Roughly three months after the pandemic halted its autonomous vehicle tests, Lyft today announced its safety operators will resume driving a portion of its cars on public roads. An employee-only autonomous ride-hailing pilot in Palo Alto remains on pause. But in a blog post, Lyft director of product Sameer Qureshi and director of engineering Robert Morgan characterized road testing as a “critical” part of Lyft’s driverless systems development.
In March, Lyft’s safety drivers — along with engineers and developers — were told to stay home until further notice as shelter-in-place orders made public road testing impossible. In the interim, the company has leaned on simulation to further refine its platform. Autonomous vehicle developers agree that simulation supplements but can’t replace real-world experience.
A spokesperson said Lyft would continue to abide by the U.S. Centers for Disease Control and Prevention guidelines and work with local governments in deciding whether to pause testing in the future. This week, governors in Washington, California, Florida, and Texas walked back some of their reopening plans as COVID-19 cases rose in more than 30 states across the U.S.
Currently, Lyft safety drivers are using personal protective equipment (including face shields) and taking precautionary steps inside the driverless vehicles. Two drivers will be paired together for two weeks at a time and subject to temperature checks, and separated by partitions installed inside the regularly sanitized cars.
While simulation may be insufficient on its own, Jonny Dyer, director of engineering at Lyft’s Level 5 self-driving division, told VentureBeat in an earlier interview that the company chose to “double down” on digital by leveraging data from the roughly 100,000 miles its real-world autonomous cars have driven and calibrating its simulation environment ahead of validation. Specifically, Lyft refined the techniques it used in simulation to direct agents (such as virtual pedestrians) to react realistically to vehicles, and it built out tools like a benchmarking framework that enables engineers to compare and improve the performance of behavior detectors.
Lyft didn’t focus on challenges like simulating camera, lidar, and radar sensor data, looking instead at traditional physics-based mechanisms, as well as methods that help identify the right sets of parameters to simulate. In addition, it reworked its validation strategy to more heavily address things like structural and dynamic simulation.
“Training inputs like weather and pedestrian behavior are limited to what’s happening in the world at each moment, and it can be unpredictable when you encounter a rare obstacle a second time. If reliant upon on road miles, it may take some number of billions of miles to test everything,” Qureshi and Morgan wrote. “Therefore, we supplement our on-road testing with simulation, which gives us a cost-effective way to create additional control, repeatability, and safety. It also allows us to test our work without vehicles, without leaving our desks, and for the last few months, without leaving our homes.”
Lyft’s resumption of on-the-road testing comes after the company revealed it would begin tapping data from its ride-hailing network to improve the performance of its autonomous systems. A subset of drivers’ cars — currently Select Express Drive vehicles, as well as Lyft’s autonomous vehicles in Palo Alto and select cars that follow the vehicles for safety purposes — are now equipped with inexpensive camera sensors. This enables them to capture challenging scenarios while helping solve problems like generating 3D maps and improving simulation tests.