CASPA's Autonomous Driving Symposium

Recently, CASPA, the Chinese-American Semiconductor Professionals Association, held a symposium on automotive driving. Having chaired a panel for them once, I seem to be a sort of honorary member (虽然我不是中国人). Even though it was a Saturday afternoon and I'd only got back from Europe the previous evening, I went along to Intel to attend, along with a capacity crowd. Unlike previous symposia, we weren't allowed to take photos of any slides.

The topics covered included Lidar, emotional recognition, safety, cloud connectivity, and more. I was going to cut the material down and put it all in one post, but autonomous vehicles are such a hot topic I left it all in and split it into two posts. So enjoy today's post and drive on to tomorrow for the rest.

Stephen Zoepf

After the welcome presentations, the first person up was Stephen Zoepf. He is Executive Director of the Center for Automotive Research, or CARS (isn't that a great acronym) at Stanford. His talk was titled Data Driven: the Future of Mobility.

He started out with some analysis he had done. Being in in the job he is, of course, he has an electric car. A Chevy Volt. He worked out that it really gets 16 mpg if the calculation is done correctly, the same as an F150 pickup truck. The issue is that he has a short two-mile commute so that the battery uses a lot of energy to heat itself up, and the cabin. By the time the battery and cabin are up to temperature, he is at his destination.

He had some more depressing statistics:

• Safety: There were 40,200 traffic deaths in the US in 2016
• Traffic: Speeds in major cities 10-15mph
• Energy: Highest CO2 contribution is from transportation

Another change that is ongoing is that we need to design vehicles for more intensive use. We are sharing more and more vehicles through mechanisms like Zipcar (22K miles per year, twice that of a privately-owned car) and Uber (60K miles per year, over five times that of a normal car). One thing that I hadn't thought of is whether vehicles should be scrapped after a few years at the end of their lifetime, or should they be refurbished like planes. Planes do get scrapped sometimes, of course, but a lot of the original 747s are still flying. Since Uber and Lyft have over 1M vehicles already, they represent a non-trivial percentage of new vehicles sold.

He had formulas for breaking down the details of this transition and the implications for the number of vehicles sold:

vehicles/year = ( passenger-miles/year x passenger-miles/vehicle-miles ) / miles/vehicle

As the use of Zipcar/Uber styles of car-sharing grows, that reduces the number of vehicles sold. That is before considering if a major societal change will take place with fewer and fewer people feeling the need to own their own vehicle. There are some signs that things are going in that direction, the age at which American teenagers get a driving license is going up and the number of millennials with no car is going up.

Louise Zhang

Next up was Louise Zhang of Lucid on En Route to Autonomous Driving—Designing a Safe Vehicle for the New Era. She has been looking at challenges to road safety, especially in the near future when vehicles will be designed for both autonomous and human driving, and many of the existing vehicle fleet will still be on the road. Nobody entirely knows why, but the crash fatality rate has increased for two years running (up 14% in the last two years) and 94% of fatal accidents are caused by human error. Since the transition to autonomous driving will take a long time (the average age of a vehicle on the road in the US is 11.4 years), she thinks that accident rate might go up due to misunderstandings. You can't make eye-contact with an autonomous vehicle, and Google has experience with being rear-ended since their vehicles "drive like your grandmother." People really do have expectations about how other vehicles will behave based on cues. For example, there are more crashes between two female drivers than two male drivers.

Lucid is really in the business of designing cars for the future. One challenge is how to design cars so that they have all the sensors without being ugly. Google doesn't bother (yet?), Tesla has them hidden. The interior of the vehicle will be dominated by screens, not buttons. But the big difference is that cars will eventually not need to be driver-centric. They will be more like your living room: open, peaceful, tranquil, comfy. They are learning from studying executive jets, and think of cars as more like an executive jet on wheels.

For safety, there are a number of different areas that can make a difference. Obviously, avoiding crashes. But also, when a crash becomes inevitable, reducing the severity and preparing the occupants. Everything from vehicle stability to reduce rollover, to reducing risky behavior when human drivers are in control (speed limiting, drunk driver monitoring).

When a crash does happen, some things to be considered are disconnecting the high-voltage system automatically (to make it safe for rescuers, and to prevent sparks causing fires). The car calls the first responders when occupants cannot, and it already knows the speed of the crash, the number of occupants, where the impact was, the likelihood of serious injury. The military have a golden hour rule that a one-minute earlier rescue results in 2% increase in the chance of survival, so this can make a big difference to survivability.

Standards are getting very out of date. For example, they do not consider automatic emergency brake when the measure frontal crashes. Nothing measures post-crash safety. There are some crazy regulations even today. For example, airbags have to be more powerful (and dangerous) than necessary since they have to cope with the case of unbelted drivers. This remains true even if the manufacturers build in an interlock so the car cannot be driven with an unbelted driver.

The most effort today goes into meeting legal requirements and getting good crash ratings, and none on things that improve safety but are not measured. The pyramid needs to be turned the other way up:

1. Do the right thing
2. Achieve 5-star ratings
3. Meet regulatory requirements

Louise used to work at Tesla and had a story about the challenge to get seatbelt warnings on all five positions, which is not required for certification (only the driver and front passenger).

 FYI, during the Q&A, someone asked when they can buy a Lucid car. It will be in production in 2019. You can reserve on the website today. They will be $60-100K.

Max Liu

Max Liu, of our host Intel, is in their automated driving group and presented on Car-Connectivity-Cloud: Semiconductor’s Enabling Role in Autonomous Driving. Of course, Intel likes to emphasize the cloud aspects of everything since 98% of cloud datacenter processors are manufactured by Intel. He had some big numbers of the savings from autonomous driving when it arrives: $5.6T per year, with $1.3T in the US. This is $488B in accident cost reduction, $507Bin productivity gains, $15H in fuel cost savings, $138B from congestion reduction. These were Morgan-Stanley's numbers, not Intel's.

A connected car will produce as much as 4TB of data per day, which will require a lot of cloud capacity (Intel inside!) to reduce from data to information to wisdom. Intel's automotive driving solution is called Intel Go, "it's not just about driving, it's about connectivity and the cloud."

To be continued tomorrow...