SXSW 2018 Day 2 Session 1: Who takes the wheel on self-driving car safety?

Session link, including audio: https://schedule.sxsw.com/2018/events/PP80321

David Friedman, consumers report

Cathy Chase, Advocate for Highway Auto Safety

Bryan Reimer, MIT

David Strickland, Self-driving coalition for safer streets

Friedman: To start, some background info: US Drivers drive 3 trillion miles per year.  Americans spend a sixth of their annual income on cars, a billion dollars a day on gas and 7 billion hours in traffic.  This is besides the 37,461 people who died in 2016, and an average of 2.5 million injured a year.

Safety is key in adoption of self-driving cars; bad safety record for self-driving cars can substantially delay in adoption of autonomous cars.

The current state of automation is as follows:

Automation level	Responsibility for safety	Driver control	SAE level	Sold today?
Partial	Driver	Yes	1,2	Yes
Conditional	Depends on the situation	Yes	3	No
Self-driving	Car	No	4,5	No

David F.: What is the future of self-driving car safety?

Cathy: We’re at an interesting point – fatalities from driving are high, but there is an optimism that the problem will be solved by the arrival of self-driving cars.  A few points to consider, though:

1.     There’s still a lot to be done to improve safety with technology before the cars are fully autonomous

2.     There is a lot of improvement that can be achieved just by enacting better safety laws

3.     There’s a slowing down in the timelines.  Car execs are more cautious about target timelines, which is good because if we get them too soon before we can prove their safety it’ll delay the adoption, delaying the reduction in accidents with it.

4.     There is a problem with issue fatigue: we are forgetting that car manufacturers have a long record of misbehavior and outright fraud, and now they are asking us to just trust them with self-driving cars and without regulation.

Bryan: the road to self-driving cars is an evolutionary one that will take time.  Meanwhile, the excitement over self-driving car is going down due to media saturation.  Also, self-driving cars have to pass a higher safety bar: being safer than humans is not enough, because media will pounce on any accident that’s the fault of the vehicle, so even a much lower rate will get higher attention.  The Tesla case is a good example.

In the interim period, humans and computers will be having a joint role in the car.

David S.: We have to work together – manufacturers, governments and consumers.  Our issue is 50 year old regulation that is not keeping up with the technology.  We need to find ways to accelerate the modernization of regulation.  Technology advancements are currently added not through standardization but voluntarily.  For example, ESC was introduced in 1990, but only made standard through regulation in 2012.  This is a long delay; many lives would have been saved if this would have been put in place earlier.

We need to be able to test new technology on the road, on mass; you can’t just test it on a few thousand cars.

Bryan: Fully self-driving cars (with no steering wheels) will not be ubiquitous for 100 years – the complexities of solving the last few problems will just not be economical.  Plus, people will not be willing to give up driving, so we are looking at a mixed environment for the foreseeable future.

Cathy: Why can’t we answer the “when” question? One of the reasons is that we’re not seeing the test data that companies have; if companies were to share the test data we’d be better positioned to assess how mature the technology really is.

David S.: There will always be human drivers.  Some people will just want to drive by themselves.

David F.: Surveys conducted show that 60-80% of Americans are uncomfortable or even afraid to share the road with self-driving cars.  How safe is safe enough? Who gets to decide that?

Cathy: When your computer or phone crashes, you walk away uninjured; when your autonomous car crashes, not so much.  I heard a quote from the head of Duke Robotics that resonated with me – we’ll know the cars are safe enough when the car executives put their families in them.  Google uses some of their self-driving cars with employee families, but they are keeping the information about those tests secret; so when the public has more information it’ll be easier to assess.

It’s also crucial that consumers have a certain sense of safety about self-driving cars as they do with regular cars: show the safety information up front, like you do with standard features.

David S.:  We need to do vigorous self-assessment to determine that safe is safe enough; but we have to be able to deploy.

Bryan: Need to think about safety as a continuum; need to build trust in the system, see a more gradual, slower approach.  It’s hard to define official performance indicators: what, for example, is considered a crash, for the purpose of measurement?  What is a near crash?  These parameters are hard to define.  Instead we need to make any incident public – even a dent must be disclosed.

David F.: Will the public accept a small reduction in the number of accidents?

Cathy: Would we accept this in an airplane?  No, so why would we accept these numbers in cars?  If we have comparable numbers of accidents between self-driving cars and regular cars people will freak out.

Bryan: Aviation provides a good comparison point.  We are still in early days of the technology, like aviation was in the 50s or 60s.  It has to be managed over the long haul.

David S.: People are all bad at risk assessment.  People tend to be unrealistic about driving risks.  Everyone thinks they’re a great driver, while everyone else is a bad driver.  This notwithstanding, every incident with a self-driving car will be pounced upon by the media.  For example, airbags – in the beginning they were unsafe for kids.  We could have said that airbags are unsafe and discarded the technology, which would have probably resulted in hundreds of thousands of deaths over the years; however, we did not stop using airbags or slow their adoption, we just moved the kids to the back until we could solve the problem.

Cathy: And indeed, there were some calls to get rid of airbags, but it was regulation that kept them in.

David F.: The National Highway Traffic Safety Administration (NHTSA) is an agency with a $900M budget, 2/3 of which go to states, with over 37,000 fatalities annually; in comparison, the Federal Aviation Administration (FAA), $16 Billion budget, covering fewer vehicles travelling fewer miles.  Can NHTSA handle this?

David S.: Eventually yes.  Certainly, NHTSA needs to be better funded.  The aviation industry is tightly regulated – you can’t make even small changes to aircraft without FAA approval.  The FAA has 57,000 employees to take care of this, while NHTSA has only 575.  It’ll be incumbent on the manufacturers to keep safety and NHTSA will need to watchdog that.

Cathy: 99% of accidents and 95% of fatalities happen on the road, and NHTSA only gets 1% of the Department of Transportation’s budget.

Bryan: NHTSA has its hands full managing regular cars, with the automation put into those.  When looking at autonomous car, we’re talking about a software world that has a different problem profile than the mechanical one has.  It is much closer to a pharmaceutical industry, so the model we should be looking for is closer to the FDA’s process for drug approval.

Question: What about liability?

David S.: It depends on the level of autonmy the car has.  If there is a human in control, the driver is responsible.  If it is fully autonomus, the manufacturer is liable.

Bryan: We will get to a situation where sensor data will be used in courts to determine responsibility.

David F.: And there is different thinking about this even inside the government.  For example in the case of the Tesla crash, NHTSA ruled that Tesla was not liable, while the National Traffic Safety Board concluded that both the driver and the manufacturer were equally responsible.

Question: Will human drivers be banned in 30 years time?

Bryan: Not in the US – possibly places like Singapore

David S: No.

Cathy: No.  You can’t even get helmet laws passed in some places in the US, there’s no way drivers will be banned.

Question: Will you be out of a job?

Bryan: No.

David S: Hopefully.

Cathy: It’s my job to put myself out of a job.