Nvidia, Intel jaw pro and con of Level 3 vehicle autonomy

Engineers continue to wrangle over Level 3 vehicle automation, which strikes at the center of AI innovations and safety.

Intel and Nvidia recently, again, voiced opposing views in the years-long debate over whether Level 3 autonomy should be used in vehicles.

Level 3 is a designation of autonomy defined by the Society of Automobile Engineers.  It is almost midway through a scale from 0 to 5 that ranks the least amount of driving automation at Level 0 and the most at Level 5.

Automation at Level 3 means vehicles are equipped with sensors, cameras and sometimes LiDARs to detect the road and traffic, as well as weather conditions, with the vehicle itself using that data and AI to automatically steer, accelerate and brake.   If the sensors cannot perceive road conditions during heavy fog or snow, human drivers are alerted to resume command of the vehicle, usually with a repeated beeping sound.

However, the amount of time it could take for a human to resume control from the Level 3 vehicle automation system, perhaps 30 seconds, has given some carmakers and Intel’s Mobileye division pause, to say the least.

“Level 3 is a very problematic niche. We do not believe in Level 3,” Intel Mobileye CEO Amnon Shashua said flatly in a recent call with reporters.  The call was originally held to discuss Intel’s recent $900 million acquisition of mobility services company Moovit.

Instead of Level 3, Mobileye is heavily focused on Levels 4 and 5 with regulated robo-taxis that would operate in a geo-fenced area like Tel Aviv, starting in 2022.

Over at Nvidia, a different attitude prevails.  “With Level 3, speaking as the platform provider, it’s really up to the customer to decide what [automation] features to include,” said Danny Shapiro, senior director of auto and robotics at Nvidia, in a recent call with reporters.  “The key with Level 3 is to make sure there’s not a handoff situation that is unsafe.” 

That usually means carmakers are left to define how a vehicle will respond in different environments. “We’ll see all levels continue to develop and some will be more popular than others. For us, it is a continuum,” Shapiro added.

With software defined cars, new features around automation can be added so that regardless of what level they are sold at, “they can always get better,” he said.

China-based Xpeng's new P7

Some models of new cars include automation features that approach Level 3, but no self-driving cars are currently on sale from any manufacturer. However, China-based Xpeng recently announced the P7 all-electric sedan running Nvidia Drive AGX Xavier AI technology that includes Level 3 features for steering, acceleration/deceleration and passing of other cars without human input.

RELATED: COVID-19 cuts into self-driving chips on top of crushing car sales

The P7 relies on 12 ultrasonic sensors, five millimeter-wave radars and 14 cameras with 360-degree perception, but no LiDAR. Its forward-looking radar can detect obstacles at 200 meters, penetrating rain, fog and haze, Nvidia said. It starts at about $32,500 and deliveries begin in June.

Audi's hesitation and upcoming L3 launches

As a counterpoint to the P7, Audi in late April held off the addition of its Level 3 Traffic Jam Pilot system in the A8, which was meant to handle driving in traffic jams or slow-moving highway traffic of up to 38 mph. Part of the reason was that a widespread global regulatory framework over the driver’s liability (for not taking back control when required) was not established as Audi had hoped back in 2017 when the model launched.

With the disastrous impact of COVID-19 and the shutdown of many auto plants globally, it isn’t clear that all the Level 3 vehicles previously announced for release starting in mid-2020 will indeed have such features.  These may include the Honda Legend for Japanese markets that is designed to allow drivers to take their hands off the wheel and their eyes off the road for use during slow traffic on a freeway.

South Korea is also set to allow automakers to sell vehicles with basic Level 3 automation in that country. By 2021, Hyundai, Kia, BMW and Mercedes-Benz are expected to have Level 3 vehicles for consumer purchase, according to Innovation at Work, an IEEE publication.

U.S. forecast for AVs amid plant re-openings

In the U.S., the future for Level 3 (and Levels 4 and 5) with major carmakers is cloudy, partly with the impact of COVID-19 devastating car sales and, somewhat, future research and development into AVs.

GM’s self-driving car business, Cruise, recently said it would lay off about 150 workers, or 8% of its workforce, according to unnamed sources familiar with the matter who told the Wall Street Journal. However, GM CEO Mary Barra has told investors there’s no plan to curb Cruise funding, and Cruise has received billions of dollars from outside investors.

In April, GM officials shut down the carmaker’s Maven car-sharing business after nearly four years of operation citing coronavirus disruption that has devastated Uber and other car-sharing businesses.  The decision affects dozens of fleets of vehicles in the U.S. and Canada.

At Ford, executives are re-evaluating business plans for AVs because the pandemic might push buyers to purchase personal vehicles to avoid possibly infections with shared services.  In April, Ford said it would delay introducing a commercial AV service slated for 2021 until a year later.

GM and Ford have spent multiple billions of dollars developing AVs since 2016, mainly with a focus on robo-taxis and package delivery.  Because their investments have been so large and because AVs are promising long-term, five analysts recently told FierceElectronics that major carmakers won’t contemplate giving up on AV technology although delays could last a year or longer. That long-term future effort by carmakers is especially likely with a strong push from semiconductor makers like Nvidia and Intel that clearly want to supply AI chips and software for greater vehicle autonomy, the analysts said.

An optimistic sign for U.S. carmakers GM, Ford and Fiat Crysler is that production lines for vehicle assembly began gradual reopening on Monday (May 18)  after nearly eight weeks of hiatus because of COVID-19.  Factories are using safety measures like temperature screenings and personal distancing to keep workers safe.

Toyota, Honda and Tesla started opening plants last week, while Mercedes-Benz restarted a plant in Alabama in April but then stopped last week because of a parts shortage.  Volkswagen wanted to make cars again at a Tennessee plant on Sunday but delayed as suppliers caught up, according to the New York Times and other reports.

Prolonged debate

The Level 3 debate goes back to at least 2017.  Recently, Intel reasserted its concerns about automated vehicle safety in a strategy and goals statement for 2030.  The company said it will work to collaborate to reduce 1.3 million traffic crash deaths each year through adoption of technology neutral safety standards such as IEEE 2846 to offer clear guidance on what it means for an AV to drive safely.

 In late December, IEEE committed to the 2846 approach to develop a rules-based, mathematical model for AV decision making. It starts with a 2017 approach Intel outlined that was called Responsibility-Sensitive Safety (RSS).

Five rules (only five?)

RSS has five safety rules based on formal logic to keep AVs safe with human drivers/passengers:

1. Don’t hit the car in front of you.

2. Don’t cut in recklessly.

3.  Right of way is given, not taken.

4.  Be cautious in areas with limited visibility.

5.  If you can avoid a crash without causing another one, you must.

It’s clear that the debate over Level 3 goes far beyond whether chips can be produced to process sensor data quickly and intelligently.  Some of the discussion centers on how governments, insurance companies, carmakers and—yes—chipmakers view the relationship between AI in a car and a human inside.

How do you feel about the Level 3 debate? I’m interested in hearing from you at [email protected].