Smart Driving Cars episode 411: Aurora, LeCun, Uber, AI & more
About this episode
Chris Hermsen of Aurora gets discussed after an MIT Mobility Forum chat focused on responsible autonomy and the “trolley problem,” with the hosts arguing perception and risk are handled differently by 360° sensing at high update rates. Princeton’s Alan Cornhouser then highlights a talk by Yann LeCun on enabling the next AI revolution via AMI and the importance of context/worldviews. News coverage spans Uber’s $10B robo-taxi push, attendant-vs-driver legal/regulatory nuances, and why Tesla/Waymo/Uber may miss real rider needs. The episode also touches AI-assisted data visualization and affordable mobility gaps.
It's episode 411 of Smart Driving Cars with Princeton's Alain Kornhauser and co-host Fred Fishkin. In this edition, Aurora's Chris Urmson fields questions at the MIT Mobility Forum, AMI's Yann LeCun on campus in Princeton, Uber commits 10 billion to robotaxis, Axios reports mobility's new big three and more. Join Alain and Fred for the latest and subscribe!
0-60 time
The transcript does not mention 0-60 time.
The transcript does not mention 0-60 time.
Aurora CEO Chris Hermsen
"Starting on top in the latest smart driving car newsletter, the MIT mobility forum and chat with Aurora CEO Chris Hermsen. [54.5s] Yeah, so and once again, the MIT forum was really good and Chris was was really forthright."
Aurora is a company working on self-driving technology. Chris Hermsen is a leader there, and the hosts are talking about what he said at a big industry event.
Aurora is an autonomous driving technology company, and Chris Hermsen is one of its top executives. In this segment, they’re discussing Aurora’s approach to self-driving systems and how it was presented at the MIT mobility forum.
MIT mobility forum
"Starting on top in the latest smart driving car newsletter, the MIT mobility forum and chat with Aurora CEO Chris Hermsen. [54.5s] Yeah, so and once again, the MIT forum was really good and Chris was was really forthright."
The MIT mobility forum is a conference where people talk about transportation technology. Here, it’s where Aurora’s CEO was interviewed or discussed ideas.
The MIT mobility forum is an industry and academic event where leaders in transportation and autonomous driving discuss technology and policy. In this segment, it’s the context for the hosts’ conversation with Aurora’s CEO.
trolley problem
"And then there was just those, you know, really bad question. I mean, we had to discuss that trolley problem again. I mean, are you kidding? Whatever. Well, Chris is one of these guys who's really making things happen and doing it in a responsible way, right?"
The trolley problem is a famous “what would you do?” ethics puzzle. People use it to talk about what self-driving cars should do in scary, unavoidable situations.
The trolley problem is a classic ethics thought experiment used to discuss how autonomous systems might make decisions in unavoidable harm scenarios. The hosts connect it to how perception and decision-making work in real driving, where the “two things at once” framing doesn’t match how a driver’s attention is actually distributed.
30 times a second
"So, of course, these things as Chris properly mentioned, they look simultaneously 360. Just because they do that, they do it 30 times a second and fast and fast faster than I could. My cognitive cycles aren't 30 times a second."
This is about how often the car’s computer checks what’s happening around it. If it updates more times per second, it can respond faster to changes than a person can.
“30 times a second” refers to how frequently a system can process sensor inputs and update its understanding of the world. In autonomous driving, higher update rates can help the vehicle react quickly to fast-changing situations, even if a human’s attention doesn’t work that way.
360
"So, of course, these things as Chris properly mentioned, they look simultaneously 360. Just because they do that, they do it 30 times a second and fast and fast faster than I could."
“360” means the system is trying to see all around the car, not just straight ahead. The idea is that the computer can track multiple directions at the same time.
“360” here describes a full-surroundings perception approach—using multiple sensors to cover the vehicle’s environment around it. The point is that an autonomous system can monitor many directions at once, unlike a human driver who typically focuses attention in one direction at a time.
enabling the next AI revolution
"[221.8s] I think they enjoy that. No, they don't. From MIT, we'll head south to Princeton, [229.7s] where there was a talk by Jan LeCun right there, enabling the next AI revolution."
They’re talking about AI changing in a big way—like a new approach that could make AI much more capable. The host connects that idea to Jan LeCun’s talk at Princeton.
The phrase “next AI revolution” is about a major shift in how AI systems are built and deployed. In this context, it’s tied to Jan LeCun’s work and the idea that new approaches can unlock qualitatively better AI capabilities.
"[237.7s] He's the co-founder of a pretty unique AI venture now. He was at Facebook running the whole AI [243.6s] industry."
They mention Facebook because LeCun worked there and helped run AI work. It’s a clue that his ideas have been tested in a real, large company environment.
Facebook is referenced as the place where Jan LeCun previously led major AI efforts. For listeners, it signals that the speaker is discussing AI leadership and research coming from large-scale industry deployments, not just academia.
AMI Advanced Machine Intelligence
"[249.9s] So his new venture is called AMI Advanced Machine Intelligence, [254.4s] and they're trying to do things in a different way, it seems."
AMI (Advanced Machine Intelligence) is described as Jan LeCun’s new AI venture. The name and description suggest a focused effort to pursue AI in a “different way,” implying a distinct research and product strategy versus mainstream approaches.
worldview, the context
"[283.4s] everything almost. And unless you have the worldview, the context, and that is at least [292.8s] your worldview. I mean, we don't even know what the correct worldview is."
They’re saying that what AI “understands” depends on the situation it’s in—its context. If the assumptions don’t match reality, the AI can give the wrong answer while sounding confident.
The host emphasizes that AI performance depends not just on raw data, but on the “worldview” and “context” used to interpret information. This is a common theme in AI safety and reliability: systems can be confidently wrong if their training assumptions don’t match real-world context.
"[305.8s] even the, then I put it in, or maybe I'll put it in if I may have forgotten. What was the thing in [305.8s] the New York Times a couple of days ago? Oh, Google is not 85% correct."
Google is referenced in a discussion about reported AI accuracy (e.g., “not 85% correct”). This matters because AI evaluation depends on the dataset, task definition, and error types—so headline accuracy percentages often hide important details.
New York Times
"[299.3s] your worldview. I mean, we don't even know what the correct worldview is. And I think I put in [305.8s] even the, then I put it in, or maybe I'll put it in if I may have forgotten. What was the thing in the New York Times a couple of days ago?"
The New York Times is mentioned in relation to a claim about Google’s accuracy. This is relevant as an example of how AI accuracy is discussed publicly, and how “percent correct” framing can be misleading without defining what’s being measured.
Robo Taxi's
"Terrific. From the Financial Times, Uber commits $10 billion to Robo Taxi's in strategy shift. Yeah, I mean, you know, if you, if you go back, you go back to sort of the beginning of Uber..."
A robo-taxi is a taxi that drives itself. Instead of a human driver, the car handles the driving, and the company still has to prove it’s safe and get permission to operate.
“Robo-taxis” are fully or mostly autonomous ride services where a vehicle drives itself to pick up and drop off passengers. The big challenge is not just driving technology, but also safety validation, operations, and regulatory approval.
Lane Herzberg
"then I mean, they were going to do, I don't know, at least from my perspective, they were going to involved in Lane Herzberg having an end of life occurrence. And I don't know, I haven't dug deeply into it..."
Lane Herzberg is mentioned in connection with a tragic incident during autonomous vehicle testing. When people talk about it, they’re usually discussing how self-driving companies handle safety and what they learn after serious mistakes.
Lane Herzberg is referenced as an “end of life occurrence,” which points to a well-known fatality involving autonomous-vehicle testing. In AV discussions, such incidents are often used to evaluate safety processes, oversight, and how companies respond after failures.
driverless trucks
"And he was also asked the tough question, you have driverless trucks out there running, is there anybody in the cab? ... there is an attendant in there, but they're going to go to a driverless thing soon."
Driverless trucks are trucks that can drive themselves using sensors and software. Even when they’re “driverless,” companies may still keep a person in the cab at first to watch and take over if needed. The big question is how safe and practical that is, and what the law requires.
“Driverless trucks” refers to heavy trucks operating with autonomous driving systems instead of a human driver actively controlling the vehicle. In practice, many deployments still include a human in the cab as a safety attendant while the system proves reliability and safety. The discussion here centers on how that human’s role affects legal and operational risk.
attendant
"there is an attendant in there, and they're going to go to a driverless thing soon... the attendant on board is not considered to be a driver, but is considered to be an overseer or something like that."
An “attendant” in autonomous trucking is a human onboard who monitors the system and can intervene if something goes wrong. The transcript distinguishes this role from a traditional “driver,” which matters legally and operationally. This is tied to whether the attendant is subject to driver-specific regulations.
business case
"Because what he's still doing is demonstrating that the technology works. He isn't out there really doing the business case and executing the business case."
A “business case” is the argument that something is worth doing because it makes sense financially. Here, they’re saying it’s not enough for the technology to work—you also have to show it can be run safely and profitably. Risk and value are part of that calculation.
A “business case” is the justification that an autonomous or driverless system is financially worthwhile—considering costs, risks, and expected benefits. The transcript contrasts demonstrating the technology works (technical proof) with proving it’s profitable and operationally feasible (economic proof). It frames risk tolerance as a key part of making that business case.
hours of service regulation
"...which someone hopefully can argue enough lawyers that that attendant is not subject to the hours of service regulation for drivers. Because if you take a class A truck... they commit themselves to the job of moving freight for 24 hours."
Hours of service rules are laws that limit how long truck drivers can work before they must rest. The goal is to prevent fatigue. The discussion is about whether an onboard safety person counts as a driver under those rules, which affects how the operation is run.
“Hours of service” (HOS) regulations limit how long commercial drivers can drive and require rest breaks to reduce fatigue-related crashes. The transcript argues that if the onboard attendant is not classified as a “driver,” the attendant might not be covered by the same HOS rules. That legal classification becomes part of the business case for autonomous trucking.
JB Hunt
"...the trucking company, the warners, whatever hunt, JB Hunt, you don't get to go home."
JB Hunt is a major U.S. trucking and logistics company often referenced in discussions about freight operations and long-haul trucking. In this segment, it’s used as an example of the kind of employer whose drivers face hours-of-service limits. That context helps listeners connect regulations to real industry operations.
truck stop
"You're stuck either in your cab or in a truck stop, or along a road. But there isn't even enough parking across the United States for these things because they must rest."
A truck stop is like a rest area built for big rigs. Drivers use it to park, eat, fuel up, and take breaks during long trips.
A truck stop is a dedicated rest and services area for commercial drivers. It typically includes parking, fueling, food, and sometimes showers, which matters because long-haul trucking requires scheduled rest periods.
truck between two white lines
"...if they're not really responsible for keeping that truck between two white lines and it works. And the only reason you really have men there is in case you pull over..."
That phrase means staying centered in your lane. The idea is that technology can help the truck not drift out of its lane, so the driver doesn’t have to constantly correct it.
“Keeping the truck between two white lines” describes lane-keeping—maintaining a vehicle’s position within marked lanes. Modern driver-assistance systems (and in some cases autonomy) use cameras and sensors to reduce lane-departure risk, which can improve safety and reduce driver workload.
triangle
"...in case you pull over and you have to take your triangle out as, as, as, as some regulations say and put a little triangle out there so people don't hit it."
That “triangle” is a warning sign you put out on the road when you’re stopped. It helps other drivers see you from far away so they can slow down and avoid you.
The “triangle” refers to a roadside warning device placed to alert other drivers when a vehicle is stopped or disabled. In trucking contexts, it’s part of safety procedures so passing traffic can slow down and avoid the stopped vehicle.
driver workload reduction via technology
"...Where's the two white lines and how the, oh my goodness, how do I not crash into something? And the technology does that for you... You don't have to be stressed every second because technology does it."
They’re saying the technology can do more of the driving for you, so the driver doesn’t have to stay tense the whole time. If the car handles key safety tasks, the driver can focus less on constant reactions.
The segment argues that automation can reduce stress and continuous attention by handling tasks like lane keeping and situational monitoring. This is often framed as lowering “cognitive load,” meaning the driver can relax because the system manages more of the driving work.
long haul trucking
"If you operated, you're beating learner. You're, you're being beating all the existing long haul trucking companies that have conventional technology. My goodness, Chris, go do it."
“Long haul trucking” refers to transporting freight over long distances, usually on highways, often with strict scheduling and cost-per-mile targets. In AI/automation discussions, long-haul is a key use case because predictable routes and high utilization can make automation economics more compelling.
Uber
"We started out talking about Uber is committing. Is that how we got there? And we ended up talking about class eight trucks on, you know, in the marketplace of, you know, 1400 mile halls or"
Uber is the ride-hailing company, but in this conversation it’s also being talked about as if it’s trying to use AI for driving-related goals. The hosts connect that to how the market valued Uber before and after major events.
Uber is discussed here in the context of autonomous-driving/AI ambitions and the company’s market valuation trajectory. For listeners, it’s a reminder that major ride-hailing companies have also tried to apply AI to broader transportation problems like freight and logistics.
1400 mile halls
"We ended up talking about class eight trucks on, you know, in the marketplace of, you know, 1400 mile halls or"
They’re talking about very long trips—around 1,400 miles. Longer routes mean the truck is working more hours, so improvements in driving efficiency or automation can add up.
The hosts reference “1400 mile” hauls to emphasize the scale of long-distance freight operations. Long distances matter because they increase total driving time, which affects how valuable automation, monitoring, and operational planning can be.
class eight trucks
"We started out talking about Uber is committing. Is that how we got there? And we ended up talking about class eight trucks on, you know, in the marketplace of, you know, 1400 mile halls"
Class 8 trucks are the biggest commercial trucks used for long-distance hauling. They’re the kind of vehicles you see moving freight across states, and they’re expensive enough that improving efficiency can matter a lot.
“Class 8” is the U.S. trucking classification for the heaviest-duty trucks—typically used for long-haul freight. These are the rigs most associated with highway logistics, where automation and AI could have big operational impact because they run high mileage and have tight cost margins.
Elaine Herzberg tragedy
"I was just thinking it's, it's just hard to believe that it's been eight years now since that Elaine Herzberg tragedy eight years already. So yeah, that was a big one."
Elaine Herzberg is mentioned as a serious real-world incident tied to self-driving vehicle testing. The hosts bring it up to mark how long it’s been since that event and to highlight how safety concerns shaped the conversation.
Elaine Herzberg is referenced as a major incident involving autonomous-vehicle testing, often used as a milestone in public discussion about self-driving safety. For listeners, it’s important context for why autonomous driving efforts face intense scrutiny around real-world risk and accountability.
on call 24 hours
"the attendant that's with the truck 24 hours today, but is only allowed to move it for 10 or 11 or whatever the number is. If he or she's not involved in there and just need to do the oversight, well, you know, I think you can be on call. A lot of people are on call."
“On call 24 hours” means someone is available all day and night if something goes wrong. For self-driving ride services, that can mean a person is ready to help or take over when the system needs it.
“On call 24 hours” describes a staffing model where personnel are available around the clock to respond to incidents, take over operations, or handle exceptions. In autonomous or semi-autonomous mobility deployments, this often maps to human oversight requirements and regulatory limits on when attendants can actively intervene.
oversight
"If he or she's not involved in there and just need to do the oversight, well, you know, I think you can be on call. A lot of people are on call."
Oversight means a person watches what the automated system is doing. If the system gets confused or unsafe, the human can step in to help.
Oversight refers to human monitoring of an automated system to ensure it operates safely and to intervene when needed. In robotaxi and autonomous trucking contexts, oversight is commonly required by regulators and is part of how companies manage edge cases, safety events, and operational constraints.
Tesla
"Moving on. And Uber is involved in this one too. Axios reports and Mobility's new big three are Tesla, Waymo and Uber."
Tesla makes cars and also works on self-driving-related technology. In this conversation, Tesla is mentioned as one of the major players in the ride-hailing/self-driving space.
Tesla is an automaker and technology company that has pursued driver-assistance and self-driving features through software updates. In mobility/robotaxi discussions, Tesla is often grouped with Waymo and Uber as a major force, even though its approach and regulatory status can differ from fully autonomous operators.
Waymo
"Moving on. And Uber is involved in this one too. Axios reports and Mobility's new big three are Tesla, Waymo and Uber. I guess you can read my comment, but I threw in there a few problem is none of those guys are dealing with people who really need a ride."
Waymo is Google’s self-driving car company. They run ride services in certain places, and the point here is that they’ve focused on particular cities/conditions instead of trying to serve every kind of trip.
Waymo is Alphabet’s autonomous driving company, best known for operating robotaxi services in select cities. The transcript frames Waymo as one of the “big three” mobility players and suggests it has already focused on a specific market (New York City) rather than serving everyone everywhere.
market supply and demand
"And once just at the beginning, all of this stuff, one doesn't really even need a really big market. I mean, for instance, the demand is so much greater than the supply. Geez, guys, get some supply out there."
Supply and demand just means there are more people who want rides than there are rides available. The point is that getting the technology right isn’t enough—you also have to have enough service available where and when people need it.
The transcript discusses supply and demand imbalance—where demand for rides exceeds available supply. This is a core business and operations concept in mobility: even if autonomous tech works, the service must still match real-world availability, coverage, and staffing/vehicle constraints to meet rider needs.
capitalism
"[1372.5s] that's capitalism. We're supposed to have lots of competitors in every market. [1377.7s] Although I also tell the students the objective of capitalism is to create a monopoly for yourself."
They’re talking about how competition between companies is supposed to benefit customers. But they’re also suggesting that competition can eventually lead to one company dominating.
In this context, the hosts are using “capitalism” as a framework for competition among companies offering transportation services. They’re contrasting the idea of many competitors with the tendency for businesses to consolidate power over time.
visualization of data
"[1454.1s] And I guess one of the things that I used to promote or used to preach was [1460.1s] was that was really interactive computer graphics because what it does is allows you to [1466.5s] learn and then probe and then search and then find and then question and critique and then [1474.4s] whatever. And this is this is the opportunity to to to learn through visualization of data."
Data visualization is the practice of turning raw information into charts or interactive views so people can understand patterns. In autonomous driving and smart mobility, visualization helps engineers and researchers inspect sensor data, model outputs, and system behavior.
household car access (one or zero cars)
"...Trenton's 61%, 60 some percent of the households have access to one or fewer, one or zero. Now, if you're one person household, you have one car..."
They’re talking about how many homes don’t have a car for every person. If a household has only one car, the other person has to use other options like walking or getting rides.
The segment discusses how many households don’t have enough vehicles for everyone to drive—citing that a large share of households have access to one or zero cars. This is a key mobility constraint that affects reliance on walking, transit, and ride-hailing services like Uber and Lyft.
Lyft
"...And if you if Uber and Lyft aren't really cheap for you, then and now..."
Lyft is like Uber—an app you use to get a ride. They’re saying it can help when you don’t have a second car, but only if it’s affordable.
Lyft is another major ride-hailing platform that can reduce the need for owning multiple cars in a household. In the segment, the hosts use Uber and Lyft together to emphasize that ride-hailing only works as a substitute if pricing is low enough.
Jalopy
"...even the the Jalopy and it's it's even worse than that because it really, it really wasn't affordable to have a Jalopy because if you had a broken tail light, then you got a ticket..."
A jalopy is basically a very old, run-down car. The point here is that even if a car is “cheap,” if it’s not safe or legal (like a broken tail light), you can end up paying fines and dealing with bigger problems.
A “jalopy” is an old, beat-up, often unreliable car that’s kept running out of necessity rather than pride. In the segment, the hosts use it to illustrate how even a cheap, broken-down car can create real costs—tickets, warrants, and legal trouble—when it’s not road-legal.
broken tail light
"...because it really, it really wasn't affordable to have a Jalopy because if you had a broken tail light, then you got a ticket..."
A tail light is the red light at the back of your car. If it’s broken, other drivers can’t see you as well, and you can get ticketed until it’s fixed.
A broken tail light is a common vehicle defect that affects visibility and safety for drivers behind you. Many jurisdictions treat it as a fix-it ticket issue, and continued noncompliance can escalate to further legal consequences, which the hosts describe.
Request an Explanation
Heard something you'd like explained? We'll add it to this episode.
Sign in to request explanations for terms you heard.
Want to learn more?
Browse our glossary for plain-English explanations of automotive terms, jargon, and concepts.
Help improve this episode
See something that's not quite right? Our annotations are AI-generated and can sometimes miss the mark. Click the flag icon on any annotation to suggest a correction.