Urban Innovate TALKS

Henriette Cornet (00:05.317)
I'm shy.

Henriette Cornet (00:11.907)
Hi William, nice to have you with us today.

William Levassor (00:15.886)
Thank you and very happy to be here with you.

Henriette Cornet (00:20.311)
Great, yeah we have a bit of a history of podcast recording together, it's our second episode, let's say, and we were discussing before that it has been a while already, so things have evolved a lot in the world of autonomous vehicles, exactly, but it's a pleasure to have you back and for the listeners who don't know you yet, I wanted to invite you to have your view as an operator.

William Levassor (00:26.795)
It's easy actually, yeah.

William Levassor (00:35.958)
Yeah.

Henriette Cornet (00:47.991)
and you'll tell us more about your experience in public transit, etc. But really, for me, you are the specialist experts on really fleet, fleet operation, fleet supervising, monitoring, all these words that we are hearing all the time and there are confusion sometimes. Mostly we will tackle the idea of safety, monitoring safety, demonstrating safety maybe as well, and also monitoring performance.

So that's kind of the agenda for today and yeah, I'm really looking forward to our discussion. But before jumping in all these topics, would you mind introducing yourself and tell more about what you are working on?

William Levassor (01:29.262)
Yeah, very quickly. So I have experience in the automotive industry. I worked in Renault and Stellantis in my early career and then moved to Transdev, a worldwide big public transport operator back in 2016 when AVs were kind of the shiny thing that was going to disrupt mobility.

Today, I'm at BETI, a small regional operator at the beginning, but we turned the strategy to really be in the middle between the AV providers and the local ecosystems, operators and authorities. What I'm looking forward to is to digitalize daily operations. What I saw in my previous...

career and all the years in the AV field that you have AV providers they are really good at creating and designing self-driving software globally and then you have local operators that know how to manage drivers and vehicles but they are non-tech industries and know service companies they need to digitalize how to

use how to integrate those vehicles. So kind of connecting the two worlds and make them work together to disrupt mobility and improve shared mobility in the future. That's what drives me on a daily basis. This is...

Henriette Cornet (03:06.617)
Yeah, that's your big topic, the shared mobility. You are very much in the shuttle area, right?

William Levassor (03:13.772)
Yeah, mobility is our thing. We're trying to improve mobility and use that particular technology to improve shared mobility.

Henriette Cornet (03:25.177)
And with shared mobility, do you mean like shared rights or shared assets? Like is a robot taxi a shared mobility for you?

William Levassor (03:34.944)
anything that can improve mobility and use less vehicles to move people in cities and outside of cities.

Henriette Cornet (03:41.925)
Yeah, that's a good definition. Yeah. Cool.

Henriette Cornet (03:53.317)
So when you say digitalization of the services, what can you dig already into that? What you mean with that? And mostly what strikes me when you say that is, isn't it digitalized already or what's missing or how is it done today? Or what's the part really that you think that needs more of this digitalization?

William Levassor (04:12.898)
Yeah, in fact when you look or when you go inside a local operators, it's a lot of human tasks. You have people charging the vehicle, you have people driving, you have people reporting what they did, are they late or okay for the missions, they are reporting the issues, they are reporting what's wrong with the vehicles, etc. So it's a very human like job.

When you integrate self-driving vehicles inside those fleets, you don't have the driver or you don't have this feedback to manage your service and your performance on a daily basis. So you need to have some kind of data from the vehicles, from the field, to manage your service and to improve it, just to manage it, to meet your contract with the authorities.

Henriette Cornet (05:10.981)
Okay, I see.

William Levassor (05:11.576)
So that's why it's basically what we're providing today at BT is field applications for the operators to guide them through the vehicle preparations to the declaration of what's wrong during the services. And then after the service dashboard to see a standard data to assess the service, the security, the safety, et cetera.

and a platform for the control centers. I won't talk about like you said supervision, hypervision, remote, everything you can say but control centers. Okay let's say like it's for metros, buses, planes etc. You have control centers. We need them and we're building that platform to gather all the functionalities the local operators needs to manage their service.

not only the driving part and supporting the driving part because we hear a lot about supporting the driving sometimes remotely and first thing on my side it's completely okay i see a lot wow they are remotely overseeing the vehicles yes it's normal we have control centers for driven buses

So yes, it's completely normal to have control centers for automated vehicles. yeah, for metros, for tramways, for buses, for planes, etc. So it's normal. What we're building is this platform to manage every aspect of your service. The driving, but also the journey of the passengers, the field and the alerts from the system.

Henriette Cornet (06:39.919)
planes as you said.

William Levassor (07:01.814)
So the four big aspects of the service has to be managed at any time during a service.

Henriette Cornet (07:08.709)
Maybe we can dig a little bit more into that. Wait, I'm a little break. Your gaze is really up. I wonder if your camera can move where you're looking. I don't know how it's going to like when I look at it like this. Yes, it's much, much better. I think at the beginning it's okay. I'm focused because you're looking at the screen. It's much better.

William Levassor (07:21.194)
You're I'll put you

Is it better I put you just above the camera. Yeah,

Henriette Cornet (07:37.411)
Okay, so I'm going my transition again.

William Levassor (07:39.63)
All right.

Henriette Cornet (07:42.117)
Okay, let's dig a little bit more into all of that or like let's decompose the different things that you said and you said a big word of data. I think that's central here. So I'm sure everybody, like people who are listening to this episode now are familiar with the data and like the AV type of data that we hear being the most important one is the number of accidents, the incidents. have been some happening with autonomous vehicles.

William Levassor (07:47.061)
Yeah.

Henriette Cornet (08:12.129)
Some have led to companies stopping their operation entirely. Some have led to a pet being killed. So many things like that that will make the headlines in the press and affect also the public discourse around them. We hear the topic of disengagement. So that's also maybe something we should talk about more. What exactly is meant with that and why is important? But the way I see it is we have all this data. I mean, we could collect so many

so many data around the vehicles and anything that's happening. But the big question for me is why? What for? What do we want to do with this data? And mostly, and also who wants the data? For what, so to say. Both goes together. Because my view, can already, and I want to hear really your view on all that and which type of data do you think we should focus on. Sometimes I have the impression in all the discussions ongoing is,

the regulators will need data to give approval for a service to be deployed either on the vehicle side or on the commercial deployments, like it is done in California, but in other states, and in Europe it's kind of similar, the separation between vehicles and service, to give the permission. But then it goes so far that for me it looks like they are acting as operators almost.

William Levassor (09:16.984)
Mm-hmm.

Henriette Cornet (09:41.221)
trying to supervise the fleet, asking for more and more more data, and it's a good example. And I want us to dig into this document, the document from SFCTA, where they propose a conceptual framework for safety, et cetera, where it goes so far, then it touches an area where I'm like, should this data be public, actually? Is it too much detail? Would it also slow down?

the willingness of companies to deploy in certain parts of the world if these type of requirements are so high, and then talk to you about the role of an operator, the role of a third party kind of player that could play something a bit more transitional. And I'm sure here what will play a big, role is the fact that in US we have these market-driven fleets, like Waymo Zoox.

William Levassor (10:11.086)
Mm. Mm.

Henriette Cornet (10:34.765)
And we hear every day about new cities where these vehicles will be deployed, where no public transit agencies is involved, the cities are not involved. It's really purely market driven and like Uber like service. Whereas in Europe, the shuttle services that are happening, super interesting, are more embedded into transit service where there, the operators of kind of a more natural role. But you see where I'm going. So who should collect the data? What for?

William Levassor (10:50.848)
Exactly.

William Levassor (11:01.79)
I see.

Henriette Cornet (11:04.375)
And so it's huge. maybe tell me your thoughts and we try to deconstruct all that.

William Levassor (11:06.702)
Yeah, each topic I wrote down like four or five different things where we can deep dive. Maybe the first thing you said, maybe it will slow down things if we ask for data or too much data. And I want to emphasize what you say, the difference of approach between the US and China.

and Europe on the other side. Without telling is it good or bad on one side, I'm not taking any part of that, or maybe later, but first there's really a different approach there. In the US companies, like Waymo, etc., they have their own safety demonstrations and they go on open road.

when they think it's safe enough in Europe.

and it's not linked to AVs, it's exactly the same for driven vehicles. When you put a product on the market, you have to homologate it beforehand. So before you put an AV on the open road on commercial services, you will need to homologate and so demonstrate the safety concept, demonstrate that it's safe enough to be put on open road, that it's as safe as a human driven vehicle.

It's not only the design of the vehicle but you have to illustrate and convince that your safety management system is consistent enough to be safe and to have people on board. So it's really different.

William Levassor (12:56.47)
All these questions, all the data, you have to provide them if you want to have a commercial service in Europe. That's the European approach. Basically, the philosophy is to protect consumers in Europe. We have those regulations beforehand. But yes, on the vehicle side,

Henriette Cornet (13:13.541)
it's really on the vehicle side, The homologation type approval is really on the vehicle. It doesn't say much about the service.

William Levassor (13:20.214)
Actually, have, and it's really step by step, you have a European type approval that homologates the vehicle.

So the platform and the autonomous driving system, and then deployment certification by your third party. So you'll have to declare where if the ODD matches the generic ODD of the vehicle that is homologated. Exactly like you deploy a tramway or a bus, you have to declare where it will operate. So there's really two different steps, generic vehicle homologation and then deployment certification.

Thanks.

Henriette Cornet (14:00.249)
But if I may pause you here, you know, in the US, the unique thing is self-certification. But the construct is kind of the same. They also have the vehicle standards that they have to apply for the vehicles. And now it's in a gray area where these standards don't exist yet for AVs, same as in Europe. It's like progressing, but it's not absolutely completely there yet.

William Levassor (14:13.422)
Yeah.

Henriette Cornet (14:26.917)
They have all these standards for the vehicle and then at state level they will approve their commercial deployment. So I see a lot of similarities in the approach but funny enough the type of services that are deployed are very different.

William Levassor (14:33.206)
And... yeah, and this...

William Levassor (14:40.76)
Very different, yeah, you're right. We see less robotaxi and more shared mobility public transportation in Europe, where you mainly have shuttles running. That may be historic because we had shuttle providers in Europe. But also because the market for public transportation is higher.

in Europe, you look at the TNC, the taxi market in the US and the public transit market in the US.

I don't have the exact figures in mind, but the taxi market is really higher in Europe. It's kind of the other way around. the public transit market in Europe is really high compared to the taxi in Europe and compared to public transportation in the US. market, it's, I'll say it's market driven. It's what we are, what we have on each side.

Henriette Cornet (15:49.605)
And back to the data when we are considering these different models and we can focus first on the motor shuttle with the operator being an integrator so to say of AV fleets and connection with public transit authorities. Maybe we start there. So tell us a bit your view on how...

William Levassor (15:54.531)
Yep.

William Levassor (15:58.179)
Yeah.

William Levassor (16:04.888)
Yeah.

Yeah.

Henriette Cornet (16:15.461)
differently.

Henriette Cornet (16:21.145)
Tell us a bit your view on the data that is needed in your opinion to assess safety, not from a regulation perspective. Maybe we can let that a little bit apart, but more like on a daily basis from an operator. How do you still make sure that the rides are safe for the riders? And how do you measure, I think that's also so important, the topic of performance, like quality of service in general, and make sure that...

there is a benefit to use this technology after all. So how do you distinguish both? What's your view on that? And can you share? know, really don't hesitate to be detailed about kind of the routine to be put in place and what type of data exists already, which data are harder to get. But like, please share a bit on the shuttle, really, on your experience.

William Levassor (16:52.802)
them.

William Levassor (17:01.101)
Okay.

William Levassor (17:12.365)
Yeah.

William Levassor (17:16.064)
operations. So first I would distinguish three kinds of data or three kinds of contexts where we need data. The first one that is quite obvious is when there is an accident.

So there you have to share a lot all the context to understand and to make sure that you put something in place so that it doesn't happen again. So you'll see four accidents. Then second context for managing for the operator to manage the service as I said. So here you will need a lot of

real-time data that you don't need to process a lot or extrapolate but just to manage your service.

William Levassor (18:15.404)
So I don't like to say real time it's at any time because as you're not driving remotely, you don't need to have data like five milliseconds. It can be longer. The latency can be longer. But you need, for example, the video stream for the passengers if they.

If there's something happening in the vehicle, need to react and they can call the control centers. you can exactly like in the elevator or in the metro, you also need the health status from the vehicles and from the control centers. For example, tire is...

William Levassor (19:02.786)
You also need the data from the health status of the vehicles. For example, you have an issue with the doors that cannot close or you have the battery that is caught in fire or you have the tires that is off. You need to have this kind of information to react and to...

to what needs to be done to restore the service and reassure the passengers. You need to have the data from where the vehicle is, the data of what's the state of charge of the vehicles to be able to dispatch it to the next mission, to the next line, or to say, okay, go back to the depot so we can integrate another vehicles. And we need to have the...

the feedback or the request from the driving stack if it needs some help from the remote operators. So that's the second context to daily operate a service at any time. The third context and the one that is really interesting for me is what data you need constantly to assess, monitor and improve your service.

We talked about regulations and safety. In safety, when you design a system, the safety lies in your safety management system. It lies in three different pillars. The design of the system, so you design your vehicle to be safe. Then you have the operating procedures.

You know in the aircraft industry that the instructions for the pilots are also very important for the safety. And the third pillar for the safety management system is the monitoring and continuous improvement. You need to have the best safety possible to

William Levassor (21:13.464)
constantly monitor and improve the system. So for that, and that's what I telling, I'm providing these kind of tools with the providers we work with. We have this generic dashboard to monitor first the global...

global view of the service mainly for insurance company, for local authorities. It's really an overview and basically it serves to measure the exposition of risk.

and split the liabilities. So we'll have to know the number of kilometers we did, the average vehicle speed, the number of passengers we moved, and who was driving. Is it on auto mode or manual mode?

and to split liability you have to track the human interventions. It's very easy for an AV provider to say, there were an intervention from a human, it's not my fault, it's your fault. If you don't have the data, you're completely blind and you don't know, you depend completely on the tech providers fault for your liability. So you have a

lot at stake here to have a clear transparent data to know and to be able on both sides to know who was driving and who had intervened in the service. So let's see the first part.

Henriette Cornet (23:09.903)
So are you saying that in Europe, when you apply for getting a permit for commercial deployment or just a trial even, is the operator who will apply and take the responsibility of what's happening? Or is it still the ADS provider?

William Levassor (23:29.006)
Today, you have two different frameworks for regulation, at least in France, or in Europe. You have the first framework for experimentation. In experimentation, you need to have an authorization, but the authorization holder takes the responsibility. So when I have an authorization,

It's my job to make sure that the technology I use and how I use it is safe enough. It's not the job of the regulators or authorities or third parties. It's only on me. For commercial services...

That's another story and as I said you need to homologate the vehicle, the system with the control centers and the deployment. So with commercial services that's another story and maybe all of this data and dashboard as we say will be more standard in the market. But for the moment I am requesting this kind of data and trying to monitor this data.

So it's really on the initiative of the operator and the authorization holder to do that or just drive and see what happens. That's it. So the first part of this continuous monitoring and improvement, the first part is so illustrating the exposition in kilometers and passengers and trying to be clear, transparent on the liability.

The second part would be to illustrate the performance. And here the metrics can be very simple. What I use is the exact same metric that we use in public transport operations with driven vehicles. know punctuality, availability, etc. So the exact same metric.

William Levassor (25:39.235)
than driven services. The third part of the dashboard that I'm focusing on to constantly monitor the service and the system is on security and safety. And it gets

interesting here. So it's basically all the emergency stop, all the takeovers, all the overrides, etc. And as you said, it is sometimes very difficult to get those kind of data.

Henriette Cornet (26:14.629)
Yeah, because it's saying a lot about the manufacturer, mean the ADS provider, etc.

William Levassor (26:20.94)
Yeah, it says a lot. It's funny when you say that. I saw a communication, a hearing of one big AV providers in the last weeks and they publicly said I won't release this kind of data because it will show my competitors my performance. From my point of view, it's not showing how you do things.

It's not a secret IP, it's just a way to measure the performance. so the clients, operators and local ecosystems can know with full transparency if they can use it at larger scale or not. And it's just transparency. And okay, you can have different performances between competitors and with any product on the market.

Usually you can see the performance and some trains go faster than other trains but they feel the user experience is better or the feeling of safety is better or anything. So from my point of view it should be transparent and that's something that I don't Yeah, after that I don't display those kind of things publicly but...

Henriette Cornet (27:33.209)
publicly available.

William Levassor (27:42.465)
as an operator and as the integrator of the system we need that to know where we can push. stops, human interventions. And the last thing I have in this dashboard is the human feeling. Today we have safety operators on board.

and they can declare if the feeling is not safe. Because for me it's very... yeah... for example the vehicle doesn't yield as expected or the trajectory feels unsafe. So there is no accident, there is no incident, but you know the feeling is not human-like or not safe enough and it's...

Henriette Cornet (28:16.837)
With feeling you mean like perception

William Levassor (28:40.792)
For me it's very important because one funny story when I started to track those interventions I won't say what company but one company

were filtering the interventions and they told me no no we run it into simulations and if you operators didn't push the emergency stop there wouldn't have been any accident so we don't count it as interventions. If the operator feels he needs to stop the vehicles it means it's not ready enough

Henriette Cornet (29:15.385)
problematic.

William Levassor (29:23.806)
to go. Maybe it would be safe in the sense that there is no collisions but that's really and that's you know the angle you take when you develop a software for driving you're looking for avoiding collisions when you're looking at an operational perspective you're looking at managing the security of your passengers

So not putting your passengers in dangerous situations. It can be from the software, can be from the vehicle, it can be from outside, it can be from other road users or other passengers. We are responsible of the passengers, not of the software that is driving. So it's different angles.

Henriette Cornet (30:07.001)
Mm-hmm, I see.

Henriette Cornet (30:11.557)
Wait.

at my notes.

William Levassor (30:26.776)
data concetva.

Henriette Cornet (30:28.067)
Bye bye.

William Levassor (30:30.638)
is the end the board

Henriette Cornet (30:38.149)
think we can move on that.

Henriette Cornet (30:47.077)
So maybe it's a good time actually after everything you said to dig more into what kind of models exist. And as I mentioned, this document from SFCTA, it's San Francisco County Transportation Authority that have produced a document called Conceptual Framework for AV Safety, something like that. I'll put the link in the description of the episode.

And they propose a way AV should be regulated and they are actually struggling in San Francisco. And San Francisco is really interesting use case for AVs because a lot started there and still some operations are starting there. But also there is still a lot of friction with the city because there is complaints that the vehicles are stopping, are blocking lanes, are blocking emergency vehicles, or that the...

the interaction with first responders is not smooth enough, et cetera, et cetera. So there have been multiple cases that have created safety issue. it's again a little bit, I mean, when it is blocking an emergency vehicles, it's quite clear, but sometimes it's a lot about feeling what you said before of like, why is the vehicle stop here? And people may take then an action to overtake the vehicle because they don't understand how it behaves, things like that.

William Levassor (31:59.182)
Yeah.

Hmm, yum.

Henriette Cornet (32:11.523)
And this has been really pointed out by SFCTA among others. They have proposed this framework, I find super interesting. But for me, I'm wondering, know, like the big issue they have is that they don't have a regulatory power over autonomous vehicles in San Francisco. It's dealt only at the California, at the state level.

by the DMV and CPUC, the DMV is the department motor vehicles that really deal with the vehicle itself and the CPUC is dealing with the commercial services. And they have already kind of levels to be reached to be able to get more and more permits linked with the ODD, linked with the performance of the vehicles and super important, the vehicle miles driven.

And I know you like this topic because we discussed already that it may not be a proof for safety that you have driven a lot of miles. And I'd like to hear your view on that. But back to SFCTA, they propose kind of this framework as a way to get more power and what they would like to see more from the vehicles. But again, I'm struggling seeing that being applicable and is there any risk that the city will get overwhelmed by

having to deal with all this operation and imagine there will be not only Waymo and Zoox but like more and more players coming, how will they deal with that? So that's where I think kind of an independent assessor as a NHSTA is calling it here in the US or an operator could be useful.

William Levassor (33:38.179)
Yeah.

Henriette Cornet (33:58.032)
So yeah, I'd love to hear your view particularly on the roles that you see from your point of view and your European experience of the roles that could play cities and if it may not be also too much for them. And the other part, sorry, I had it. I had it just now and I lost it again. and the other part.

William Levassor (34:19.352)
Yeah.

Henriette Cornet (34:22.627)
What's your view on these kind of progressive levels, like building level up? And I like to compare that. It's almost like a video game. You have to go. You need some points to be able to go until you reach the boss level. So maybe that's a comparison that can be made. And I know you had some thoughts on that. so both aspects, the role of the city or this kind of level approach to gain trust at the end. It's about trust.

William Levassor (34:28.238)
Exactly.

William Levassor (34:42.198)
it.

Henriette Cornet (34:51.063)
and demonstrating safety. wherever you want to start, it's a lot and it's a bit different kind of approach.

William Levassor (34:55.948)
Yeah, so maybe first by the level approach, as you know, I love it and I use it. That's how I go with the AV providers I work with in the last 10 years. I think that's very smart. And that's the way to first get to know the.

Henriette Cornet (35:02.2)
Mm-hmm.

William Levassor (35:19.704)
the tech providers you work with and go step by step and not do only paperwork at the beginning. Because yes, I'm a paper guy and we haven't seen that yet, but when I see a vehicle...

driving, an AV driving, I hear a lot of people say you see it can drive so it works. No, doesn't mean it works, since it can drive a few minutes for a drive. It's fun, it's extraordinary, but it doesn't mean it's safe enough, it doesn't mean it can scale, and I have a lot of examples with that. So often my first questions in my first weeks with an AV provider is only papers.

Henriette Cornet (36:04.677)
Mm-hmm.

William Levassor (36:05.132)
to assess safety, is it safe enough? And then we go step by step because it's important to go on the field and as we said earlier to continuously monitor to see if we can go further. So yes, I think the video games approach and to follow different steps or levels is very smart and you have specific metrics that you can look at to validate the next phase. For example, I have a check

list with different steps for example can we go on open road can we take the safety driver

back from the steering wheel just with an emergency stop, can we put passengers on board, can we remove the safety operator from the vehicle. So these are different steps, big steps. Then you can imagine steps like growing the ODD or putting standing passengers in the vehicles. And for each...

Henriette Cornet (37:13.701)
But then how do you, who decides on where to put the threshold? know, like, how is it visible after how many miles, after which events or non-events, probably, do you say, okay, it's enough and let's move to the next phase? It's kind of tricky.

William Levassor (37:32.297)
It's very tricky and by experience it's very different when it's the authority or the operator or the tech provider that is speaking. Very different. Sometimes I heard yeah it's L4 you can drive without anyone on board and when I look at it, no let's

do it step by step and that links with the second question you have what is the authority we have and the opportunity we have to have data and to challenge the providers in Europe as you said we are organized with

PTAs, the local authorities that have a contract with the PTO that operates and take provider, an AV provider. So the AV provider can sell its vehicle and autonomous driving license only if the operator is willing to use the vehicles in its contract. So we have kind of natural authority on the provider and we can challenge them.

It's still not easy, for example, because we are at the beginning of things and if you are a small operator with like hundreds of vehicles in your network, you have less authority than a big operator with hundreds of thousands of vehicles if they're going to scale or...

If you're Uber and you can buy thousands of vehicles, there you have an authority and power on the the AV provider. So it's kind of difficult and

William Levassor (39:23.838)
Win-win discussion with the AV provider. Most of the time it's difficult at the beginning but then they understand it's a win-win and this is how you build trust with the local ecosystem by having a local, as you say, third party that is monitoring and assessing the deployment and the service to make sure it's well integrated in the local ecosystem.

using integrated and not only safe for the driving part but well integrated in local ecosystem. to answer, Phase's approach is great. For example, we can...

we can put standing passengers in the vehicle. That's an example. I think you don't have it in the US. It's not a question in the US because you're more focused on robotaxi. here where we are in the public transportation business, we have buses. So the dream is at the end to have people standing to put more passengers in the vehicles. If we have a lot of emergency stops and

heartbreaking we cannot do that so that's one thing I monitor do we have strong breaks and what's the frequency what's the occurrence the occurrence rate of these heartbreaks so that's a metric I look at before we can we can do that so a lot of metrics for each phases and

Jobs under IPT is a science IPT outfit.

Henriette Cornet (41:11.961)
ticket from there. But if people are interested in this metrics, can they contact you? You can share with them your experience on that. So like if anybody wants to know more, but is it publicly available? Is it like what, how do they do that? And the back of my question is, I think we should not reduce us only to robot access. It's, it happens a lot when I, when I am contacted by my European fellow fellows.

There are many transit agencies in the US that would really benefit from the technology. Mostly as a feeder, first and last mile services, for paratransit, for microtransit. are many, many use cases that could be a big contribution to AVs and to at least explore the technology.

William Levassor (41:44.974)
Yeah.

Henriette Cornet (42:04.869)
The message I want to do is like, look at it, look at what it can do for you. Imagine you are a transit agency in the US, you own your fleet, your bus fleet, you own it, you are operating it, so you are both kind of an agency and an operator at the same time. You are curious about the technology. How would you do, how would you start and would you need this third party integrator or how, yeah, because the risk I see is if they contact directly AV providers,

they may also get some kind of deployment but may not match exactly their expectations.

William Levassor (42:41.528)
Yeah, they can try to ask for as a condition, ask for specific reporting, dashboard and metrics to follow. Maybe not publicly, but at least at the local authority level to look to. But as I said, having a third party is kind of the...

confidence part and know...

Henriette Cornet (43:10.916)
Mm-hmm.

Henriette Cornet (43:15.127)
It's a trusted partner.

William Levassor (43:17.068)
Yeah, a trusted partner that is outside, know, of the tech providers that has all these push from the commercial side, the stakeholder part, they need to say they can scale and they need to say they can drive without purchasing, but they need... So there are a lot of that are very serious, are, you know, top performances, but still...

Henriette Cornet (43:30.595)
and blisters.

William Levassor (43:45.935)
they have this push internally. When you have a third party, it can build more trust and I think it's, as I said, a win-win also for tech providers because if it's a third party that is pushing their technology, it's always better for everyone.

Henriette Cornet (44:09.605)
Yeah, and for competition as well among themselves.

William Levassor (44:13.388)
Yeah, you can compare, it allows comparison with the exact same metrics. That's something that I really wanted.

Because in my experience it's impossible to compare different operations, different cities, different services or different vehicles. When you have those kind of metrics you can compare everything and know challenge and improve. The goal is not to challenge to say you're good or you're bad, it's to globally improve. There are a lot of industries where again performances and safety are shared.

Maybe not with the whole public, but with authorities it is shared to be able to feed a common good to improve. Because as you saw then, you said at the beginning, if you have an accident, it will be very bad for, obviously, the company that had the accident, but for the whole ecosystem. So feeding this...

Henriette Cornet (45:02.149)
Come on,

William Levassor (45:18.63)
A improvement at least on safety, I think will be, that's the beginning of the industry, and that's what is also fascinating, we are at the beginning of the AVs, maybe in 10 years time, all of this safety, you know, in the railway industry and in the aircraft industries, all the safety events, even if it's not accident, they are shared, they are shared.

if you have an incident with an aircraft, it's shared. So they are all improving and avoiding accidents worldwide.

Henriette Cornet (46:00.527)
But so when you speak so much about metrics, what I hear a lot is a need of standardization, I suppose, would be something you would value and encourage.

William Levassor (46:12.194)
Yeah, that's at least at my level. That's what I initiated and pushed. So for every operations or every technology we deploy for our operations or for other operators, we provide a dashboard with the exact same metrics in any city or any technology.

Henriette Cornet (46:40.383)
That could definitely help.

I'm if we have... Otherwise we're approaching the end, I think. I'm going to look if there's another one. And tell me if there's something you haven't said yet.

William Levassor (46:57.806)
Oh yeah, I forgot to the time to block the road. I put in the different levels. I took the example of the standing, but there was this that was interesting. I have in my dashboard too.

Henriette Cornet (47:06.606)
tu voulais reparler de ça?

William Levassor (47:23.755)
It's not very practical.

Henriette Cornet (47:23.823)
You want to like that, cold, and I'll try to put in...

William Levassor (47:28.302)
We can try a sentence like that and you'll if you do or not.

Henriette Cornet (47:34.125)
I think I've answered all my questions.

Henriette Cornet (47:53.425)
Okay, come to the room.

William Levassor (48:02.968)
So again on these levels

For example, have these steps where you want to remove the operator from the steering wheel. You will look precisely how many takeovers you had and clearly you want zero takeovers in the vehicles. If you want people standing in the vehicles, because that's what we are looking for in the public transportation industry here in Europe, because we also want to address more people in the vehicles like Bossez. What the metric I'm looking at is the hardware

What's the occurrence of hard breaks? Because obviously if you are constantly hard breaking, people will fall in the vehicles and that's not what we're looking for. If you want to extend the ODD of your service, one of the metrics I'm looking at is how long does the vehicle's

stay stuck in the middle of the road if you get stuck and you have exemplarizing in San Francisco that's one of the things you talked about so the worst case is you're blocking emergency vehicles but even without emergency vehicles if you get stuck two minutes in the middle of the road you're blocking traffic and interesting stories when you block the traffic is one of the metrics in the

transportation contracts in driven services. So nothing to do with AVs. If you block the traffic, if your vehicle has an impact on the traffic and block the traffic, that's a metric we have from the PTAs in public transportation industry. that's also one metric we're looking for AVs because that's important again to build trust. Again there is no collision. So from a

William Levassor (50:02.128)
safety of the driving part it's safe because there is no collision but from the operational perspective from the security of the operations that's important because you will put your passengers in dangerous situations because as you say people will try to overpass the vehicles you can have accidents then that is not your vehicles but you will have accidents if you get blocked too much

Passengers will try to get out of the vehicles and it's even more dangerous Remember in the metro When the metro is blocked you hear an audio message Please stay in the metro and don't get out on the roads because it's dangerous You can have vehicles passing in both ways. So stay in the vehicles. So again, it puts passengers into dangerous situations

Henriette Cornet (50:34.853)
Mm-hmm.

Henriette Cornet (50:42.233)
Mm-hmm. I'll try to.

Henriette Cornet (50:59.417)
Yeah, now I can see that definitely and how this type of metrics is necessary. Still, I'm curious to see how it will evolve in the US and having kind of a sorting between the information that should be publicly available for the trust that we mentioned already for this feeling. you know, a big part of my work these days is on safety perception and actually safety perception

is one of the arguments for choosing between one mode or the other. Taking the bus at night in certain areas, sometimes some people will say, no, I don't feel safe. I prefer to take a cab or a Waymo because I feel more safe doing so. It's just an example, but safety perception has really an impact on how people move around. So I get it that authorities

William Levassor (51:50.166)
Yeah, it does. Yeah.

Henriette Cornet (51:56.351)
feel like they have to take care of kind of the social good and of what that people feel feel safe the way their cities organize etc but still how much data should be shared it's it's really like i feel there's still some granularity and some work to be done based on everything you say by the way and i think you are helping really making that a bit clearer that there could be these layers of a of third party that can

William Levassor (51:58.959)
Mm-hmm. Yeah.

William Levassor (52:22.638)
Manage the data and simplify the complexity of the operation and of the technologies, simplify this complexity to a standardized dashboard for cities.

Henriette Cornet (52:24.421)
Exactly, work more on it and manage. Yeah. Yeah.

Henriette Cornet (52:35.885)
Yeah, exactly. Yeah, fantastic. I think that's a good way to also advertise your work and what you're working on. And I would advise people to reach out to you directly if they want to do so and dig more into this topic. As a last, very last question to end this episode, you know, I'm also working on the quarterly update on the different AV services around the world. And I wanted to ask you what are the ones that...

you find interesting these days, which are the ones that you are following up more closely than others? Are there any AV deployments that you're kind of intrigued by? yeah, so is there anything worldwide worth watching more closely?

William Levassor (53:09.422)
Thanks.

William Levassor (53:18.702)
Obviously, looking at

how Waymo manage the scale and the number of different cities and how do they approach the European approach. I saw that they chose to attack through UK and not directly European Union and I see that's not for nothing. So very curious to see how they will go there and how they will tackle everything

Henriette Cornet (53:36.762)
Mm-hmm.

Henriette Cornet (53:44.218)
Mm-hmm.

Henriette Cornet (53:51.597)
enters the market in Europe.

William Levassor (53:53.791)
that we discussed and also curious about what will happen in Europe I will say to answer to your questions because so first there is Waymo and then we have also the Chinese actors coming Europe and it feels like you know the American actors at home in the US, in California, the Chinese are at home in China

and maybe Europe will be the battlefield to define the state of the art of operations, maybe not driving technology, at least how do we operate, how do we build that trust, because it's more in the culture. So maybe that will be...

Henriette Cornet (54:31.769)
The host.

Henriette Cornet (54:43.131)
Interesting, I like it.

Henriette Cornet (54:48.986)
battlefield back to the video game comparison.

William Levassor (54:50.132)
Yeah, I don't like the pattern of your word, but you know where everything will melt and the different cultural approach will merge and I think that's very interesting to be there and to see all of these different actors.

trying to come and all saying I'm the best in the world, don't look at what I do, it will be fantastic, it will be L4 without operators on board tomorrow and you know seeing that, yeah the reality is far different but it's very interesting to be there obviously and to see everything.

Henriette Cornet (55:18.713)
The reality is different.

Henriette Cornet (55:26.181)
Cool, cool, yeah. So we'll need more episodes to monitor, using the right word, and which deployments will come and what's come next in Europe. And in the US, I really hope we're gonna see more shuttles. That's my hope, that there will be more. And even if it's, know Waymo has these new vehicles and some other providers are experimenting different form factors.

William Levassor (55:31.502)
Yeah.

William Levassor (55:42.349)
Yeah.

Henriette Cornet (55:51.933)
more shared mobility would be great and more transit agencies like public authorities being willing to try things that's would be yes yes

William Levassor (55:52.195)
Yeah.

William Levassor (55:59.183)
to integrate this tech because as I said in the introductions for me and that's what drives me. That's a fantastic opportunity to improve shared mobility and public transportation. It's been a long time, decades and decades, it hasn't been improved a lot and it needs to be improved to compete with just driving your car.

Henriette Cornet (56:12.196)
Yep.

Henriette Cornet (56:20.48)
yeah.

Henriette Cornet (56:25.465)
Yeah, absolutely. We shared a view on that, I know that.

William Levassor (56:27.918)
So PT needs self-driving technology and self-driving technology also needs optimization.

Henriette Cornet (56:34.957)
And starting being a competitor to cars, as you say, to be more and more aggressive in how to get more modal share of how people travel around. Yeah, that's my wish too. Yeah.

William Levassor (56:45.696)
and mobility solutions for people also outside of cities. In cities you have a lot of solutions but outside of cities you need solutions, we need solutions, people need solutions. So that will help.

Henriette Cornet (56:58.541)
Yeah, let's maybe take that for another episode. The rural mobility US has so much of these cases where there is no infrastructure as well for transit. So we'll need something else. So let's take that for the next episode. Thanks a lot, William. It was great. And I really advise everybody to reach out to you and get to know better about your work and just your mindset. I think it's very insightful.

for this type of discussion. So I wish you luck in everything you'll be developing and we keep in touch definitely. Thanks a lot.

William Levassor (57:32.578)
Thank you.

Henriette Cornet (57:36.677)
Cut! la queue! Ça a été? Ouais, on va voir, on va voir. L'outil est assez chouette, il fait des transcripts automatiques, on va voir s'il a réussi à nous comprendre. Parfois j'ai démi.

William Levassor (57:49.268)
Yeah. Yeah. Well, it's not bad anyway because we managed to break stuff. Data, what I was it's not necessarily the sexiest but it's... No, I think it's... We managed to get stuff out.

Henriette Cornet (58:05.669)
Yeah, after I could very well, when I show the thing, can already show, when I acronyms, usually when I talk with acronyms, I the acronyms on the screen. If you have really simple to show I can also do, you know, a screen that interposes and shows something. That, it doesn't I'll maybe at it, listen to it. And when you know, when we talk about levels, etc., it could maybe help to have two or three things.

William Levassor (58:12.556)
Hi,

William Levassor (58:20.62)
William Levassor (58:25.422)
OI

Henriette Cornet (58:35.255)
I think if I stop, it will stop the recording without turning off the computer.

William Levassor (58:39.304)