But 10 months in, they are pleased with the progress of a team that contains nearly 300 engineers. Lyft is developing software for the self-driving systems, while Magna is focused on outfitting the vehicles with systems, and eventually manufacturing them at scale.
“We’ve made software changes, probably multiple times a day, based on what we’re learning, ranging from perception to planning, and planning is one of the biggest challenges. ‘OK, I’m perceiving these objects. What is the right path I take?’” said Kapoor. “That is a very hard thing to do, because there are a lot of unknowns.”
Testing has included a 10-week pilot in which self-driving vehicles — with human safety drivers behind the wheel — ferried Lyft employees from their Level 5 office in Palo Alto, Calif., to and from a nearby Caltrain station. Helping connect commuters with public-transportation options and their homes or offices is a complementary role that Lyft foresees for robotaxis — something the company has piloted in traditional-driving projects in recent years. So it makes sense that similar routes are the focus of early AV testing.
At the same time, Lyft has been integrating technology from Blue Vision Labs, a London-based augmented-reality company it acquired in October, into its self-driving plans. Software from Blue Vision Labs can use cameras in smartphones to better perceive the environment, and this could help add context and information for self-driving operations — potentially diminishing the need for more expensive sensors.
“We can leverage 1.4 million drivers that we have in our network, and get a ton of data into the scenarios our drivers are actually going through, and look at the frequency of that in a histogram,” Kapoor said. “We drive that back into development, and it helps us understand the question of, ‘When are we ready?’”
For Magna, the partnership gives the company firsthand experience in preparing to manufacture self-driving systems. It further provides insights not only on fully autonomous systems, but on how the company might use some of those learnings in driver-assist systems that are in production today or in the production pipeline.
Sometimes those lessons learned trickle down from full-autonomy testing into ADAS systems, and sometimes they are carried upward from driver-assist technology into the full-autonomy testing.
“With our knowledge of functional safety requirements, that can be taken and used as a foundation for Level 4, Level 5,” said Magna chief technology officer Swamy Kotagiri. “On the algorithm and software side, that absolutely can bring faster development on different features. … So this is a significant piece of what we are doing, and fits within our roadmap for getting design wins at Levels 1, 2 and 3, and then looking ahead to the mobility systems of the future.”