Jeff Wishart levels with us about safety

Introduction and Welcome

Anthony: You are listening to There Auto Be A Law. The Center for Auto Safety Podcast with executive director Michael Brooks, chief engineer Fred Perkins, and hosted by me Anthony Cimino. For over 50 years, the Center for Auto Safety has worked to make cars safer.

Yeah. Fred, your hair looks great. Hey, welcome to today, Mac.

Special Guest Introduction: Jeff Wishart

Anthony: Seventh, 2025 special guest, Jeff Wishart is back and he has a long title. You ready for it? Okay, let’s go. Fellow at Science Foundation, Arizona, VP of Innovation Mobility at the Arizona Commerce Authority. Now vice chair of the SAE Onroad automated driving committee and honorary member of this podcast as he is appeared.

This is your third time, fourth time th

Jeff Wishart: well, hopefully third time’s a charm, right? Yes.

Anthony: Oh, the other two were quite charming. They’re quite good. Great. [00:01:00] Thank, welcome back to the show.

Jeff Wishart: Thank you. Happy to be here. Great to see you guys.

Anthony: Yeah.

Understanding Technology Readiness Levels (TRL)

Anthony: So one of the things we’re gonna talk about with you as. The TRL, the technology readiness levels, and I’m starting with that one.

’cause we did that podcast about a year ago, I wanna say, where we got all into full self-driving and how self-driving cars work. And we had a whole section there on levels and and how SAE does it one way and then it doesn’t make sense that way and I don’t understand it talk to us about adapting the technology readiness level framework to automated vehicle development.

Goodnight. Sounds good.

Jeff Wishart: I’ll start by saying that this is not really related to the SAJ three 16 driving automation levels. I still support those. I know you guys have some issues and some people do.

Application of TRLs in Automated Vehicle Development

Jeff Wishart: Fellow frequent guests, Phil Koopman doesn’t, isn’t enamored with them, I guess I should say, but so the technology readiness levels are basically a way, it’s a two-prong thing that is a [00:02:00] way of communicating where an automated vehicle is in its development and also as a stage gatekeeper of so you don’t progress from one stage to the next without proving some level of efficacy.

And so the TLS or technology readiness levels tls were developed by NASA for for technology. Generically, it’s not any particular technology that needs to be necessarily considered by them. So what we’ve done in this paper that we published at w S-A-E-W-C-X, which is a conference in Detroit every year.

Is make these TLS more applicable directly to automated vehicles. Basically the scale is from one to nine. So one is you’re just basically developing the concept and then by TRL nine you are you have a product that’s ready for deployment. And so you can imagine all the stages in between as you’re proving your technology and you wanna make sure that you are not.[00:03:00]

Say, going on public roads before you’ve proven some efficacy in simulation or even in a closed course. So you want to do some show, show some real capability in simulation and then you build your prototype and start testing on closed course, and then you prove that out. And then you maybe graduate to going to on-road testing, but you always have a fallback test driver.

To make sure things if something goes wrong, that you’re there to operate the vehicle safely. And then you can take that fallback test driver out before you deploy it on public roads with no fallback, with no driver at all. So it’s really a way of, again, communicating it so we’re all agreeing on where the vehicle is in its development, but also making sure that we set performance.

Levels and thresholds for, to prove that the vehicle’s ready to move from one stage to the next.

Anthony: Okay. And these are all guidelines, right? These are not laws, they’re not any legislation. This is what you want a company, ideally, [00:04:00] internally, to follow.

Jeff Wishart: Exactly. The way I see things happening is, I’m putting out research.

Like a mini consensus, a around some research and it becomes a best practice. And or maybe something like an organization like the Automated Vehicle Safety Consortium puts out a best practice. And then SAE, which is a standards development organization, takes that on and makes it into a standard, which gives it the imprimatur of.

Something with, that’s a standard, right? Which is not the same, just a higher level of, authority than a best practice. But still not a regulation. But we do want to make these standards become a regulation because even our standard doesn’t need to be followed by anybody. It’s a good idea to do it, but it’s not necessary.

So then hopefully state and federal regulators can take the, these standards and put them into regulations. So I’m hopeful that NTA will [00:05:00] adopt not necessarily whole cloth or without some modifications, but take these standards and put them into some regulations.

Anthony: Okay. Has anybody started using these?

Challenges and Future of TRLs in AV Industry

Anthony: Like we have Aurora just is saying, Hey, we have, no drivers in our semis and they’re driving between Dallas and Houston. Did they, do you know, or I don’t know if I’m putting you on the spot here, if you know that they internally went through this does, I don’t know if anybody knows.

Jeff Wishart: No, the paperwork came out in April, so it’s a bit early to to expect anyone to be following it just yet. Okay. I’m hopeful. It’d be great. It’d be great to find out if someone is, and maybe someone’s listening to the podcast who will come back and say hey, we’re already doing this. That would be great.

But I’m working with some AV developers and I’m hopeful that eventually we can get them on board. But I’m also I guess even if we skip the best practice side of things I chair the, in addition to being the vice chair of the OA committee, underneath it is there’s a, the verification and validation task force.

Fred’s a member and [00:06:00] we will be talking next week about what kind of documents we want to. Producing in the next couple years, and this is gonna be one of them that I will present and see how people feel about it. So even if we don’t get a like I, as I said, the best practice, perhaps we can jump right to make it into a standard.

SAE Standards and Safety Practices

Jeff Wishart: So I, and maybe I talked about this in the last podcast, but. There are three different types of SAE standards, each with a kind of growing level of consensus. You start with an information report, which is like a literature review. Once you have some more consensus, you have a recommended practice, and then when you have a lot of consensus, you have a standard.

So this may end up being at lower level, perhaps an information report. But it still has, again, the imprimatur of SAE, which is a well respected SDO or standards development organization in the community.

Anthony: Okay. That’s great. I think I’m, now, I’m up to speed on levels again, Fred. [00:07:00]

Jeff Wishart: Excellent.

Fred: Jeff, I wanted to thank you for putting this paper together.

I think it’s really important and useful to map things out. From my perspective, what you’ve done is. Collected a lot of loose ends and put them all together in a way that people can understand how to walk between all of these divergent or potentially divergent documents that are floating around.

The TLS for our listeners are technology readiness levels have been developed for a long time by, and are used by a lot of organizations. They’re used by military and commercial. Developers of machinery for the most part that are complex and that require a lot of not only engineering knowledge, but also a lot of inputs from the user community as well as the environment that the machines will be used in.

And the environment could be a public [00:08:00] environment, it could be a military environment. It could be a lot, but. If you dive into it and look at the TLS themselves and as part of the TRL process, you have a technology readiness assessment group, which are human beings who look at the evidence presented for the different TLS and say, okay, yeah it’s good enough to move on to the next level, or it’s good enough to move on to the, the public arena.

One. One of the things you wrote in that paper, Jeff, is that quoting from it, TRL frameworks are being widely adopted by participants in the AV industry to address the maturity level of their technology. Close quote. I’m completely unaware of that, and I was surprised to see that. Can you amplify on that a bit?

Jeff Wishart: Yeah that’s a one of my coauthors. I, as I said to today I don’t know of anyone [00:09:00] who’s using my version of tls. But as I think you pointed out early on previous podcasts the federal governments and the trust came out with a paper, I think it was the GSA that came out with a different TLS for different agencies.

And so there was a DOT version of it. So I, and there was some participants from the AV community on that group that developed the TRL specifically for the DOT. And I guess I would say with maybe it’s been more a casual, at least publicly casual adherence to these, rather than a publicly strict adherence where they’ve said, here we are at TRL.

Three, and we’re now a TRL seven.

Public Awareness and AV Testing

Jeff Wishart: So I guess I would say I would like to see people being openly and explicitly following these tls because I think part of the problem is that the public is unaware of when. What stage of [00:10:00] development a particular AV is, are they testing? Do they have fallback test drivers in them, but are they now deployed or are they still testing?

And I think that’s been part of the problem in the, that the industry has with the public and perhaps even with governments. I think sometimes governments don’t know where these vehicles are in their development as well.

Fred: Thanks. Just for our listeners. It’s actually by the GAO, the Government Accountability Office.

That’s what I meant. Produced in January, 2020. And we will have the link up on our website as well, or on the, in the notes exactly where they’re gonna be. Anthony,

Anthony: yes. They’ll be in the notes. And if you’re listening now. Just pull up your phone while you’re driving down the highway and I think Michael’s okay with this.

And just start looking and clicking on links, michael? We’re good. He’s shaking his head at me. No. If you

Michael: are in, Nevada in a level three Mercedes, yes. If you’re willing to accept liability.

Anthony: So speaking of Mercedes, ’cause what you were just talking about Jeff, there is the public doesn’t [00:11:00] know what mode these are in.

And I like what Mercedes does is doing, we’re saying, Hey, we’re in level three. We’re putting all these teal colored lights on the outside. Good idea.

Jeff Wishart: I think it’s an excellent idea. I think that they, we need to know whether the automated driving system a DS is actually on engaged and in control of the vehicle or not.

So any way that they can do that. Now the problem is it’s not standardized. Or I guess one of the problems is it’s not standardized, I may know that blue light is means that the a DS is in control. I may not. And then the next vehicle might have a red light or a green light. And how do we know which is which?

But Michael said something interesting that I, I. I wanted to touch upon briefly the, can you what secondary tasks as they’re called, can you engage in when you, these level three systems are in control of the dynamic driving task. And can you read, can you be [00:12:00] looking at your phone? Can you be sleeping?

Can you be reading? Can you be watching tv? All these

Fred: things. Lemme jump in for a second. Oh, and you said level three, we’re not talking about TRL three when Yes. Jeff just referred to level three. He meant the SAE automation levels. So a little confusing, but I wanted to make sure that we clarified that.

Jeff Wishart: Yeah, thanks, Fred. Yeah, again, my TRL the framework that I developed are not related. Not specifically. They’re different from the s se automated driving levels. So just be clear on that. Thanks, Fred. Yeah. So I, back to what I was saying. I think that all of these secondary tasks need to be enumerated explicitly and specifically for each one.

Can you engage in that task or not? Because some of these take longer to regain situational awareness. So when the, if you’re, let’s say you exit your operational design domain. And the vehicle, it says, okay, so the fallback ready user, so the other person sitting in the driver’s seat, okay, you now need to take over the dynamic driving task or [00:13:00] DDT.

You need to have situational awareness, obviously, when that occurs. So if I’m sleeping, that’s gonna take a lot longer for me to regain situational awareness than if I was just daydreaming, for example. Yeah. So I think that those different tasks need to be specifically enumerated.

Anthony: Okay. And it, that’s gotta be done per manufacturer, per their system.

This sounds insane.

Jeff Wishart: It, yeah. I, yeah, it is challenging for sure. I’ll choose that term. Term challenging is

Anthony: not the correct word we’re going for, but thanks for playing.

Jeff Wishart: I agree.

Anthony: So I’m thinking of sorry, that teal light, ’cause there’s a there’s a driving school around the corner from me and I don’t know if this is required, but they always have student driver on the car.

And and that’s the good indication to everybody. Is that a requirement, Michael? Is that some legal requirement or is that just this, the company just being like, they don’t get too mad at me.

Michael: I’m not aware of the, any legal requirements that maybe it would be a state requirement, if anything.

But I think [00:14:00] it’s mainly just, just to warn people that are around the vehicle, to be patient with the driver who, and careful around a driver who’s not quite ready yet.

Anthony: I think Uber or a Waymo should start doing that when it starts driving on the opposite side of the street. So I’m not drunk, I’m just a computer.

Oh boy. Yeah. Hey, have you have you had more experiences with Waymo out in Arizona where you are? Because the last time you were on, you said it stole your girlfriend.

Jeff Wishart: Yeah. Yeah. It, no, no more stealing girlfriends since, although she hasn’t been here since, wow. Yeah, that’s, we’re still together.

Okay. She’s in Panama. Okay.

Anthony: The Waymo took a really long turn. Exactly.

Jeff Wishart: Took her back to the airport. No I’ve ridden in a few more still the same kind of, it’s still the same kind of experiences. I think that they are, they’re they’re very good at nominal driving, I think that they are. Better in my from what I’m reading, my experiences the data show that from, for nominal driving [00:15:00] they’re quite good. They struggle, and I think you guys have talked about this, when things, when they encounter things that they haven’t seen before, they haven’t been trained for.

Yeah, they don’t extrapolate very well, but they’re very good at driving. Whereas humans tend to be pretty good at identifying ed or at dealing with edge cases. But not so good at nominal driving. So it’s it’s interesting. A mirror image of each other.

Michael: Yeah. The, I think some of the rep, the reporting this week, I know Waymo released some new safety data, and I’ve seen it say, it says that Waymo’s are better than that average driver at certain things, but also that humans are better at taking risks, which I think.

Sounds like what you’re saying.

Jeff Wishart: Yeah. In some, yeah. Yeah. So I have a report, an SE edge research report Fred’s one of is on the development team to help me as re reviewing these. And so I, this reports focusing on the safety benefits of the three what I see as the three. Types of features that are out there.

Active safety features, driver support, which is SAE level [00:16:00] one, level two, and then automated driving features. And so focusing on that one yeah, Waymo’s come up with data and I think it’s very promising. I think that, but we don’t have enough data. Even this most recent one is not enough miles driven.

The human drivers in the US accumulate more than 3 trillion miles a year. So Waymo is at, in my report at 33 million, I think it’s around 50 million for their most recent report. But Rand, which is a consulting company, did a report a few years ago that showed that you would need 275 miles million miles, I should say to be driven before the impacts on fatalities can become clearer.

So we’re still early days, but it’s very promising. But having said that, I would say that the data that we’re seeing while promising are not extrapolatable as Waymo starts to drive on highways. It drive an inclement weather and expand its ODD to different cities. [00:17:00] And it’s not extrapolatable to any other AV developer.

So if AV developer X comes in next we can’t say Waymo did really well here, so this one will also do very well. So I think there’s the, we’re still very early days understanding the impacts of these vehicles in terms of their safety or their impacts to safety. But it’s at least promising for Waymo.

Discussion on TRL Levels and Safety Cases

Fred: Hey, I wanna go back to urologist for a minute here. Because this underlies a couple of comments or a couple of questions I want to address to both Jeff and to Michael. TRL level six, the technology readiness level six as described in the GAO document has three points in it. First is the operational environment, including use of community, physical environment, and input data characteristics fully known.

Second question is, was the prototype tested in a realistic and relevant environment outside the laboratory? And [00:18:00] the third question is, does the prototype satisfy all operational requirements when confronted with realistic problems? Now, the way the TRL system works is you have to address the answer to each of the questions and get a satisfactory answer before you can claim.

Operations at that level. So if you’re not able to satisfy all of those three questions adequately at TRL level six, then you are at best at TRL level five. That’s the way it works. As far as I know, nobody has been able to satisfy all of these questions or parameters for, or I guess qualifications you’d call it for TRL level six.

And Waymo’s got well-documented problems figuring things out. The environments that’s been tested in are pretty restrictive and Tesla keeps killing people in full cell driving. So [00:19:00] I think that the claims for high TRL levels are specious at best, really hard to support. Two questions.

One for you, Jeff. First. Is, do you agree that the TRL levels for automation right now today can be claimed to be no higher than TRL five?

Jeff Wishart: Yeah. No, I’ve I guess I wouldn’t agree. Fred and I’ve heard you say this and I’m glad we’re having the opportunity to chat about it. I. So if you’ve got the benefit of having read the paper.

So you know that I’m have slightly different take on what TRL six is for avs. And so I, while I think that I’ve we do address those questions that you mentioned from the GSA. Implicitly, we don’t have them explicitly in the, in our document. So in for TL six for us is, it’s been [00:20:00] tested in a closed course, which gets to, I think, questions two and three that you mentioned, and you’ve, your results are such that are good enough such that it’s now congratulate, I’m using air quotes to go, to begin testing on road.

But prior to that, you will have been testing and simulation where you would’ve covered a lot more of your operational design domain in your testing. Now, of course, you need to have a connection between your simulation and reality. It simulation is never has the fidelity of one, but you need to know what that fidelity is.

Even close course doesn’t have a fidelity of one. They use targets that are not exactly like a human or exactly like a car. So you have to be able to. Make that connection so you, for every test you understand what the level of fidelity is. So I would say that what we’ve done is say that, okay, at tier all six, you’ve done testing in a relevant environment.

And you’ve done well enough that you can now graduate. So you is answering those questions [00:21:00] implicitly. Now, have you done every single have you covered every single thing in your ODD? No. That’s impossible to do. I would say, but you need to, we need to. That’s one of the big questions, I think in the a v community, is how much testing needs to be done, how much of your ODD needs to be covered, and we’ve got some work in this space.

That you can obviously need to do a certain amount. It has to have a certain level of complexity because if you’re just doing all driving with no other road users, for example, that’s not all that useful. So you need to do all sorts of, you have to consider both the different types of weather or all.

In your testing while, and beha as well as your behavioral competencies, sometimes people confuse those two things, but you need to be testing probing enough of your of those two subsets. What to make sure that you’re doing enough testing. But can you do all testing? [00:22:00] No that’s not feasible.

Fred: Oh, thanks. Okay. Good response. I appreciate that. Michael, a couple of questions for you. First is, who should be on the Technology readiness Assessment Team that’s evaluating? The evaluating the TRL level for each vehicle, who should be on that team? Should it include just the manufacturer?

Should it include manufacturers and academics? Like Jeff and Jeff could probably use the work, by the way. Be the public, or who

Michael: should be on that? The state police. I definitely think you’re going to need to have, oh, if you’re looking at a specific operational environment, for instance, the, say the ODD is downtown Phoenix.

I would suggest that in addition, you’re gonna have to have a manufacturer there. I. As part of that team, I would suggest that you might want someone from the federal government as well, someone from NHTSA’s Automation Safety Group or their their research group there. And then a [00:23:00] local representative would probably be nice.

Just because every specific area has its own peculiarities. And then I, I don’t know. I also to throw in, you might want an independent expert from academia or from, another AV manufacturer who doesn’t have an interest in the manufacturer that’s being evaluated to provide that perspective as well.

I, I don’t think we should. Pick random citizens off the road and thrust them into this that, it doesn’t sound like, I don’t think they would have too much fun doing it, but beyond that, that they might not be able to contribute that much. But I think you should have a range of expertise on board in making that determination.

Jeff Wishart: I. I think that’s a great point, Michael and I’ve got some thoughts on that because, so what we tried to do with this paper is and thanks Fred for describing it. We, for each TRL to satisfy the each TRL to move from one TRL to the next, you have to satisfy. We, our safety case framework, so safety case is really becoming a, I guess [00:24:00] there’s a consensus around that using safety cases is the way to go to, to ensure safety for automated vehicles.

I think the community is really rallying around this concept. And the safety case is basically saying, don’t just make a claim. You have to provide evidence for your claim. And so you can have a, you, if you, people are interested, listeners are interested, you can go to say, the Aurora Safety Case framework, where they decompose a top level claim that they’re safe enough to be on public roads and they decompose it all the way down.

And then they, for each of these claims that there’s evidence to support that claim. And so what we did is for our safety case framework, it’s three pillars. We’ve got a safety management system. So basically the organization has a culture of safety, has a way of making the organization itself values safety, prioritizes it.

We’ve got design methods. So did you do followed accepted design standards when you were developing the design of this automated vehicle? [00:25:00] And then third testing and testing gets probably most of the attention. But it’s just one of three. And so for each of these pillars, there are standards because there, as Fred knows there are.

A multitude of standards out there. And so how do you know which ones to follow, which ones you can ignore or pay lip service to? What we want to do is, okay, I’m at TRL three. The, for each of these pillars, there are these standards that you or regulations, as the case may be as standards become regulations, I want to, am I compliant with this regulation or conform it with this standard?

Compliance and Certification Challenges

Jeff Wishart: And I wanna be explicit about that, that I am. Stating unequivocally that I am compliant or conformant with those documents. And so unless I can show that I can’t move to the next PRL, but how is, how it gets. I guess audited or assessed that’s open and that’s where Michael was going.

There are companies in, if we’re in [00:26:00] Europe companies like to Rhineland who will do that evaluation because they do that they do type one certification over there. And so maybe that’s something that we can have here as well. We have approved. Evaluators of a safety case, and so that they approved evaluators of the tls.

So they are, they sign off on moving from one TRL to the next. I think that’s another option, but I think involving a swath of the public and government, I think I could see that happening as well.

Michael: And that sounds like what Edge case research, who we’ve spoken to before. Yeah. Does in the United States.

Technology Readiness Levels and Commercial Service

Michael: I’m also looking at the TRL six that the Department of Transportation has in this geo ga o document. And the first question just seems really a kind of a difficult barrier generally, is the operational environment fully known? That question is a, I don’t know how do you fully know anything, even if you’re just looking at a specific operational design domain like [00:27:00] downtown Phoenix.

I think it would be hard to be honest and say, yeah, I fully know everything Oh, before proceeding past that. I don’t wanna dip too deep into epistemology and philosophy here, but that’s a, that just seems like a stumbling block for anyone trying to get technology ready using those, the dots, readiness levels.

Fred: Yeah, I’m not, let me jump in here again ’cause I got a corollary question for you, Michael. Okay. So recognizing that TLS go from zero to nine, right? And recognizing that there’s some question about whether the current technology is a TRL five or T six or something, what do you think is the minimum technology readiness level for any given?

Automation level before that vehicle with the technology is allowed to enter commercial [00:28:00] service?

Michael: I mean it’s gotta be the, it’s gotta be past, at least what. Now it’s gotta be at nine.

Jeff Wishart: That’s what I would say as well. Yeah. That’s how we define it. TL nine is when you’re in commercial service. Now that doesn’t mean that you’re not continuing to improve, but you’ve met a bar and you could regress.

And go back to TL eight if things don’t go well, but at TL nine, yes, you’ve proven efficacy with and without a fallback test driver. And now you’ve graduated again, air quotes into TL nine where you can enter commercial service.

Rant on Developer Practices and Legislation

Fred: So this is my chance to rant because right now Uhoh, right now Corey starts

Anthony: ranting.

Go to auto safety.org and click on donate.

Fred: There’s plenty of time to do that. That’s right. So here’s the rant. Right now the developers are spending enormous amounts of money to do two things. One is to, create the illusion that commercial service is just fine. ’cause everything’s going just fine and we haven’t killed anybody yet, so why don’t we do more of it?

[00:29:00] And yet not believing that themselves really, they’re entering legislatures at the state and federal level to try to skip around this whole process by saying that, yeah, you don’t have to do anything. You are good enough. And by the way, this is a nationwide standard that says. You’re good to go.

You can do what you want with as much frequency and as many places as you want. So I think that really, if we’ve got a discipline process as you described, we’ve got the requirements that are being developed we can put together a technology readiness assessment team, including all the appropriate stakeholders and, why don’t we do that? What are the manufacturers trying to avoid? We know how to do it. We know where to do it, and we know why. What’s missing

Jeff Wishart: here? Yeah, that’s a great question, Fred. I, and I wish I could answer that.

Data Sharing and Safety Case Audits

Jeff Wishart: I [00:30:00] will say that what I’m trying to do with my work is always being aware that the.

Right now we’re not getting a lot of data from industry, right? We get the standard standing general order data, which are useful in terms of crashes, but we don’t get simulation results. We don’t get them showing that they’re conformance with one of. Standard X or standard Y. And what I’m trying to do is make it so that we don’t have to get into proprietary details, but showing that you are conformed with something and perhaps there are trusted third parties, like Edge Case research, like two.

They’re independent and they can audit a safety case. Get to us to a point where we’re sharing some of the data, but not all of the data from the AB developers. I know that’s been a sticking point for them. They don’t wanna just share everything, right? They want to know what is gonna be done with the.

With [00:31:00] what they’re sharing. And so they, and on the other side, we need to have good reasons why we’re asking for data to be shared. So we need to get to a point where here’s what we agree will be shared and here’s what we’re gonna do with it, and this will allow us to make that assessment.

Fred: Oh, thanks.

That’s a great point by the way. People do that all the time. And in my experience in the military, proprietary information is included in all the design reviews. And it’s protected quite well and it’s protected from competitors as well, but all the competitors have to reveal its proprietary details in order to go through the procurement process.

So I do think that particular point is a red herring brought up by the manufacturers to try to avoid any exposure to this disciplined qualification process that we’re talking about here.

Anthony: What if we have all the manufacturers put all this data in a signal chat. And they’ll be better. That

Fred: that’s only for higher security.

Anthony: Okay.

Michael: There actually, there [00:32:00] was actually a, I think there’s a bill pending on Capitol Hill that basically sets up a sharing program with manufacturers. But the whole point of the bill is to prevent any type of access. To that information by the public. But even beyond that, it wants to prevent NHTSA from using any data derived from that sharing process in its enforcement action.

So it’s basically saying, yeah, we know, you see there are, we know that you can see there’s a defect in our vehicle in the data, but you can’t use that to make us pull it off the road. So it’s really problematic.

Anthony: That’s just more insanity.

Software Updates and Safety Concerns

Anthony: I wanna know with the current avs that are out there, when they move from market to market, from ODD to ODD, are they using the same software?

Because the reason I ask is ’cause we found out with some Tesla whistleblowers that they’ll, they’re artificial intelligence is somebody drawing lines on a map. And being like, Hey, don’t hit this. [00:33:00] So there’s little to no proof that they’re not using anything more than just basic object detection software.

You can download all the internet. Do we know is Waymo, do they have a driver specific for Arizona and then they adjust it for, for Austin? Because I can go, I can take my Dodge Dart and it’ll drive in Cincinnati and it’ll also work fine in Miami and, it’s, it will adjust to the road conditions.

Maybe I need snow tires, but.

Jeff Wishart: Yeah, that’s a question. I I don’t think that well.

The versions of software that are being used in different deployments. There is a process that they go through when they’re planning to enter a new city. They come and they spend a lot of time first mapping with lidar to get the HD maps, high definition maps, and then they will drive around with fallback test drivers in there.

And then they will pull that out before they go into commercial service. So they, like I said they’re implicitly following the tls and [00:34:00] then they’ll enter commercial service. So there is a process that they go through. It’s not just they drop a vehicle in and go. So I imagine there are differences in the software, but the extent to which those differences are visible to us.

Yeah, that’s very minimal.

Fred: I’m thinking because then it’s more, Hey, can I jump in? Yeah, I’m sorry. I just wanted to ask Jeff this question. So Michael mentioned earlier that there was a report that says Waymo is the safest thing in the world and it’s better than sliced bread and all kinds of nice things.

In their report, they talk about the Waymo driver. As though that’s a stationary thing. And the Waymo driver has put together, 500 million, 50 million, whatever number of miles it is. But it’s my understanding that there have been many different versions of software that are included in the Waymo driver.

And wouldn’t it be important for them to say that we have allowed all of this [00:35:00] data to be accumulated? Within the Waymo driver rubric because it’s all similar enough and we’ve done regression testing and these things have changed and they haven’t done that. And Jeff, what’s your opinion about how much stability is required in the configuration of a vehicle before it can be reasonably presented as, a single driver or a single AV technology?

Yeah,

Jeff Wishart: it’s a great question, Fred, because how much, how big a change needs to occur before you, you have to say we’re back at TRL three, right? Yeah. Yeah. So at some point, if you, obviously, if you were to say, okay I’m replacing my lidar with this unit, or taking off the ladder from the top and putting it just on, on the front bumper, right?

Pretty obvious. Okay. We, that’s a major design change. And so we’d need to really go back and make sure that all of our previous results [00:36:00] are still valid, but a software change. And as I said, even at TL nine, when you’re deployed, you’re still gonna be changing, you’re still gonna be encountering new scenarios and new objects.

And so you’re always gonna be updating your software. So how, but how much of an update. Requires a regression of TRL to a previous one so that you can and at the same time, or I guess looking back at all your results and making sure that this design change means that all the other ones are still valid.

I, I, it’s a great question. I don’t think that we’ve answered that, that goes along with how much testing needs to be done, how much testing needs to be done after. A design change and of a certain size, right? It’s a, I don’t think that we know, it’s certainly not in any standard, but it needs to be there.

Fred: Thanks. This actually would include your perspective on this, Anthony. My belief is that, or my expectation is that it would be legitimate to include an [00:37:00] updated software in a version that you’re accumulating for safety statistics. If all of the software upgrades have been regression tested and passed that regression testing, and also all the new features that are in that software have been validated through discipline, test process and anything.

I think that this is, this matches your experience in the commercial software industry, right? Isn’t that more or less what they do before you go from, thought. Do two to dot three.

Anthony: Yeah. And it goes wrong all the time. All of us experience it going wrong all the time, but your phone’s not gonna kill you.

You your com it’s gonna get inconvenient. But that’s the thing. And especially with, your phone and your computer and whatnot vast majority of ’em today are not some unknowable neural network that. I know what my input is, I know what my output is, but I don’t know how it got there.

And with the claims [00:38:00] of a lot of the AV companies saying hey, we’re using artificial intelligence. First of all, define what that means. And that, that’s a, that’s, that gets into epistemology. But then it’s just seeing what is this change actually mean here? What happens? What? I think that’s from a safety point of view.

I don’t know if that’s knowable. It’s gotta be very open to the public of saying you can’t have these opaque release notes when, Hey, here’s the latest version. I don’t know if there’s some way that if I’m in my car and hey, there’s a new software update that I can be like, I, no, I don’t want that.

Because a lot of the software updates are, Hey, we’re putting advertising on the screen while you’re driving, and a whole bunch of shit you don’t want. But then there’s other stuff like, Hey, we put in security fixes here because, ’cause China is gonna hack you, or something like that. Okay, I want that too, but, oh, this change here for advertising.

All of a sudden my backup camera doesn’t work. But it’s an av. I don’t care about a backup camera anymore. Computer’s doing it. Solve the backup camera problem, [00:39:00] Michael. Done.

Jeff Wishart: I think Fred, that, you wanna make it incumbent upon the developer to show that any change that they make does hardware or software does not invalidate any of their previous test results.

And so how that’s done, how that process is done, I guess is up to the regulation, the regulator, but I think it’s a critical thing. We can’t just assume as and was describing there that the, you go from version 0.1 to 0.2 that everything is now still the same and just better. Yeah, I think we, that’s a critical part of the process that needs to be developed.

Anthony: Yeah. How do we do that? Hey, listeners, if you know how to do that you’re probably sitting on a ton of cash that you can donate to the center for Auto Safety auto safety.org. Click on donate.

Edge Research Report and Recalls

Anthony: Hey Jeff, before we jump into recalls I wanna make sure I didn’t miss this. This is the edge report or we covered this.

Jeff Wishart: That’s the Edge research report. Yes. It’s still in [00:40:00] development, hopefully published in the next few months.

Anthony: Okay, great. So this is what you and Fred are both working on.

Jeff Wishart: I’m the lead author I asked Fred belatedly, apologies, Fred, to, to join the development team. And basically these are se e publishes these reports to, discuss topical issues in automotives.

And so mine’s on the safety benefits of advanced features like active safety, driver support, and automated driving.

Anthony: Okay. All right. We’re looking forward to that. You’ll have to come back when that comes out. And mainly just explain to us what it’s like having to work with Fred that closely all the time.

Jeff Wishart: You still hear Fred it’s a joy.

Anthony: Shall we do recalls? Sure. That sounds great. Sure. Okay. First one through company called zoox 270 vehicles. This is, I like this ’cause it’s their a DS software version 24.32. And Fracted prior leases to April 17th, 2025. Zoox is Amazon’s [00:41:00] level five SAE level five card.

Has no steering wheel, no brakes. Maybe it has a big red button. I don’t know. It can’t deal with backing out of driveways when someone’s coming down the road or it can’t deal with people backing outta the driveways. I don’t know what’s going on with it. I don’t want to get into the thing.

Michael: Yeah. It looked like that the, there was a vehicle backing out of a driveway perpendicular to the Zoox vehicle that was traveling down the road.

I think the Zoox saw the vehicle slow down or stop and decided it could continue the vehicle then. Continued, and it’s not totally clear, but from the, from what they said at Nitza, that’s what I’m getting. The Zoox vehicle continued and hit the car that was backing into the road or coming out of the road from the driveway.

And at the, I believe they suspended their operations in Las Vegas and then have now resumed after updating those vehicles with a new software update. So they fixed that one [00:42:00] situation. Whether or not they have completely fixed that problem is yet to be known.

Fred: We don’t really know whether or not that was one of the design objectives.

They could be, they’re trying to swat the competition, or it could also be that Ukrainian avs drones somehow their software gets switched into the vehicle. We simply don’t know.

Jeff Wishart: Just a quick clarification at it’s level four. Level five means you, you’ve got basically no operational design domain.

You can go anywhere. Oh, okay. They clearly do have an ODD, so they’d be classified as level four.

Anthony: Thank

Jeff Wishart: you. Thank you. Level five

Michael: is also known as the unicorn.

Anthony: Exactly. I thought level five just meant no steering wheel, no pedals.

Jeff Wishart: No, you could have no steering wheel or pedals in a level four if they design it like that.

No, no one’s done that just yet. Although the cyber cab may be that if eventually because we may see that because they, that’s what they release that, I guess that two person vehicle. The, so we [00:43:00] may see vehicles like that. But it doesn’t have to be you could choose to put that in or not.

The driver controls.

Fred: And to just be mind numbingly clear, you can be an SAE automation level four while simultaneously being a technology readiness level three. Okay. Absolutely. They run in parallel. But they’re unrelated to each other.

Anthony: Yeah. Thanks. I just like the fact that zoo’s suspended operation when this happened.

That’s, no one else has done that. None. Like Waymo didn’t, hasn’t done that when they’ve had issues saying, Hey, let’s stop everything and figure out what went wrong. GM crew. Ha. What happened to them? I

Fred: think they may have seen what happened to Cruz and took a lesson from that.

Anthony: I think that’s good.

Next up, Polestar 27,816 vehicles. The 2021 to 2025 Polestar two rear view camera. Gee, [00:44:00] I, you know why? Even why, of course it’s the rear view camera.

Michael: Yeah. It’s a rear view camera. It’s also, anytime I see a recall on a rear view camera and the word infotainment comes up. I cringe because I still won’t, I know they’re not listening to me, but I still think it would be smart for manufacturers to isolate safety systems from infotainment and all the potential software problems and stuff that goes through infotainment systems because we continue to see this issue of.

Entertainment screens interfering with safety, with the safety of the backup camera. So once again, I’ll make that plea. I think that there’s an update coming. I don’t know if pole stars have over the air updates, but owner, it’s gonna be a software update. And owners are gonna hear about that from Polestar in about a month.

Anthony: Okay. Last one, Volkswagen, 5,637 vehicles. The 2025 Id buzz. Wait, really? They didn’t they just recall all them like [00:45:00] a week ago?

Michael: Oh, yeah. They’ve had a few problems. I think they’ve had a few recalls. I actually saw the first one up. The first one I’ve seen the other day in a garage in DC It was.

Pretty neat looking. Didn’t quite, it still looks like a hippie van, but it’s, it’s a neat looking van. This one’s an odd recall, so it, they built the rear bench seat. Too wide. They basically have two seat belts on it, but the federal motor vehicle Safety standard 2 0 8, and also another reg that NSA has 49 CFR part 5 7 1 point 10 requires a certain number of seat belts if you have a dictated amount of space on the seat.

And in this case, they built the seats large, and they only included two seat belts when they should have had three seat belts on that bench seat. But instead of, replacing the seat or adding a seat, belt or seat, which is, may not be all that safe and it’s probably very expensive, Volkswagen has decided instead to add some [00:46:00] padding to the rear bench seat.

That decreases the amount of space you have in those vehicles on those seats. So I’m not. Too sure how many owners are really going to want to get this recall performed. ’cause it’s basically taking cargo and person space out of their cars. But at the same time, you can’t put a passenger in those seats without them being buckled.

So it’s an interesting, it’s not really a fix so much. It is adding panels to reach a compliance level. So an odd little recall and an odd approach there from Volkswagen.

Anthony: They should just, give them a thing, a duct tape just to be on with it. That’s, that will work, right? No hey, that’s all of our recalls.

I got nothing else then. That’s it. Yeah, that’s it. Hey Jeff. Thanks for coming by again. Thanks Jeff. Always

Jeff Wishart: good to have you on. Appreciate it guys. Great to chat with you again.

Anthony: Be love it. Be.

Final Thoughts and Predictions

Anthony: Before you go, I gotta ask do you think we’ll have an a EB standard [00:47:00] before the cyber cab is released?

Jeff Wishart: We do have a regulation.

We have a final rule, right for a EB that includes a EB, which is great. It’s it might be revisited by the current administration. Yes, we’ll see for questions, but yeah that’s, I don’t have those tea leaves in front of me, so I can’t read that.

Michael: Yeah. Alright. I’m gonna say we will have more certainty on the a EB rule before we have certainty on cyber cab.

I think that’s a pretty safe prediction,

Jeff Wishart: I think. I think that’s fair. Yeah. All right. I would agree with that one.

Anthony: Alright, with that, thanks everybody for listening. We’ll be back next week. Thank

Jeff Wishart: you. Bye-bye. Bye everybody. Bye-bye. For more information, visit www.auto safety.org.