Q&A: ASU researcher says self-driving cars aren't foolproof

By

Scott Seckel

From push-button parking to hands-free autopilot, self-driving technology has arrived to the U.S. auto industry as developers push to deliver features that have previously been the stuff of science fiction.

Despite a recent fatal crash, carmaker Tesla this week reiterated a commitment to advancing autonomous vehicle technology, and the nation's top road safety regulator said such innovation remains a priority because of its potential to reduce accidents. 

The developments raise several questions: Exactly how autonomous are self-driving cars? What remains to be done to improve safety? Is this technology roadworthy, or is there still a bit of work to do?

To find out, ASU Now turned to Ram Pendyala, an incoming professor at Arizona State University who teaches and conducts research in multimodal transportation systems planning and engineering. He has published nearly 80 articles in journals, books, and proceedings, completed more than $4 million in sponsored research, and holds leadership positions in professional associations such as the Transportation Research Board and American Society of Civil Engineers.

Pendyala's answers come as investigators look into a fatal crash in May involving a Tesla Model S and a tractor-trailer making a left turn. It was the first traffic death involving a self-driving car. Tesla officials have said the vehicle failed to tell the difference between the bright white truck and the sky and that neither the car's systems nor the driver applied the brakes. The tractor-trailer driver, meanwhile, told The Associated Press that the man killed had apparently had been watching a Harry Potter movie. The Tesla driver’s relatives have stated that they will cooperate with the investigation and that they hope it leads to safety improvements.

Tesla promotes the Model S as having autopilot, but it never claims the car is fully self-driving. The cars will auto-steer, self-park and even change lanes on their own in some cases. But the company characterizes the technology as an “assist feature,” something of an advanced cruise control, rather than an opportunity for drivers to check out for long stretches.

Q: How safe are these cars?

A: That’s a bit of a loaded question. There’s not a lot of cars on the market that boast of having autopilot capabilities. A lot of enhanced automotive features (already exist) like automated braking technology and the automatic parking technology, and cars that begin to sense if you’re getting drowsy and starting to drift, they self-correct. But in terms of real autopilot, there aren’t a lot of cars. In terms of being available to the mass market, there aren’t a lot of cars that have that feature. As Tesla states, the autopilot feature does permit you the ability to have arms off and legs off, and it will even change lanes, but that doesn’t mean the driver can be completely disengaged from the driving task. There is the expectation that the driver will be focused on the driving task. They have to be ready to take control at any moment. In this case, the autopilot capabilities aren’t foolproof.

Q: What can be done to improve safety?

A: There are probably some scenarios out there that have not undergone sufficient tests, yet. Before (fully autonomous vehicle) technology comes mainstream, there will be billions of miles of additional testing and thousands of scenarios that have not been covered so to speak. I would say drivers need to be focused on this task. … A semi-autonomous vehicle is exactly that; the driver has to remain engaged.

Q: Are people expecting too much of the technology?

A: If you look at a lot of the research we do at ASU, it’s about driver behavior. Let’s say this driver used this feature 100 times, and on a 100 trips it worked beautifully. The driver now has a certain expectation. You’re thinking the 101st time it’s going to be just as perfect. There now is a driver expectancy and driver behavior aspect to take into consideration. … Human driver behavior as it is, expectations begin to grow as you become familiar with the technology.

Q: There are other issues beside the automotive technology. Infrastructure needs to catch up as well, doesn’t it?

A: It’s not just an automotive technology issue. It’s as much an infrastructure issue. A lot of the research at ASU involves thinking about what we need to be doing on the infrastructure side. … Indeed, a lot of these automated self-driving vehicles need well-delineated lanes in order to sense the position of the vehicle.

Q: Where is the technology headed?

A: There are two types of technology: connected vehicle technology where they talk to each other through dedicated short range communications. Those are not automated vehicles. They are human-driven vehicles with a view towards assisting the driver and the vehicle to act and react in a safe manner.

The autonomous vehicles are meant to be standalone cars that can drive themselves in any type of environment without a worry. They sense everything happening around them and navigate the path.

A lot of this is beginning to emerge as CAVs — connected and automated vehicles. They are both autonomous as well as connected. On the one hand you have the autonomy, so the human is taken out of the equation. On the other hand, all of these vehicles are connected so they act in a more responsive manner ... You have to harness the power of both.

 

Top photo courtesty of Tesla Motors