Autonomous Driving Software

The Navigator: Traversing the autonomous trust gap

Safety and robustness are critical in automated driving, and the threshold for minimum viability is much higher than typical Silicon Valley software projects.
During a recent conference hosted by semiconductor maker NXP in Silicon Valley, I had the honor of moderating a panel on the Autonomous Trust Gap. The timing was fortuitous in the wake of several recent road incidents involving partially automated vehicles that had been developed by companies based in the region. While the panel didn’t include any of the incumbent automakers, it did feature executives from EV startup Byton, software companies Renovo and Cloudera, and chipmakers NXP and Nvidia.

The entry of the SilV tech community into the automated driving space over the past few years has been a somewhat worrying development. Not so much because of what they could do to the incumbent industry—after all, every business sector must eventually evolve or die. My personal concerns have been more related to the tech mindset that has developed over the past two decades: Move fast and break things—and don’t be too concerned that you’re shipping minimally viable products.

Tesla has been selling its $5,000 AutoPilot driver assist product as a Beta, while startups like Comma.ai and its founder George Hotz have been claiming they can do automated driving with the equivalent of a smartphone. It has often seemed like the SilV entrepreneurs were more interested in just growing revenue as fast as possible while turning existing businesses outside out, than actually providing a real safety benefit.

With at least three known AutoPilot-related crashes including one fatal one this year that involved running into stationary objects, and an automated Uber test vehicle running down a pedestrian, it is increasingly being demonstrated that the threshold for minimum viability in automated driving is much higher than for typical SilV software projects.

The good news from the panel, and increasingly from the conversations I’ve been having across the region in the past year, is that this message seems to be sinking in. Startups that previously thought they could disrupt one of the world’s largest industries in the way Uber has turned the taxi business upside down are realizing that safety and robustness are critical to ensuring that people will be willing to give up the act of driving. An automation system that randomly decides it doesn’t know where it is or where it should go—and throws control back to a person that has been disengaged for an indeterminate period—isn’t going to cut it.

Collaboration among OEMs and suppliers on technology development remains key. Nearly half of the intellectual property in new vehicles now originates from the supply base. With OEMs knowing they are going to be on the hook for the behavior of automated driving systems, they are looking for the best technology they can get to enhance those systems.

Conversely, startups are recognizing that they have a better chance of building a sustainable business by partnering with OEMs and Tier 1 suppliers than trying to go it alone.

However, with recent surveys indicating a large majority of people still don’t trust self-driving vehicles, technology itself isn’t going to close the gap. The way we as an industry communicate with average users is perhaps even more important. SAE J3016, the taxonomy for automated driving is increasingly seen as part of the problem. The levels, particularly those from partial to full automation, have been both too precise in labeling and too open-ended in definition.

Written by engineers for engineers, the SAE Levels were inappropriate for communicating to a non-technical audience and have been frequently misused and abused for marketing purposes. The premise around the strategy and messaging for partial or conditional automation needs a rethink. A newly updated version of the definitions provides some added clarity about where certain functions fit. But inconsistent branding by OEMs is still likely to cause confusion. Driver augmentation or envelope protection (as Toyota is developing for its Guardian system) seem far more robust.

My panel was in agreement that more candor is needed when companies discuss the capabilities and limitations of their technology. Even if that means it falls short of the hype, you can build more trust in customers by under-promising and over-delivering than the reverse. Continue reading »
X