 systems face driver takeover timing and non-response risks, while L4 Robotaxis/Trucks must balance safe pullovers against traffic disruption, road geometry (curves vs. straights), and unexpected route detours.
• Operationalization & Real-World Tooling:
§ Defining ODD limits on paper is easy, but enforcing limits in real time requires overcoming sensing, inference, and dynamic environment challenges.
§ Transitioning from static ODD definitions to live enforcement exposes significant operational gaps in real-time confidence measurement, onboard vs. offboard sensing capabilities, and automated decision-making.
Collision Avoidance Driving Behavior Reference Models (CADBRMs): An Emerging Safety Tool for Developers, Regulators, and Researchers
Collision Avoidance Driving Behavior Reference Models (CADBRMs) are a rapidly evolving field where human-like responses to critical scenarios are modeled. These models are increasingly influential in AV development, validation, testing, and homologation.
This breakout focused on understanding CADBRMs and how they fit into participants role as developer, regulator, researcher, etc. Topics included standardization efforts in SAE; understanding how CADBRMs are designed and built; how to leverage CADBRMs to develop, validate, and test AV stack behavior in critical scenarios; understand the current state of maturity of these models and what use cases they currently serve, including homologation; and understand how driving behavior reference models can be validated (i.e., what it means for them to be “good enough”).
Researchers shared the current maturity of CADBRMs, key methods for designing them, and likely future development directions. AV developers shared how these models currently help them develop, validate and test AVs. Regulators discussed how these models are used in the context of ADS regulation (presently, in UNECE R157 and, potentially in the future, L4 regulations).
Experts spoke to the state of the art in CADRBMs, scenario-based model evaluation, and ADS industry developed models, as well as how CADRBMs can be used in safety evaluation, virtual scenario testing, system development, and homologation. Speakers and panelists came from cogniBIT, European Commission, fka, Fortellix, JARI, Leeds University, NHTSA, and Waymo.
Key findings were:
· Different approaches to models are available and different CADRMs are available using these different approaches
· Different validation approaches are available
· Driver modelling is a growing field but requires more attention and research (driver models are included in regulation (see UN R157))
· Sharing of model libraries is necessary to find consensus
· Driver model is key assessment tool, yet NHTSA is not using driver models
The group detailed what’s needed going forward:
• Shared language across all stakeholders necessary – this is the goal of the SAE J3330 Recommended Practice currently under development)
• Validation of driver models is still incomplete
• More discussion and exchange on international level is needed
• Critical analyses of benefits and limitations are required
Three Years of AV Safety InfoShare: Where Are We Now?
This breakout was organized by Partners for Automated Vehicle Education (PAVE), with the intent to understand the AV Safety InfoShare and how this program works to unite AV stakeholders in tackling the foremost safety challenges.
In 2024, top AV companies launched the AV Safety InfoShare – modeled after the longtime aviation InfoShare model – to facilitate collaboration in tackling the biggest challenges in AV safety.
The group identifyed key insights, trends, and lessons that have emerged from four previous AV Safety InfoShare sessions and considered how the AV Safety InfoShare program can adapt to meet future safety, policy, and public-trust needs as AV deployment scales.
Panelists will assess how InfoShare has evolved alongside the technology and discuss how the program can continue to support transparency, trust and safety outcomes in the years ahead. Panelists came from organizations such as Edge Case Research, GM, Kodiak, MOIA, Pivot Point Strategies, SAE, TORC, and Zoox.
Testing What Actually Breaks: Evaluating Automated Vehicle Performance at the Boundaries of Design Assumptions
With automated driving systems transitioning from pilot deployments to scaled operations, the industry challenge is shifting from demonstrating capability to building confidence in performance under real-world variability. Simulation, closed-course testing, urban test environments, and operational deployments each provide valuable insight, but many critical system limitations only become visible when evidence from these environments is evaluated together.
This session examined how organizations across the automated driving ecosystem are using simulation, physical testing, and real-world operations to evaluate performance at the boundaries of operational design domains. Discussions focused on degraded conditions, edge-case behavior, deployment readiness, and how different validation environments reveal different types of system limitations.
Perspectives included proving grounds and urban test environments, public-sector deployment leaders, and ADS developers discussing how validation evidence informs deployment decisions and scaling strategies. Experts provided talks from organizations such as Aurora, Drive Ohio, City of Detroit, TRC, UMTRI/Mcity, and Waymo.
Key Findings were:
· Different testing environments reveal different aspects of automated vehicle performance; confidence is built by integrating evidence across simulation, controlled testing, and public-road operations.
· Data is the foundation for identifying performance limitations, translating controlled test results into real-world expectations, and making informed deployment decisions.
· Transparency and collaboration among developers, test facilities, researchers, and public agencies are essential to closing gaps between research, testing, and deployment.
Going forward, the group identified research needs:
· Establish shared data frameworks and greater transparency around testing methods, performance indicators, system limitations, and deployment-readiness decisions.
· Expand collaborative research among developers, test facilities, researchers, and public agencies to identify priority scenarios and evaluate performance under degraded, regionally specific, and ODD boundary conditions.
· Double-down on data-driven research into degraded conditions, complex interactions, adverse scenarios, and long-tail events that are difficult to evaluate through routine operations alone.
Evaluating Apples-to-Apples: Applying the RAVE Checklist to AV Safety Impact Studies
Measures of AV safety are vital, but is every assessment of equal quality? Can the results of one study be compared to another? It is difficult for stakeholders to make sense of the many studies and results being published. Organizers of this session noted that there is a lack of consensus on the standardized methods (although progress is being made), while at the same time the process of obtaining and analyzing the data sources (both human and AV) are unfamiliar.
In this session, participants were appointed as guest editors of leading scientific journals tasked with performing technical evaluations of statistical and epidemiological methods used in one of several AV safety impact studies. Participants used the RAVE Checklist, a best practice for conducting AV safety impact research and the basis for the forthcoming ISO T/S 25536, to evaluate several safety impact studies.
Discussion included how each study should be interpreted in the context of other studies, what limitations or uncertainties are present as a result of the approach chosen, and how potential next steps to address limitations.
The session concluded with a panel discussion about implementation barriers and paths forward to continuing to advance the state-of-the-art in AV safety impact studies from a number of diverse perspectives. Panelists came from organizations such as
General Motors, Insurance Institute for Highway Safety, NHTSA, San Francisco County Transportation Authority, Tesla, University of California Berkeley, Virginia Transportation Research Council, and Waymo.
Key Findings included:
· There is a need to continue safety impact research, given new ADS entrants, new ODDS, and more refined benchmarks.
· There is need for developing consensus on methodology. Even with existing best practices, stakeholders need to become educated.
· There is a need to break down silos. Data is sparse or nonexistent (especially on the human benchmark side).
Lost in Translation: Turning Real-World Incidents into Safety-Critical Scenarios
Accident reports are written for legal, statistical and administrative purposes — not for simulation — so key scenario fields (exact geometry, trajectories, vehicle mechanics) are often inferred or missing entirely, even in a complete report. This intriguing breakout focused on one core safety question: Why is it still so difficult to transform real-world incident reports into testable, safety-relevant AV scenarios?
Plus, how do regional reporting differences affect this process?
The session used real accident and incident reports from the United States, the EU, and Japan. By working with reports from three regions, participants compared reporting structures, data richness, terminology, and implicit safety assumptions, and assessed how these differences influence scenario generation and safety validation.
The participants considered human-in-the-loop AI applied to real safety data, emphasizing validation and trust. Additionally, integrated comparison of expert, AI, and crowdsourced scenario creation methods. All with the aim of framing what must change in data, practices, and standards to enable globally consistent, incident-based safety scenarios.
Experts came from Applus+ IDIADA and the Japan Automobile Research Institute, among others.
Takeaways included:
· Reporting is not standardized; each country records and structures accident data differently, making cross-region scenario reconstruction inconsistent.
· Training should shift toward education: explaining why a step matters lands better than simple mandating a new process. If reporting adds too much extra work, people won’t do it. Further research is needed on making accident reporting less difficult for the person filing it, while still capturing simulation-ready detail, balancing completeness against real-world adoption.
· Current investigation approaches remain too data intensive; participants called for a more human-factor-oriented focus in how accidents are investigated and reported.
· It is important to harmonize accident-report fields across regions (US, EU, Japan) so that scenario-relevant data (such as geometry, trajectories, vehicle mechanics) can be captured consistently and reduce the translation gap.
· US ADS standards were flagged as an area with room for improvement.
AI / VALIDATION
To Train, or to Validate — That Is the Question
The development of end-to-end (E2E) automated driving systems shifts the development process and resulting safety burden from manual rules to data-driven learning. This shift creates a critical dilemma between the need for training (optimizing the model to mimic expert driving in both nominal and critical scenarios) and the need for validation (proving the model can safely handle the broad range of conditions and critical scenarios on the road).
A basic assumption is the fact that specialists will need some differentiation between the training data and the validation data – they cannot be the same. Ultimately, effective validation for E2E systems requires a decoupled strategy: using real-world data to train for “normality” while employing high-fidelity simulations or targeted “edge case” datasets for validation to probe the limits of the system’s generalization.
This is a paradigm shift which raised key questions about how to produce safe AVs:
· How to perform training and ensure its sufficiency?
· How should V&V differ for AI-based stacks?
· What are the key steps to ensure the ultimate products are safe, robust and predictable. How is this different for AI-based stacks?
· What are the set of metrics to be used to track the process and ensure safety?
· How can simulation and real-world data be combined to most effectively train and validate AI-based stacks?
· Should regulators treat AI centric systems differently?
Plenty of challenges and questions, setting the stage for this intensive breakout session. Organizers and speakers came from organizations such as Ernest and Young, Foretellix, Inchor, Laplace Intelligence, Motional, Streetscope, Torc Robotics, Transport Canada, University of Michigan Transportation Research Institute and Mcity, and Wayve.
A key focus was on effective and efficient training and validation of safe AI-centric ADS /E2E systems with these factors in mind:
· ADS development and validation is shifting from codified to data driven
· ADS systems need to minimize the need for remote assistance
· Regulation is not distinguishing between types of ADS systems but probably should
Challenges and approaches for effective and efficient training (development) and validation of safe AI-centric ADS /E2E systems were noted as:
· Handling AI and E2E ADS system challenges pertaining to validation of bug fixes, changes, and other enhancements.
· The “traditional” AI challenges of hallucinations, unpredictability, lack of explainability.
· What should be the approach to demonstrate safety, given AI centric systems, training data and validation requirements?
Key Takeaways included:
· ADS development and validation is shifting from codified to data driven.
· Various methods evolve to handle data, curate relevant data, measure safety parameters and values.
· No robust, fool proof methods exist.
· The ADS systems themselves need to evolve and develop more predictive capabilities, in order to minimize the need for remote assistance.
· Coverage and exposure metrics are a must for safety case, and for quality assessment.
· Regulation is not distinguishing between types of ADS systems. It is technology neutral, but probably should not be so.
Open questions were:
· How many data sets are needed – for what purposes ?
· How do you approach validation failures ? How do you root cause, fix and re-validate given AI models?
· How do you find the long tail of scenarios and hazards?
· How do you know when you are done validating?
· Safety case for regulatory demand – any unique needs for AI?
The Evolving Landscape of Safe AI in Mobility: What’s Working, What’s Coming, and Where We Go from Here
With artificial intelligence now driving many of the most transformative advances in autonomous mobility, and presenting the most pressing safety challenges, this session explores how AI is being applied responsibly across perception, planning, and control, and how industry, academia, and standards bodies are working to ensure its safety.
A moderated panel discussion brought together safety and AI leaders from across the industry, with panelists will discussing the latest approaches to applying AI safely in perception, planning, and control systems, as well as methods for hazard analysis, assurance, and system-level integration.
The conversation highlighted the practical use of ISO/PAS 8800, alignment with UL 4600 and ISO 21448, and recent extensions to Machine Learning Failure Mode and Effects Analysis (ML FMEA) for assessing AI-related risks. The experts explored what’s working in current industrial practice, what new safety strategies are emerging, and where international standards are heading next.
Organizers and speakers came from groups such as Edge Case Research, General Motors LLC, Qualcomm, Reynolds & Moore, Virginia Tech Transportation Institute,
Waabi Innovation, and Zenseact.
Key takeaways were:
· There is no single “winning” AI architecture for autonomous driving. Panelists described multiple approaches, including modular architectures (AI and traditional software working together in specialized components) and end-to-end architectures (a single AI system learning driving behavior directly from sensor data rather than relying on many separately programmed decision-making modules). The discussion emphasized that different architectures offer different advantages in safety, explainability, and performance, and that hybrid approaches continue to emerge.
· Nature may offer valuable lessons for AI safety. The discussion highlighted the human brain as an effective model for autonomous driving architectures, balancing learned behaviors with specialized reasoning and safeguards. One memorable analogy described a spouse as an “excellent collision warning system,” illustrating how humans naturally combine perception, judgment, and external feedback to improve safety.
· Building public trust requires more than a technical safety case. Panelists agreed that demonstrating technical safety alone is insufficient to earn public confidence. Improving transparency, developing a shared vocabulary across engineers, policymakers, and the public, and communicating safety evidence in understandable ways were identified as essential steps for connecting rigorous safety engineering with societal trust.
Going forward, the group emphasized:
· Strengthen government-industry partnerships to advance AI safety. Continued collaboration between government, industry, and the research community is needed to accelerate the development of AI safety methodologies, evaluation frameworks, and standards that enable both innovation and public confidence.
· Advance AI safety as a shared industry responsibility. Public trust in autonomous mobility depends on the collective performance of the industry. Individual safety failures can undermine confidence in AI-enabled transportation as a whole, highlighting the need for continued investment in robust safety engineering, transparency, and responsible deployment.
· Sustain open, cross-disciplinary dialogue. Ongoing engagement among policymakers, industry leaders, standards organizations, and academia is essential to develop a shared vocabulary, align technical advances with societal expectations, and ensure AI safety evolves alongside rapidly advancing autonomous technologies.
REMOTE ASSISTANCE & REMOTE DRIVING
Remote Assistance: Facts vs. Myth
This session aimed to define Remote Assistance in terms of what it is and what it is not. This was accomplished in a panel of both ADS developers/operators and stakeholders who will provide real world examples/use cases of Remote Assistance (vehicle guidance, scene confirmation, First Responder interaction, etc.) as well as discussion of myths and misconceptions.
A clear distinction was also made between Remote Assistance and Remote Driving by highlighting the conditions and requirements by which they each occur.
The panel discussed Remote Assistance Operator selection and training and Remote Assistance in operations, i.e. how remote assistance events are monitored and used within fleet supervision frameworks to assess safety, performance, quality of service, and scalability. Also discussed was how remote assistance is currently perceived by the public, with a focus on identifying communication strategies to more clearly convey the role
of human oversight in ADS operations. Importantly, the group addressed how remote assistance is likely to evolve as the AV industry scales.
Organizers and speakers came from organizations such as Aurora, City of Austin, Loughborough University, Urban Innovate, Waymo, and Zoox.
Key take-aways were:
· The approach to remote assistance was similar for all companies on the panel. No remote assistant at any of the ADS companies represented had the ability to remotely drive their vehicle.
· Cities need to know what a Remote Assistant can or can’t do. In particular, first responders need to know if, in an emergency situation, they can move a vehicle.
· Riders need to know how to get information during a ride especially when an incident or event is in progress. There may be different responses needed depending on the specific rider’s needs.
Ctrl + Alt + Drive: Remote Driving
With remote driving emerging as a critical component to autonomous systems and other use cases, this panel brought together leading voices from industry, government, and academia to explore the rapidly evolving landscape of remote vehicle operation — a technology that enables human operators to control vehicles from distant locations using real-time data feeds, advanced telecommunications, and sophisticated control interfaces.
As the organizers noted, remote driving sits at a unique intersection of automation and human oversight. Whether serving as a fallback mechanism for autonomous vehicles encountering edge cases, enabling new mobility services, or expanding freight and logistics capabilities, remote driving presents both extraordinary opportunities and complex challenges. Panelists examined the current state of the technology, human factors research informing how remote operators interact with vehicles in dynamic environments, and the regulatory landscape surrounding remote driving.
Organizers and speakers came from organizations such as Exponent, Federal Motor Transport Authority (Germany), Kittelson & Associates, Varnum LLP, and Vay.
The Human Side of Automated Transportation: Inside, Outside, and Remote
The premise of this breakout session was as follows. To reach their full potential, driving automation systems require more than advanced environmental sensing; rather, they depend on a deeper understanding of the human beings sharing control, riding inside, and moving around these vehicles. As automated vehicles move towards scaled operations, safety outcomes increasingly depend on the communications and coordination between technology and a set of human roles: the human inside the vehicle, other road users outside the vehicle, and remote assistance personnel. These roles are often treated as separate domains with different requirements, yet they are linked by their operations in a shared location.
As their joint success and experiences depend on communication, shared understanding, and, at times, coordinated decision-making, this session introduced a perspective that treats these roles as a connected system, focusing on sensing and information exchange across the ecosystem. The intent was to synthesize real-world operational data and research insights into concrete challenges, such as mismatched expectations and communications, interventions, and responsibilities, with the intent to guide the development of actionable research questions, design principles, and policy priorities.
In essence, automation changes human roles rather than eliminating them.
Organizers and speakers came from organizations such as FHWA, IIHS, Kodiak, Loughborough University, MIT, Tesla, Transportation Research Center, and the University of California-Los Angeles.
The conversation focused on topics such as:
· how increasing automation is reshaping human roles across the transportation ecosystem, from drivers and occupants to other road users, remote support, and operations.
· Examining human-system interaction across driver behavior and cognition, occupant protection, infrastructure and mixed traffic, remote support, and real-world automated vehicle deployment.
· the importance of integrating human factors across technology development, infrastructure, operations, and policy as automated transportation moves toward broader deployment.
· Human involvement extends across drivers, occupants, other road users, remote support, infrastructure, and operations.
· Observable behavior is not the same as cognition. Behavioral indicators can help estimate attention, readiness, or intent, but understanding human state requires multiple signals and consideration of individual and situational variability.
· Expectations and responsibilities need to be clear. Mismatches between system capabilities and what people expect automation to do can create confusion, particularly around transitions of control and responsibility.
· Remote support introduces new challenges in situational awareness, uncertainty, workload, and task prioritization, particularly as automated operations scale.
· Human touchpoints are important sources of learning, revealing opportunities to improve technology, interfaces, operations, and support.
Key takeaways and action items were:
· Develop and validate approaches that combine behavioral and contextual indicators to better estimate human state while accounting for variability across people and situations.
· Better understand expectations, responsibility, trust, and transitions between humans and automated systems across different levels of automation.
· Expand research on occupant behavior and protection, including diverse occupants, seating positions, restraint use, and evolving vehicle interiors.
· Establish evidence-based approaches for remote support, including situational awareness, uncertainty communication, workload, task allocation, and workstation design as operations scale.
· Continue learning from real-world deployment and simulation, while also strengthening collaboration across disciplines to incorporate human capabilities and limitations into evolving safety approaches.
Executive Committee for ATS 2026
It is important to acknowledge the Executive Committee for ATS. Some of these folks have been doing this for well over a decade. Their contributions are highly important to the AV community.
· Jane Lappin (Blue Door)
· Edward Straub (Velouria)
· Steve Shladover (PATH)
· Valerie Shuman (SCG)
· Marisa Walker (AZ DOT)
· Steve Ridella (Zoox)
· Chan Lieu (Aurora)
· Laura Fraade-Blanar (Waymo)
· Chris Bartholomew (Denso)
· Jiaqi Ma (UCLA)
· Hyungjun Park (ITS-JPO, USDOT)
· Joshua Cregger (USDOT Volpe Center)
· Julie Evans (TTI)