This week, I’m at the 2018 edition of the Automated Vehicle Symposium in San Francisco, and will be providing daily recaps of the proceedings (as I’ve experienced them). The symposium runs from July 9-12, and while yesterday was technically the second day of the symposium, Day 1 featured only afternoon breakout sessions. Here’s a rundown of the programming I found notable:
The morning program started off with Kristin Kolodge of J.D. Power and Tina Georgieva of Miller Canfield, who shared findings from their joint March 2018 report on AV liability and consumer attitudes toward the technology. Some key points:
- Lawsuits are going to get a lot more expensive when involving AVs, due mostly to a prolonged “discovery” phase full of engineer depositions and expert witnesses.
- Passengers will be more willing sue operators of fully automated vehicles (L5) than for either non-automated (L0) or partially automated (L3), instances in which they are typically comfortable with the damages covered by liability insurance.
- The legal field is still poorly informed on situational complexities associated with partial automation, and has difficulty understanding the spectrum between no automation (L0) and full automation (L5).
- Nearly half of vehicle owners currently expect dealerships to educate them on proper use of ADAS (advanced driver assistance systems), though this currently falls significantly short of expectations. A majority of consumers would also pursue additional certifications allowing them to use ADAS features, if offered.
Lyft’s VP of AV Programs, Nadeem Sheikh’s presentation illustrated a daunting task for rideshare operators as they attempt a transition to autonomy. Bottom line: customers have greater demands for ride share convenience and efficiency than ever, and they will not handicap AVs (beyond an initial novelty) in terms of mobility expectations. Also, AV development costs aren’t coming down quickly enough to provide rides at a fraction of current prices (as promised by many).
This means AVs will have to provide service at the same level or better than existing options. Customers will not accept a poor ride experience, and will not tolerate longer wait times for dispatch. As Sheikh noted, if Lyft can’t demonstrate a wait time of less than 3-5 minutes, customers will hail a competitor’s service instead. Moreover, the economics of fleet ownership and management likely preclude service in areas that will result in significant travel time deadheading (no passengers in vehicle). As a result of these factors, Lyft is designing its AV routing algorithms to:
- Avoid streets with bicycle lanes
- Avoid complicated intersections
- Geofence vehicles to zones of operation that either have the population density or enough traffic to economically justify operations
This programming avoids uncontrolled environments with lots of pedestrian and bicyclist (and scooter) interactions, and situations where low visibility or challenging edge cases would complicate the AV driving task and reduce level of service. Also, Sheikh noted there will almost certainly be a driver in vehicle for the foreseeable future. He didn’t fully elaborate on their role, but noted AV technology would be introduced and expanded incrementally, in terms of features, speeds, and operational design domains that allow Lyft to expand its area of service.
Editorial note: Based on what I’ve seen/heard, Lyft’s AV capabilities aren’t all that advanced at this time, but Sheikh’s speech largely echoed my own sentiments about the challenges of implementing AVs in dense urban areas. Ultimately, cities may have to coax TNCs through either carrots or sticks to serve lower-density or unprofitable areas out of equity concerns.
US Department of Transportation Secretary Elaine Chao gave a brief keynote address to cap the morning session. Her prepared remarks outlined the aspirations and challenges USDOT has associated with AVs, and priorities for the agency as it prepares to release its “multimodal” Federal Automated Vehicle Policy 3.0. For more on this plan, see Eno’s explainer from March.
Other morning sessions included presentations on:
- The state of play in AV sensor architecture.
- AV safety standards and validation.
- The utility of deep neural networks.
- NTSB’s findings from Tesla’s fatal Autopilot crash in Florida (2016), and preliminary findings from Uber’s fatal automated vehicle crash in Arizona (2018).
Finally, I chose to attend an afternoon breakout session titled “Trucking Automation: Deployment Challenges and Opportunities,” which featured a lot of research and analysis from tests on highway truck platooning. The fuel cost savings potential is clearly there, but we still have a significant amount of research and testing to perform before we understand platooning well enough to see large commercial fleets make significant business decisions. Platooning-enabled following distance impacts air flow to the radiator (and as one questioner noted, tires), which could negatively impact fuel savings and asset depreciation. Ideal truck platoon chain length accounting for lack of dedicated AV facilities still isn’t definitive.
More to come from Day 3 (July 11), stay tuned…