Karl Iagnemma, MIT rock-star scientist, as Gary Silberg likes to call him, may have more perspective on the inner workings of autonomous vehicles than most people on the planet today. But that didn’t stop him from acknowledging the two most basic questions the public asks about AVs: “When are we going to have these cars on the road? And why aren't they out there now?”
The first thing Iagnemma would say in response to these questions is that Aptiv autonomous vehicles have already arrived. In May 2018, Aptiv launched the first and largest commercial autonomous vehicle service available to the public in Las Vegas. These AVs provide access to over 1,600 entertainment venues, hotels, restaurants, and popular locations throughout the City of Las Vegas and Clark County – including the Las Vegas Convention Center and City Hall.
Today, Aptiv autonomous vehicles have provided over 35,000 public rides with a 4.95 out of 5-star passenger rating. In addition to its AVs in Las Vegas (which can be hailed by passengers on the Lyft network), Aptiv operates R&D vehicles in Boston, Pittsburgh, and Singapore.
“I think we all appreciate that, today, autonomous technology is really expensive,” said Iagnemma, president of Aptiv Autonomous Mobility. “The average end user will likely not be able to afford a personally-owned autonomous vehicle for a very long time from now. I'm talking about full level four—hands off, eyes off driverless technology."
"But what we’ve identified at Aptiv is that autonomous mobility services will be adopted at a more rapid pace. That is, the use of robotic vehicles as taxis and ride-hailing vehicles. Offsetting the cost of a human operator allows us to provide better access to affordable transportation while adding richer content into the car. This is exactly what we’ve created at Aptiv Autonomous Mobility."
While Iagnemma leads a team that has successfully deployed AVs in four cities, on two different continents, and is the first to enable a public autonomous mobility service – the question of scalability is still at hand
For Iagnemma, the key to scaling lies in two dimensions to safety: Technical safety and perceived safety.
Technical safety: Deploying a vehicle that verifiably responds to the rules of the road, avoiding collision and operating safely in its environment.
Perceived safety: In order to adopt autonomous technology, passengers not only want to know that the vehicle is technically safe, they also want to feel safe. This means riding in a car that drives like we drive-- in a safe, human-like manner.
Which begs a new question: How do we develop autonomous vehicles that drive like a human? What do we, as human drivers, do? In Iagnemma’s view, human drivers:
Understand and obey the rules of the road.
Actively prioritize comfortable, safe driving while avoiding accidents (even when it means breaking the rules).
Adapt to our cultural surroundings. For example, driving in Singapore has its differences from driving in Boston.
"When you put these things together, you can imagine this is a very dynamic set of rules that we as humans operate by,” said Iagnemma. “And this is hard for autonomous vehicles and computer systems, which in general don't like flexibility and uncertainty.”
As a developer, Iagnemma believes the need to program autonomous vehicles to drive like humans is one of the industry’s core problems and one of the factors hindering the ability to scale.
To address this deficiency, Aptiv is creating an autonomous vehicle “rule book,” called Structured AI, a system for compiling various types of rules of the road. “It’s a system of common practices and driving preferences for autonomous vehicles,” said Iagnemma. “It also defines a rule hierarchy, because, of course, not all rules are created equal.”
The primary rule—safely avoid collision—sits at the top of the hierarchy and must be followed, even if it means violating another rule, which is therefore a lower priority. The goal, according to Iagnemma is to make the rules easy to explain, understand and audit, as needed.
The thinking goes, on one hand, if you try to innumerate every possible scenario the car may encounter, you're going to spend a lot of time doing just that. On the other hand, if you're going to try and find tangible examples of every scenario the car may encounter, it's a similar issue, just with a different technical approach. “There is a sweet spot in the middle where we’ve found another solution,” said Iagnemma
With that said, Aptiv is focused on building positive perceived safety with passengers, and the numbers are encouraging. Not only are riders giving Aptiv AVs a nearly perfect rating, nine out of 10 say they would ride again. On the aforementioned issue of perceived safety, this rating system is a simple, instructive measure of how passengers enjoyed the ride and how comfortable they felt with the technology.
To learn more about Karl Iagnemma and Aptiv Autonomous Mobility’s approach to safely scaling autonomous vehicles, read: Reaching Safety at Scale May Involve Some “Rule Breaking”