Speaking at the United Nations AI for Good conference in Geneva earlier this month, Carnegie Mellon University Professor David Danks commented on self-driving vehicle testing on the streets of his Pittsburgh neighbourhood: “They’re boring. You don’t notice them anymore because you see them five to ten times every day. People started to walk more and more frequently directly in front of the Uber cars, disproportionately more than they would ever in front of a human-driven car.”
“I would stop somebody and say, hey, I noticed you jaywalked. Across the board, with unanimity, they would say, oh, yeah, I knew it was an Uber car and it would stop.”
Danks says that jaded attitude changed this March, when a self-driving Uber vehicle struck and killed a pedestrian in Tempe, Arizona. Uber immediately shut down its Arizona autonomous fleet and halted testing in other cities.
Although Uber is rumored to be planning a resumption of autonomous vehicle road testing this summer in Pittsburgh — where Uber ATG is headquartered — they are facing opposition from Mayor Bill Peduto. A former advocate of ride-sharing and self-driving car technologies, Peduto appears to be having second thoughts in the wake of the Tempe accident, and recently lashed out at Uber on Twitter:
Danks has been researching public trust in autonomous systems over the past few years, with an emphasis on self-driving cars. Earlier this year he co-authored the paper Regulating Autonomous Vehicles: A Policy Proposal, which proposes regulatory entities borrow experience from the US FDA, which regulates drugs and other medical equipment.
The paper calls for a new process called “Roadway Translation,” which Danks likens to the US Food and Drug Administration (FDA) approval process. But instead of drugs or MRI scanners, officials would be regulating Autonomous Vehicle Ensembles, comprising 1) Hardware components including sensor packages, processing platforms, and automotive hardware; 2) Software systems including visual recognition, planning, and control algorithm; 3) Models and constraints that specify the range of contexts the vehicle can be expected to identify.
Roadway translation would take three steps. In early-phase testing, the vehicle will “explore and simulate” in virtual environments to learn how to codify different road situations. The transitional testing that follows will enable developers to uncover additional, unanticipated contexts of operation. Confirmatory testing, which puts the vehicle on the road, will confirm whether the car can respond to a wide range of operational contexts. This last phase is “analogous to the prescription-and-monitoring stage of drug development”, to quote from Danks’ paper.
Danks’ proposal ensures that regulatory bodies could suspend tests or sales of self-driving cars until benchmarks for safety and reliability were demonstrated. They would also have control over all relevant data.
Danks has now teamed up with Professor Aimee van Wynsberghe from Netherlands’ TU Delft, Wee Shann from Singapore’s Land Transport Authority, and China’s Tencent Research Institute. The group will study how autonomous vehicles are regulated around the world, and run a series of workshops on self-driving policies.
“If you just look at the variability in terms of how pedestrians and vehicles interact with one another. It seems to us that we have enough variability both in terms of the regulatory dimension and in terms of the cultural dimension,” explained Danks in Geneva.
Whether Uber will resume testing in Pittsburgh this summer remains unclear. But one thing is certain: If Pittsburghers are reluctant to welcome self-driving cars on their streets, stakeholders will have to win public trust by taking measures to increase vehicle operational safety.
Journalist: Meghan Han | Editor: Michael Sarazen