Fear and Loathing on the Path to Level 4 Driving

March 1, 2018

Lindsay Brooke

Tech trust: Demonstrating Valeo’s Cruise4U during rush hour in Paris.

One of the perks of my job is access to the latest vehicles. Such hands-on product exposure is essential, I believe, for understanding the rapid advances of automated-driving technology and other technologies we cover.

This driver’s vantage point has shown me the inconsistency in production ADAS (advanced driver assistance system) performance. The functional deltas that are apparent in lane-keep assist (LKA), lane-departure warning (LDW), automatic cruise control (ACC) and automatic emergency braking (AEB) and blindspot detection are pronounced. This is true brand-to-brand as well as among vehicles of the same brand.

This situation does not bode well for convincing the public that a self-driving future is a good thing. It also shows how much further the industry must go in validating its systems performance at near-aircraft-levels of safety.

Many of the new vehicles I sample have difficulty keeping lane. On a newly-painted highway, for example, various systems will ‘bounce’ the vehicle left-to-right-to-left, like a pinball, within the shoulder-line borders. Others allow the vehicle to drift frighteningly across a freshly-painted center line, until it’s literally within the oncoming-traffic lane. And without alerting me to correct.

That’s the responsibility of lane departure warning systems. They’re supposed to deliver haptic feedback through the steering wheel when the sensors detect the vehicle has wandered. LDWs are consistently annoying, as a 2016 study by the Insurance Institute of Highway Safety determined. About 66% of vehicle owners turn off the feature, which is prone to triggering ‘alert mode’ on curvy roads and when encountering irregular lane markings. IIHS also found that drivers who are less likely to use their turn signals experience false LDW alerts.

When I report the system anomalies to friends working in vehicle integration, their response is often a matter-of-fact, “That’s not uncommon.”

Sometimes the automated driving systems deliver close to the performance that the OEM claims. On a roadway without shoulder and/or centerline markings, or with lines so faded they blend into the concrete, some vehicles do track, keep lane, and uncover blind-spot intruders with high fidelity.

This is not a trust-builder. In this nascent period of ADAS evolution, building 100% customer confidence is paramount. Most people are risk-averse to some degree when it comes to personal safety—I’m an avid motorcyclist, for example, but I’d never skydive.

So how can we feel confident to travel in any SAE Level 4 vehicle if we can’t trust today’s Levels 1 and 2 “driver assistance” features? And how rocky will the transition steps to “self driving” be?

I posed this question to Glen DeVos, CTO at Aptiv (formerly Delphi). He agreed. “The worst thing a driver-assistance system can do is have false positives, where it is constantly alarming or doing something when there is no problem,” DeVos asserted.

The uneven state of today’s system performance, he said, is “a byproduct of where we are in terms of technology deployment into the market. Over time that will normalize,” DeVos said confidently. The performance differential across platforms will diminish.

When that time comes, the industry hopefully will have defined what is “good enough” in systems performance. Will it be 95%? 99%? Or 99.999%? The new era can’t come too quickly.