ai.bythebay.io: Josh Hartung, Bringing Automated Driving Software to the Production Vehicle
So, yeah, I came here from Portland just to speak to all you nice folks. PolySync is a company that's creating an underlying infrastructure framework for fault tolerance on the vehicle. So, this kind of starts out in the development phase of autonomous cars where you're using, you know, typically something like, you know, Wolfgang mentioned like Ross or various other frameworks. But as companies get closer to production, the need for fault tolerance, for resiliency, for diagnostics goes up. And so, that's kind of what PolySync is building is that framework. The key to that is that a lot of companies today are heavily focused on building exotic hardware to fulfill reliability requirements in the car. And so, for an analog, what this kind of looks like is building a mainframe in the car. Taking the very, the utmost of reliable hardware, making it more reliable with redundancy and things like that
And then finally deploying this like crystalline beautiful thing that will never fail and never breaks. We've seen in data, in data centers and cloud infrastructure that a more effective approach is to replace exotic hardware with exotic software. And that's kind of the direction that, that PolySync is going with our core product and some, future products. This is not a pitch thing, but I want to kind of give you a, a sense of, a sense of what our position in the industry is and why I'm a person that is talking to you today. And that'll lead, and that'll kind of lead into some insights that have been gained in our kind of unique position in the industry. So the focus here again is, is scalability. This is something that you can scale. This is a framework that you can scale from, from ADAS, from level two requirements all the way up to, you know, the mythical level five requirements
We started out with, with a variety of sensor drivers. You know, as Wolfgang mentioned, you can plug into a whole bunch of different, of different sensors with PolySync, immediately get data from them. And like, like any good framework or middleware, the data is abstracted into higher level message types that you can, that makes your software much more portable and much more resilient, things like that. We also have one of the best in the industry or the, the best in the industry record and replay system. And I would, I would say this is interesting from a, from a data collection perspective. Many of the frameworks that, that are out there used in, in R&D have issues with, with data fidelity and reproduction, especially with timing. Data ordering is really, is really important in some applications. We nail all that stuff with some proprietary timing mechanisms and a, and a distributed record model
We're also integrated with other frameworks to, to give kind of a tool chain path from, research, from research applications into mass production. So Ross is the one that's commonly used. Mathworks, Matlab Simulink is one that's maybe less common out here, but very, very prevalent in, in, in Detroit and auto, and automaker circles. And we have various applications to be able to get data in and out of different data science frameworks. So in launching that thing out to, uh, the, the automakers, what we ended up doing was working with, uh, OEMs, tier ones, tier twos, uh, other miscellaneous suppliers, startups, research institutions, academia. We worked across the industry, uh, and tried to sort of understand the different constraints that, that, that were being placed on this, our type of framework software, uh, from the very early stages of development all the way on to mass production. And consistently we were seeing issues, uh, at the very early stages of, of, of, uh, of development with getting into, you know, getting into a functioning vehicle. Uh, the, the buried entry on, on a by wire cartons actually be quite high
Uh, although George, by the way, George hots has done some really great work on reverse engineering, some production vehicles. Um, we had an internal project in reverse engineering, the Kia soul that produced a by wire car, uh, for very, for very low, low cost. So that's our open source car control project or project. Some people may have noticed that we launched it just a couple of months ago and had a bunch of interest on GitHub. It's pretty cool. What it does is it allows you to get, uh, full control of, of steering throttle and brake of a 2014 plus Kia soul can be extended to other, to other cars. We're currently looking into that, but, but you know, the key for us is getting really good on, on the Kia soul. So there you go
Controlled the, with a remote controller. Uh, and there's no, and there's no torque limitations on this. There's no, there's no, um, uh, there's no situational limitations and critically, there's no spoofing of the, of the can network. This is all sort of native control of the vehicle. Um, also you can grab the steering wheel, brake or throttle. It disables. It's all built in. Pretty cool
The way that we do it. Oh, excuse me. Cost. Other, other solutions out there tend to, tend to top a hundred K plus. And those are for the, for the folks that can afford that. We wanted to shoot for something that was more like 10 to 15 K. So that includes the vehicle. Uh, so for those of you who are looking to get into this world and, and start hacking on cars and, and, and, you know, applying, uh, AI to, to vehicles on the road, this is a really great way to get started with a vehicle you can control by wire
10 to 15 K includes the vehicle. But you've got to build it. Also, if you're going to be at the 10 K end, it's like ridiculously, well, you're going to have to like really do some sketchy stuff, you know, like salvage title cars and, and junkyard parts and things. But it can be done for that cheap. We like, we made sure that that's possible. So our approach here is, is we have, we use the, the, the, the electronics power steering of the vehicle. And uh, and we're able to, uh, we're able to essentially spoof the torque signal, which normally gives you power steering, uh, electronic power steering. We spoof that and, uh, and we're, with a, with an Arduino board and we're able to, uh, to control the steering
We have a separate module for that, separate module for, um, or actually throttle, which does about the same thing. Electronic throttle modules have been available on cars for a number of years. And essentially that's the same, that's the same situation. We spoof signals, uh, coming from a potentiometer on the, in the, um, in the gas pedal and are able to control those via, uh, via canned signals. And then finally brakes, we actually add a actuator and we take a, uh, a brake by wire actuator from a 2007, yeah, 2007, uh, four to seven Toyota Prius and you retrofit that on the car and it's a messy job. But in the end of the day, you end up with a hydraulically controlled braking system and all of that gateway through, uh, another Arduino module into the vehicle. You can get canned signals, wheel speeds, steering angles, all the kind of basic stuff you need to, to control a vehicle. So again, all linked up over can
Kind of cool. I'm rushing through this because I don't really want to pitch you guys on this, but, uh, I kind of, I came into this thinking that, that this was going to be a group of people who are probably thinking about starting startups and thinking about, uh, or thinking about moving their company in, in this direction. So here are a couple of tools that you can use to do that. This one of course is a very low buried entry, easy one to get a car on, to get a car built. Uh, of course the hardware, firmware and software is all open source. You can check it out on our GitHub and you can see OSCC.io. You can check out all the site stuff. Okay
Enough of the pitching. So in coming to this, to this conference, I, I thought a lot about what I would want to share with you that would have some value. And in doing that, I kind of, uh, I, I started to go back through what did 2016 look like. And I just, I ended up with just off the top of my head, I came up with these different videos that I grabbed from YouTube of amazing things that have happened last year. Uh, probably everyone who's been following this space can identify, you know, there's auto, comma, Waymo, uh, crew, no, cruise is there in the middle. Someone, oh, uh, Tesla on the left, drive.ai, Neo, RoboRace and Ford, uh, all making really amazing announcements and showing, you know, some incredible capability all last year. On top of that, we have Udacity developing an open source autonomous driving car and a course where people actually pay money now to become self driving car engineers. Uh, this is all very interesting
It's got, it feels like a frenzy, right? Feels very like we're, like we're being swept away by this. That's a good indicator that we're at peak height. So I, so here we are Gartner's, everyone's probably familiar with the Gartner hype cycle. You've seen it many times in many, in many articles. I actually have no idea how they measure this. So it'd be very interesting to, for anyone who could tell me that, but it rings true intuitively. So I'm going to use it as, as true in, in this example. So this is the 2016 hype cycle
And the idea here is that on the bottom axis, you have, you have these kind of phases of development of te, of various technologies. And on the curve, you have this characteristic curve that they define as, you know, the phases that these technologies go through. So 2015 was actually peak hype. I don't know if anyone knew that. 2015. This last year, we're on the kind of, the down slope there. And you can see that we're there with a nano tube electronics, software to find anything, slightly behind natural language question answering, which I'm definitely out of hype with by now. And I'm going to reality and virtual reality kind of coming out of the, out of the trough of disillusionment
So this is something to be very aware of, right? If you're thinking of, to, to, of starting, of starting a company or trying to innovate in this sector, what does the future look like? And the future probably looks something like this. It probably looks like we're going to have a harder and harder time doing things that are interesting to the, the world at large, right? We made, we made, we have a lot of work left to do, but it's probably that we're going to have a harder and harder time getting headlines. Did anybody see that Neo did the fastest autonomous driving lap around the circuit of the Americas? A couple of people saw that. Let's see. At the time that I checked this, 23,000 people had seen that. Incredibly amazing a feat, right? That was, that was just, I think it was a week ago or two weeks ago or something. It's going to be harder and harder to get, to get headlines. Roborace is another one
Roborace has been, has been steadily developing a more and more advanced autonomous driving technology. And like the big standout headline that anyone can tell me that they've heard about Roborace is that there was a dog on track and that they crashed. That was pretty interesting, right? Roborace has been a bit of a good, so it's going to be difficult to garner attention. The, you know, in some ways, I think that early stage startups that are really innovating on, on the, the self driving task may be already established. And I think that, and I think that they're, they're, one of the things I worry about for this industry is that we may be at somewhat of a 3D printer sort of moment here where, where the initial, the, the initial doing of the autonomous driving task is not particularly hard. And it is particularly impressive now. Uh, but it is the, it is in the trough of dissolution where tech, where technologies go to die, right? If they're not executed on and then, and then brought to the, uh, brought to the consumer in a meaningful way where they can start to generate revenue. And I don't think that anybody here can, can, can tell me of a, of a company that, that like nobody here can, you can't buy an autonomous car
It's not a, it's not even an industry, right? It's like, it's a, it's a, it's a science project that's being heavily invested into by companies that believe it's important for their future. But you and I can't go out today and buy that technology. So, uh, we're coming into a difficult time. And, uh, uh, I think that the, the difficult time is going to be the most important and, and simultaneously most, most difficult time to answer the question that is the thing that stops us from getting out of the trough of disillusionment. What is safe enough? I think that's the thing that we're all trying, we're all trying to answer right now. There's so many facets of this. There's not one question or excuse me. There's not one answer to this, right? NHTSA, uh, National Highway Traffic Safety Administration, released last year these guidelines for, for safety for autonomous vehicles
There's no solutions in them whatsoever. In fact, there's a lot of hard problems. There's like data sharing problems, collaboration problems. There's, uh, uh, they want you to use a common, they say a combination of simulation and real world data. Oh, and by the way, all that data should be shared among the industry. I can tell you from being deep within the industry, there are zero mechanisms for anything that looks like that. Um, so certification is, is, is a huge challenge there. Uh, there's a standard called ISO 26262
Anyone heard of that? ISO 26262 is the, is, is the functional safety standard, which is the standard by which software and hardware are deemed to be safe enough for operation, um, on, on, you know, the, the US and worldwide traffic, you know, through, through affairs. Well, these like don't even have, like you can't actually apply those standards to autonomous driving. Uh, they're, they're, they're actually specifically calibrated in, uh, for, for a world where everything that these systems do is deterministic. And where the input space is actually, is very tiny. It's like within, you know, things I can count on, you know, several people's hands and toes, right? So that's, so big challenges there. Security is one that we have, we see now increasing in, in the news and in areas of interest, but has, I, I have seen zero viable solutions there. I see a lot of people hand waving to it, but like most security things, um, it's like you should definitely not wait till the end for security. And then that's what they do every time
So, uh, there's, so yeah, there's very few, and there's very few viable solutions for anything to do with security. Testing is another big question. I, I see in almost every case where a company has gotten an, like to an advanced form of autonomous driving is they built their own simulator. In fact, they built most of their tooling, right? Um, annotation systems, uh, cloud storage systems, you know, uh, scalable training systems. All those things tend to be built in house because there is almost no ecosystem. There are all is almost no tooling for this type of activity. Where else can I go? Yeah, no tooling. Um, embedded is another, is another really weird world that when we go closer to production, where we have to, where we trying to answer what is safe enough, where we get into a really confusing thing
Because in the world of embedded, uh, again, there's no tool chains. There's very little like helper stuff. We just write code that goes on a microcontroller, not a microprocessor, different thing. And, uh, and, and, uh, you know, sort of, we're used to doing that all in house. There's not even like, there's not typically a collaborative model in, in embedded development. Um, there's, you know, there's very little processing capability. I mean, even the, the computers that are typically deployed on these cars, uh, just are completely incapable of running the algorithms that you all are so excited about designing. Now, NVIDIA is making some good headway there, but there's still issues of making that type of technology safe
Consumer acceptance is another one, right? These levels, there's always debate about the levels, right? Consumers have no idea what the levels mean. They just think autonomous car, great. And they have a, they have an issue there. And then finally, like legislation is, is one that is a systemic risk for every one of our companies, right? Because, uh, you know, for, as a, as a great for instance, GM's been running around the country right now, uh, sponsoring this save act. Has anyone heard of that? No. So the save act is, is a, essentially a bill that was passed in Michigan and then is being kind of recycled around the country. Uh, that says that you can't do autonomous testing unless you're a vehicle OEM. Uh, it's going to, it's trying to go through an organ right now
So that's why I'm, I'm aware of it. But, uh, you know, that's a, that's a huge risk to the industry. Wait, wrong button. There we go. So I think the challenge here is that Silicon Valley is actually not really well prepared to handle these type of issues, right? We're really good at being a room of amazing data scientists that make computers think and do incredibly great things. But when it comes to building, uh, an ecosystem and building like these types of tool chains in a collaborative manner, it's not like such a strong, it's not such a strong competency of, of, you know, this, this culture in this area. Move fast and break stuff is one of the most threatening and scary phrases that you could ever say to an auto OEM. Uh, on top of that, because the auto OEMs have most of the customers, it's very difficult to build a go to market strategy for your startup
So many of the startups today, like, uh, you know, the ones on, on that list, there's, like I said, there's none of them that are selling. There's no way there. It's, it's, it's, it's, uh, it's delicate and, and limited to try and go around the OEMs to go into the aftermarket, to do a dash cam, uh, to try and deal with some sort of telematics play or something like that. There's not very many of those plays because the consumers typically can't be accessed directly. You have to go through the vehicle manufacturer if you're gonna create something that's robust, that's resilient, that's, that's of any level of quality. Uh, there, uh, I mentioned that there's not much of an ecosystem there, but that's something to be, like, really aware of, which is that most of the code that goes on a car, all the car that goes, the code that goes on a car today is C. Uh, all the helper systems and high level abstractions and languages that you're used to working with, uh, don't exist when you get to production. Maybe they should, but nobody's made that safe yet
Nobody's figured out a way to make somebody agree with that's a good thing to ship on the car. And that is a very long journey. Um, and I think one of the real risks here is as we head into the trough of disillusionment, right, we have to ask ourselves, when will investors start asking us for business models, right? As those because, as, as, as it becomes less exciting, less amazing, the fact that you were able to drive a car down the road and not hit stuff. Okay, what happens next, right? Like, w, how, how many more millions of dollars do I need to pump into this thing before it gets somewhere? And how savvy are, you know, are the, are the, uh, the engineers or the, are the founders in terms of getting to market? I keep hitting the wrong button. There we go. So conversely, Detroit, and I'm going to use Detroit as the surrogate for all auto. It's not like accurate, but it's one that tends to be a, a, a, a key, a key sort of abstraction is, is also not well prepared to handle this. I've mentioned a lot of these issues
This is the V model. So, so if you're an automotive engineer, you know this by heart, and this is how you work every day. Uh, and this is how you define success in your company and your job and how you get product to market. This is, uh, this is a, a, a sort of mandated, uh, process specification from ISO 26262, the functional safety standard. And it is by definition waterfall development. It's like lock, waterfall development is like locked into these guys, uh, to, you know, uh, development processes. Uh, and very, very difficult to move away from. So the way this works is basically you can see here on the left hand side, you start in the sort of design phase, you write specifications, you, you do more design, you do even more design, and then you start building it and going back up and testing and make sure that you met your design specifications
And when you get to the other end of the V there, hey, we ship. Good job. It takes five years. Um, this is a hundred year old industry. Electronics started coming on the scene in the late seventies, early eighties. And there's a, there are a massive number of enablement companies of, uh, of proprietary tools that they use, MATLAB Simulink being, being one of the, of the most prevalent. Um, there are engineers that all they know is this. There are engineers that like the engineers that write the software, that goes on the cars today, don't code
They generate code from MATLAB Simulink. And it's all in C. So there's like a complete disconnect between what those developers know and do and what you guys know and do, right? Never the two shall meet. Uh, the, uh, the other issue, well, there's two other issues on my list here. Uh, I think culturally there's a tendency in Silicon Valley to, to renew, uh, technology on a constant basis. I think there's a, there's always, there's always a new thing, right? Like, uh, you know, it was chef and puppet and now it's like Kubernetes, right? And everyone wants to, everyone wants to have the big shape of data problems that Google has or, or, uh, you know, uh, you know, these big companies. And, and so we're always sort of embracing the newest thing. I want the new thing
I want the update. Automotive is exactly the opposite. And in fact, this can be a real challenge, you know, even for, definitely for a company like ours, they don't embrace the new thing. They don't want the new thing. The new thing is threatening and scary. And by the way, the things I have already work pretty well. And so keep that new thing away from me. So this is great if you can get in and solve specific problems
You can actually be locked into auto for a very long time. That's why most of these companies look so ancient and deprecated in a way is that they like, they don't, there's, is no motivated to change because once you have something into auto and it works well, you're there, you're locked in. Um, so the, I guess, and the other issues that I think, um, probably on the technical side, I want to mention is, uh, is that neural networks present a huge problem for, uh, for, uh, for, this whole thing. Like there is a fundamental disconnect between the way that you qualify automotive systems for safe, for, for safety today and what neural networks do, right? So neural networks are like a probabilistic statistical approach. You end up, the way that you ensure safety on cars today is a bunch of people sit in a room and they think of all the ways it could break. They write it down on a piece of paper, like a virtual piece of paper, but you get the idea, right? Like there's actually, there's actually like some very limited scope of things that, you know, that they, that they identify that could go wrong. And they have this, and because of that, they have this very limited input space based on what amounts to like analog sensors, right? Things with linear type values that change, you know, uh, incrementally. When you go over to neural networks and dealing with autonomous driving, you end up with this like fundamental disconnect, which is that you end, which is that the, these systems are probabilistic and they have a massive input space that you cannot possibly quantify on a piece of paper, right? All the way, you know, one pixel different may, may produce a different result, right? So we can't sort of enumerate all these areas
And so I, I think a very interesting area of exploration is statistical validation of safety based on, uh, based on AI systems. I think that like that's probably one of the most difficult and interesting areas, uh, moving forward is how do you prove these things are safe? There's folks insurance would be very interested in that. Um, automakers obviously would be very interested in that, but you need to have some level of, of rigor behind it. Wait, always the wrong button. Okay. So I, uh, I wanted to move on to sort of how, how do we bridge that gap? Like what can we do as, as people that want to start companies? Like this, this all sounds very grim or whatever, but hopefully I'm enumerating some problems that, that are at least interesting to some of the folks here. I put this up here because I couldn't like, I was, I was Googling graphics and I was like, okay, bridge the gap and it came up with terrible graphics. And I was like collaboration and that came up with terrible graphics
So I put a funny cartoon. Um, yeah. So for most companies, so some companies will be able to stand alone and it remains to be seen who, who those companies are and how they'll go to market and all those things. But you know, there'll be some, there'll be some of them that can survive in a, in an aftermarket capacity and a retrofit. Some of them will be acquired, but the companies that want to actually stand alone as companies within the automotive ecosystem are going to need to deal with about 25 customers. That's about all you've got, uh, of customers that are meaningful. We need to get good at collaboration, really good at collaboration, really good at partnerships. And I think for anyone who's married, what that means is that we need to get good at listening and asking questions
Um, one of the, one of the major things here, I, I mentioned repeatedly is safety. And I think there's a tendency to sort of hand wave like there is the security to get rid of, you know, we'll make it safe later. Um, and I think that that actually may be, that may be totally valid, but you can't, like art, you can't come into something and naively do without understanding the history, without understanding why the things exist the way that they do. And I think that that, like that within safety is a very important, uh, is a very important consideration, but across the automotive industry. This is a hundred year old industry. They're incredibly risk averse. They have a supply chain system, which is like, like expressed in the, in the like categories of companies. And that's all they do
Like you don't get to come in and just sell to an OEM. Right? Because that's what tier one suppliers do. And those are billion dollar companies. And that's all they'll buy from. Right? You have to, so, so you may be able to come in and innovate and change people's minds and come up with a better way to do things. Absolutely. There is opportunity for disruption. Although don't use that word
That's a scary word. Um, nobody likes to be disrupted. Um, you may be able to come in and innovate on those things. But if you come in without that risk level of respect and understanding, uh, then you will be rejected because you don't understand the world. You don't speak the language that your customers speak. Um, yeah. So we did this. So this is a lot of this is kind of part of my own personal journey and revelations in trying to build a startup in auto that doesn't just do all of it, that tries to break down the stack and do what I think is a critical component in an industry where there is no mass production or spend
And like basically the only exit is just gargantuan, insane acquisitions. I had to take the headphones off. Right? I had to stop thinking that I knew what this industry was about and where it was going and the best way to do things and start asking questions. We actually stopped and my whole team stopped pitching anybody. We were just, we, we set up, we, we talk about what we do publicly. We show people that we, we have some like chops and some credibility. We come in and we just ask questions. And what we have found from asking questions is fairly profound
Um, we, yeah, I think that the thing, the, the, the biggest takeaway that I, that I have found is that as if, if, if you talk to just people in Silicon Valley, you've just talked to these innovation labs, you know, in, in auto, you get this picture and those were, and we worked initially primarily in R and D and we got a picture, which was that these R and D labs tend to be pretty scattered. You talk to them and you go, well, these guys, you know, they don't know what they're doing. They don't have systems for this and that systems for that. And look, they haven't, you know, they have this many disengagements and they're not. Yeah. And then, then, and so you get this picture in talking to the folks in this area. And, you know, as we know that every OEM and tier one have some type of innovation lab here and they're trying to sort of collect, uh, the folks. Once we started talking to a later phase of production and in auto, these are called advanced engineering and, and production engineering
We got a very different picture, which was almost exactly the opposite, which is almost that the, the labs here that are working on autonomous driving are sort of like recruitment. And disinformation campaigns. Um, as you go closer to production engineering, we, we had this, we had this question, we had this hypothesis, which are not a hypothesis and assumption, which was initially that the, like is sort of the popular view of these, you know, the companies don't know what they're doing. And we said, like, let's go in and find out. And what we found was that predominantly these companies that you were, that are saying they will be a market and 2020 or 2018 or when it, whatever they're like extremely aggressive looking date is. And they're, they're going to ship a hundred cars out. We kept saying to ourselves, like, they don't have that technology. Like what, how, why, how are they saying that? Is that totally marketing? What is going on? We got closer to production and they, I, I think that they're not doing things the most efficient way, but they have like all the basics in place
Right? They are doing the right things. They have engineers that are pretty smart. Right? That are, that are doing, that are aware of the same problems that you all or the, you know, the vanguard of technology are aware of. And they're doing the work to solve those problems. Often it is not the most efficient. Uh, often it is not the smartest way, but it is, but they often don't care. We talked to one of the, one OEM where we were, we were, we were thinking, okay, maybe there's a, maybe there's an interesting thing to do here around, um, decontenting, uh, or, or, um, focusing data collection. So, so, um, like George was saying, they're collecting essentially just the edge cases based on disengagements
Right? So we said that would be a really good idea. You wouldn't have to collect so much data. We all know that there's a lot of data. We went and talked to these companies and we said, what if we could do this for you? And they said, they were like, what? We just want all the data anyway. Like what if we threw out something that was important? Like, and that's, that's true, but wrong, but they don't care. And we said, but you're going to have so much storage. How can you handle that? Like you can never pay for that. And they said, we don't care
We'll just convert our old factories to data centers. Uh, so, so that was a very interesting finding, which is that, which is that this is, we're at a time where automaker have the biggest war chest that they've ever had, right? They had banner sales year last year. Uh, you know, the industry is cyclical. It will go down, you know, just as the autonomous driving industry will, but they, but, but they have money and they have time, you know, basically, right? Uh, and they have space to do this work and they're going to get it done, you know, whether you help them or not. And in some cases they don't want your help, but in some cases they really need it. And so I think that my, uh, my, like, my takeaway for, for you guys as, uh, as people who want to start companies and should start companies and solve these very important and hard problems is to listen. hard. Have reasons like OSCC, uh, to go in and to ask questions and to understand the, the, the pains and the problems
They're very different for every company. I have struggled for four years of this company to try and come up with something that was uniform and like generalized in terms of the problems that every company have. Uh, they're very different. So go in and listen and then push. The industry needs you to push, but they need you to do it intelligently and, uh, uh, and with, and with respect. Okay. We have, uh, Hi. Thank you for your talk
Um, and I really wanted to touch on that last point about listening. Um, and, um, you spoke about how you heard things from these manufacturers and, and I particularly like the part about they're just going to convert all of their manufacturing plants into data centers. Um, but what are other types of questions we can be asking? Um, what other things should we be considering and taking in? I mean, obviously every company that we might be approaching is different, but like in, in terms of the space, what, what are some areas we can be thinking about pushing? Sure. Um, so, yeah, I tried to enumerate some of the areas, you know, there's, there's kind of the general areas of interest like security, safety, things like that. I mean, I think that those are all interesting, interesting things and I would encourage you, there's a lot of, of published information about like the safety stuff. For us it's been, it's been really key to go in with some type of focus and to know, uh, and to know who we're talking to, right? So my, to my point about like the Silicon Valley labs being totally different, we get, uh, amazingly to me in any given OEM there might be five, six different F like parallel efforts to build autonomous driving in different type of capacities with titles that are totally meet like senseless and meaningless. That like some person feels like they're like a director, but this person is actually their boss and, and this person who like just graduated from Udacity is now in charge of this thing. And like, so, so there's all these different efforts
So, um, I think the, one of the points that I didn't make, uh, is, is start focused on something like some, some area because you can't ask questions of these people that are outside of their, outside of their like normal, like scope. But they will answer them and they'll answer them and they'll answer them wrong. So it's been a really big, uh, thing for us to actually map these organizations. We, and we do this, like we put it down on a board. We like this person reports to this person, works with these people. Like, uh, this is in an R and D capacity. This is in a production engineering capacity. Where, where does each of these people sit and who's the best one to ask these questions of? Cause I, I think it's less of a like, let's send out a survey and get, and get a big, and get a big cross section of, uh, vaguely useful answers
But one answer from one, from the right person who's like mostly in charge of the thing that you care about can inform everything. Yeah. So what questions? I don't know. You know, uh, you've got to find those out for yourself. Hi, uh, sorry I missed a little bit of your presentation. So I hope you didn't cover this in that part. So you started your startup, try to focus on everything and decided to concentrate on something. So are you now a services company or what does your startup actually do? Uh, sure
So we talk, I talked about that early, early on, but tried to kind of get through it to get to the, the meat of my opining. Hopefully. Uh, uh, so, so we build an, uh, under the, uh, underlying platform and if you're familiar with something like Ross, um, then it's like production grade Ross. And the intent, the intent of it is to sort of help manufacturers address, uh, address reliability and safety issues on the system level by replacing exotic hardware like, you know, highly rated, you know, automotive grade systems with flexible software, uh, and distributed system type resiliency. Yeah. Good. Hi. So it sounds like you're kind of suggesting that Bay Area companies basically become tier three, tier two suppliers, but there is no, uh, history of the Bay Area doing that type of thing
And the cultures would match up. Do you see that actually as a viable option or do you see a split and, uh, there being like different efforts to do different things or something? Oh, that's a, that's a hard thing to answer. I don't know. I mean, that's, that's the crazy thing about this industry is like there is no go to market, but the go to market changes every three months. So, uh, can Bay Area companies or any, uh, or any small startups actually become direct suppliers? I, I don't know. Like I'm hoping we're collecting information on that right now, but, uh, but I think that the, that's like still very much to be determined. We don't really have much of a precedent of a standalone company, uh, um, going to market with, you know, with various, various automakers without being acquired first. The one standalone or the one, the one exception to that, which I would mention, I think is Mobileye, right? Mobileye has done a great job of going to market and, and maintaining their sort of value, uh, value segment as, as they grow up
And I think that like hopefully they'll be able to continue that. Um, and, uh, but you know, I know there are a lot of people gunning for them. So how, how I, you know, I don't know, uh, you know, start with solving the right problem. Yeah, over here. Um, in your conversations with auto makers, do you get the sense that any of them are thinking, uh, these legacy systems are so aft that we should just start over? Um, so in any automaker, there's a lot of schools of thought. Um, inevitably the momentum is, is like just do more of what we do, but like do it harder. Uh, oftentimes what we're seeing is that, is that they're, uh, the automakers are pushing reliability concerns out to their, to their suppliers. So one of our, one of the, uh, tier ones that we work with, we were talking to about like the types of RFQs, requests for quotes, that they're getting
And you know, a year ago it was for like level two capable systems. Now everything is level four. And we asked them level, level four being, you know, being a complete redundancy, you know, uh, fail, fail, fail operational. They call it meaning something in break and the thing still works, um, at 100%. So that's been a, that's been a very interesting change. So I think that's indicative of like things going like in the same direction that like in, in a refinement type direction. I think the challenge is when you go to level four, uh, there's just no, I mean, there's just no, uh, there's no existing systems that run this type of software at that level of, of performance that, uh, you know, that like there's, that's just not deployed anywhere. And so lots of companies are thinking that they can do it
One example of a disruptive technology is, is, um, is Audi Z fast familiar with that. It's like a four chip computer. There's one, um, SOC. There's like a FPGA, there's a mobile eye and there's a something else. A still D, uh, controller. So that's kind of an interesting hybrid architecture. But I, our belief is that they're eventually going to get to a point where they go, well, crap. Like this thing's just not, not gonna, not gonna go there
Um, you know, a great, a great example I like to use is, is, you know, today I think Ford runs six Xeons. Um, like they're public on that. There's like six computers stacked up in the back of their car. Uh, a couple of GPUs. You know, how do you make a 16 core Xeon ASIL D compliant? Like usually that's done with like a multi, like a lock step, uh, synchronization and, and full redundancy and things like that. I mean, that's, I think that that's like a $12 billion fab to build that chip. I don't see, I don't, I just don't see that type of thing like going directly into production. So I, I think that disruptive architectures are probably the only way and there may be starting to awaken to that
There's always like one or two people that are really into it. Now, is that a good answer? Okay. All right. That's it for this talk. Thanks Josh. Appreciate it. Thank you.