汽车 > 车联网博客



AVS2016: The Autonomous Week That Was

by Roger Lanctot | 7月 26, 2016

Last week began with news of Germany pondering the requirement of a crash-reporting black box for autonomous vehicles and China considering a ban on testing of autonomous vehicles on public roads.  The week concluded with U.S. regulators encouraging autonomous vehicle testing in public comments at the Automated Vehicle Summit 2016 put on by the Association for Unmanned Vehicle Systems International and the Transportation Research Board.

Autonomous vehicle advocates at the AVS2016 event decried California's draft automated vehicle testing policies as too restrictive while critics called for more severe U.S. Federal regulation.  In fact, a representative of Consumer Watchdog called for Tesla to reprogram its autopilot function to require hands on the steering wheel at all times and also demanded, in a letter to Tesla Motors Chairman Elon Musk and National Highway Traffic Safety Administration Administrator Mark Rosekind, that Tesla and NHTSA accept responsibility for the recent fatal Model S crash.

Consumer Watchdog emphasized its point by driving a white truck (pictured above) to the Union Square Hilton, venue for AVS2016, to remind attendees of the white truck implicated in the recent Tesla Model S crash in Florida.  The Model S's radar- and camera-based autopilot system is thought to have been confused by the white truck implicated in the crash - an outcome that might have been avoided or mitigated had a laser-based LiDAR system been in use.

In their remarks at the event, Transportation Secretary Anthony Foxx and NHTSA's Rosekind announced no new initiatives nor the finalization of promised testing guidelines.  Rosekind, in particular, indicated that NHTSA is strongly interested in fostering advances in automated driving by providing a regulatory framework that favors no single solution.  Perhaps equally important, he said the agency does not wish to impede this development progress, noting that there will be setbacks such as the fatal Tesla crash, but that no single event should shut down technological advances.

There is good reason for caution and ambiguity in the regulatory process.  Several speakers on the topic of machine learning noted the hundreds of millions if not billions of miles necessary to be driven to achieve even a 90% confidence level for certifying automated driving as safer than human driving.  As if to put an exclamation point on the regulatory vapor lock, a senior executive of Polysync proposed replacing the automotive industry's familiar, but demanding, functional safety criteria (focused on anticipating all possible system inputs) with a so-called "operational safety" protocol focused on outcomes.

While Germany is reportedly considering the fitment of a black box in autonomous vehicles to report crash data, California is preparing the way with its own data sharing protocol - tracking miles-per-disengagement and type of disengagement for autonomous vehicles being tested on California public roads.  ("Disengagement" refers to conditions when the driver retakes control of the car from the autopilot.)  Google was an early critic of California's proposed testing guidelines, but the company is acing the exam with steadily improving scores and detailed accounting of event types.

SOURCE: California Department of Motor Vehicles/Google results

More details: www.dmv.ca.gov

No other car company is providing as much detail in its disengagement reporting to the California DMV and no other car maker has equaled Google's Level 4 performance.  A similar picture emerges from twin studies, one by Car&Driver and one by Motor Trend, of semi-autonomous (ie. Level 2) vehicle systems, with Tesla far in the lead based on disengagement analysis.

http://tinyurl.com/jo4cnu3 - The Physiology of Semi-Autonomy and Test Results - Car & Driver
http://tinyurl.com/zfj4wvs - Testing (Semi) Autonomous Cars with Tesla, Cadillac, Hyundai, Mercedes 

What emerges is a picture of automated driving that originated with the Defense Department of the U.S. government being adopted and promoted by disruptive organizations, Tesla and Google, and now being promoted by the U.S. Department of Transportation.  The automotive industry could not take the lead on vehicle automation because the concept ran counter to everything every electrical and mechanical engineer had been taught about functional safety.

The USDOT's unambiguous support of vehicle automation was the single most important take-away from AVS2016.  The second most important message was the possible necessity of adding LiDAR technology to the sensing portfolio of autonomous vehicles. 

It's good to see the USDOT hold its ground in its support of automation while China and Germany dither.  It's especially heartening to see that California's seemingly restrictive approach to testing has nevertheless produced an enlightened approach to data gathering.  It makes more sense to understand what is happening BEFORE crashes happen or when disengagements occur rather than after - as in the case of Germany's proposed black box.

Previous Post: Car Theft Making a Comeback | Next Post: Beware of Competitors' Praise
Leave a comment