PALO ALTO, Calif. — In the realm of self-driving vehicles, the number of miles driven in autonomous mode is often used as a proxy for progress.
So it might seem alarming next month when an annual batch of mileage statistics scheduled to be released by the California Department of Motor Vehicles shows Aurora Innovation drove significantly fewer miles in 2019 than in the previous year, when it logged 32,844 on California public roads. But that’s by design.
“We’re going to drive about half as many miles this year,” said Sterling Anderson, co-founder of the company, during a debriefing with reporters here this month.
For Aurora, that constitutes progress. Unlike others, the company wants to limit the number of miles driven on public roads and instead emphasize testing in simulated environments.
Those efforts gathered momentum this year, as company engineers created an internal set of simulation tools that can vet self-driving mettle. As on-road testing halved, simulated testing at the company increased by a factor of 100.
There are countless third-party providers of similar simulation software and services. But Drew Bagnell, another co-founder, says most of those run on game engines. While they are often visually appealing, they don’t necessarily deliver more accurate results.
Having the bespoke simulation tool, called the Offline Executor, allows Aurora to make seamless comparisons between scenarios collected in the real world and those run through simulation.
By reducing its on-road mileage, eschewing flashy simulators and shying away from highly controlled test rides, Aurora has bucked industry trends. It has happily charted its own course in 2019 toward realizing commercial deployment of self-driving technology.
“We don’t do dog and pony shows,” said Anderson, who previously was chief engineer of Tesla’s Model X program and led development of the automaker’s Autopilot driver-assistance system. He co-founded Aurora in 2017 along with Bagnell, previously the architect of Uber Advanced Technologies Group’s autonomous systems, and Chris Urmson, who previously led Google’s self-driving car project.
In June, Volkswagen Group announced the end of a self-driving partnership with Aurora. In September, Hyundai Motor Group, another key automotive partner for Aurora, established a self-driving joint venture with Aptiv. These developments would have shaken many executives, but Aurora’s leaders appear unconcerned.
Hyundai remains a partner and investor. More broadly, Anderson said Aurora intends to deploy its automated-driving system with a wide variety of partners. Not tying itself too closely to one helps the company ensure it can build a system that meets the needs of the widest possible market.
“We want something that’s industry agnostic, vehicle agnostic and use-case agnostic,” Anderson said.
Aurora has raised $690 million to date, including an undisclosed sum from Amazon. Further, the company established a partnership with Fiat Chrysler Automobiles in June. It will use Chrysler Pacifica minivans as the basis of its on-road fleet going forward, and that will be the first vehicle from which it conducts commercial operations with no safety driver aboard.
No date or time frame is yet set for those developments. But whenever Aurora gets there, it will be without much outside assistance.
Beyond its in-house simulation, the company has created its own high-definition mapping system, called Atlas, which enables rapid updates across a fleet. Third-party companies often make maps that Aurora says are “a morass of dense, heavyweight data that requires significant time and resources to maintain.”
In May, Aurora brought its lidar development in-house by purchasing Blackmore, a Bozeman, Mont., company that forgoes traditional pulse-based lidar in favor of continuous beams of light.
Randy Reibel, Blackmore’s former CEO and now vice president of lidar at Aurora, says the company is detecting and collecting data on objects at a distance of up to 450 meters, which can enable and enhance high-speed operations, such as trucks rumbling down interstates.
Last week, Aurora said it was developing its own “teleassist” to help vehicles that get confounded in unusual situations. Aurora is careful to differentiate its approach from that of companies that are developing systems by which remote operators can control vehicles or provide them with waypoints. Rather than tell the car what to do authoritatively, information provided by a remote assistant is considered more of a suggestion that the on-board self-driving system can reject.
It’s one more way Aurora has forged its own, distinctive path toward making self-driving technology a widespread reality.