view episode transcript
Essentially when you are talking about level four, level five autonomy, especially in urban kind of environments, you are looking at the problem of what is called artificial general intelligence. That is not only perception, but also cognition. It is not a solved problem. Artificial general intelligence has been an open research problem for the past 70 years.
Jay Clouse 0:21
The startup investment landscape is changing, and world class companies are being built outside of Silicon Valley. We find them, talk with them and discuss the upside of investing in them. Welcome to Upside.
Eric Hornung 0:49
Hello, hello, hello, and welcome to the Upside podcast, the first podcast finding upside outside of Silicon Valley. I’m Eric Hornhung, and I’m accompanied by my co-host, Mr. Rendez-Vous himself, Jay Clouse. Jay, how’s it going, man?
Jay Clouse 1:02
Great. I’m very full.
Eric Hornung 1:04
Still?
Jay Clouse 1:04
Very full still. I just told you that I’m not even hungry for lunch right now. We are here at day two officially of CES 2020. And yesterday after, Eric and I got done recording, I lost you, bud.
Eric Hornung 1:17
I know it is a very sad moment. I was walking around. We were at the automotive floor. There were trucks everywhere. And we just got separated,
Jay Clouse 1:26
Got separated, I put out a desperation tweet to say, hey, lost Eric at CES 2020. He’s probably scared and alone as he is when he’s by himself. I thought he might be sustaining himself on impossible burgers or sliders. Then, lo and behold, much to our, I don’t know, excitement, glee…
Eric Hornung 1:46
Pleasure, curiosity.
Jay Clouse 1:48
Much to our glee and curiosity, I get a tweet back from Here Technologies, which is a location mapping company that apparently has a lot of corporate backing. They said, What you guys need is a rendezvous point. Go to top of the world at the stratosphere at 730, and not only will you be reunited, but you’ll have a four course meal with tasting menu. Just the magic of CES, Eric.
Eric Hornung 2:12
This place is crazy, man. But we had a free dinner. We got free wine. We got a free view.
Jay Clouse 2:17
Yep.
Eric Hornung 2:18
And, man, I don’t know how to cot this segway.
Jay Clouse 2:22
Well, thank you to Here technologies. Smart cities and transportation seems to be a theme of interest here at CES 2020, and today we are talking with a founder of an autonomous company. His name is Takin Meriçli. He is the cofounder and CTO of Locomation. Locomation develops safe and reliable autonomous trucking solutions. It was founded by five autonomy experts from Carnegie Mellon University. They’re building autonomous trucking, trains, platoons…
Eric Hornung 2:54
Convoys.
Jay Clouse 2:55
Convoys. We’ll explain a little bit more about that in a second. But Eric, we haven’t really talked to anyone in autonomous on the show yet.
Eric Hornung 3:01
I think it’s really hard to find pre-series A autonomous companies outside of the valley.
Jay Clouse 3:06
I agree.
Eric Hornung 3:06
So it’s it’s really a supply issue more than demand issue because you and I have talked about autonomous on the podcast before, but just in vague generalities, never a deep dive.
Jay Clouse 3:18
Locomtion has raised five and a half million dollars to date from investors including Plug and Play, our friends at Firebrand Ventures and Homebrew. We’ve had a little bit of a trend of trucking on the show, but there’s definitely going to be a different look at trucking and automation. And that’s exactly why we come to places like CES to get a look at that.
Eric Hornung 3:34
So if you guys have any thoughts on autonomous trucking after this interview, you can reach out to us @upsidefm on Twitter or if you have something a little longer send us a note at hello@upside.fm. Jay, come with me on a quest. Come with me on an adventure. Are you in?
Jay Clouse 3:51
I’m in.
Eric Hornung 3:52
All right. I want you to start a high growth startup right now. What’s its name?
Jay Clouse 3:57
The Haberdashery.
Eric Hornung 3:58
The haberdashery. It’s a very Traditional startup name. Google. Zynga. The Haberdashery. All right
Jay Clouse 4:06
Tip of the hat to that name.
Eric Hornung 4:07
That’s a, yeah, that’s that’s a nice one. So let’s say that you go out and you want to hire some great engineering talent to get this thing off the ground. What are you going to do? Where you gonna go?
Jay Clouse 4:17
Well, I would probably first go to my own network and realize quickly that I don’t know enough engineers.
Eric Hornung 4:22
Right, and with a name like the haberdasher, you’re gonna have a tough time recruiting.
Jay Clouse 4:25
Tough time. Tough time. I might need some outside help, Eric.
Eric Hornung 4:28
Yeah, I think if I were you, I would go to our friends over at Integrity Power Search. I mean, they’re the number one, full stack, high growth startup recruiting firm between the coasts. They partner with venture capitalists, private equity groups, and hypothetical CEOs like you, Jay, to build amazing teams for the world’s most disruptive companies. Since 2012, they’ve successfully executed over 600 searches. So that sounds like you’re starting engineers, Jay, we could get those. All right, you can get those with IPS. And they are on track for over 200 in 2019. Their clients have collectively raised over 2.5 billion with a B in venture funding, and, hey, maybe with the haberdashery, will be counting.
Jay Clouse 5:08
If they can help The Haberdashery, they can help you. Learn more about Integrity, Power Search upside.fm/integrity, and they may just be able to help you find your first engineers.
Jay Clouse 5:26
Tekin, welcome to the show. Welcome to CES 2020.
Tekin Meriçli 5:28
Thank you very much. Thanks for having me.
Eric Hornung 5:30
On upside, we like to start with a background of the guests. So can you tell us about the history of Tekin?
Tekin Meriçli 5:35
Sure, I mean, it’s actually a long history. I’ll try to kind of sum it up a little bit. So my background is actually in computer science and engineering. Maybe I should even start with my birthplace. So I’m originally from Turkey. My hometown is Istanbul. At a very early age, I actually got the opportunity to meet computers. So I had some computers in my life and started programming at a very early age. And that led to pursuing a degree in computer science. And throughout my studies, I got fascinated by AI and robotics. So that’s how I started studying AI, because when you’re doing computer programming things are mostly abstract. But when you’re programming a robot, you actually start seeing things moving in the physical world, which is even more exciting. So that’s why I decided to actually go into robotics and particularly intelligent robotics, not necessarily manipulate their arms and such working at factories. Then I progressed in my academic career. Back then, I had a dream of becoming a professor, actually even like studying at Carnegie Mellon or becoming a faculty at Carnegie Mellon, which I eventually achieved. My masters and, and PhD, I concentrated on intelligent robotics applications, worked on a multitude of projects. And eventually I ended up at Carnegie Mellon as a postdoc, and then senior engineer, and eventually special faculty. So my last installation was the national robotics Engineering Center. It’s a semi-auto autonomous r&d division under Carnegie Mellon robotics Institute. So we build a lot of autonomous systems there, including autonomous vehicles intended for various applications ranging from military context to, to industrial context, agriculture, mining, etc, etc. So, I was also very lucky, I’m privileged to have a lot of brilliant people around me, I was part of the autonomy group there. Some of those people loved getting together to start a Locomation, my latest startup. And we are focusing on building semi-automated semi-trucks in Locomation. So that’s kind of the quick introduction to myself.
Jay Clouse 7:37
I want to go back to picking up computers in Istanbul. How is the tech community in Istanbul? Is that something that’s popular there? Or how did you even come across this is an opportunity
Tekin Meriçli 7:46
Actually, it’s a boutique kind of thing. Maybe I should say back in the day, it used to be, especially like pre-internet era. It wasn’t like…You could actually count people. Okay, like these are the computer people, a very, very small community. I guess nowadays with the democratization of computers and, and smart devices, etc, etc., probably that’s seen grow a little more, but there’s still like room to improve and grow.
Jay Clouse 8:12
So how did you come across computers in the first place? Was that in your family or..?
Tekin Meriçli 8:17
Yeah, that was actually a very happy coincidence. So maybe I should also step back and say that my brother, who is also our CEO with the current company, Locomation, has been my role model and has been the trailblazer, if you will. So as for the computer, it was a very happy coincidence. My uncle, my dad’s brother, is a very enthusiastic guy in terms of technology. So he likes trying gadgets as they come out. And back in the day in the 80s, this little computer, Sinclair Spectrum, came out and apparently he purchased one. But his passion for computers was not very long lasted, so he used it for a while, and then his kind of excitement died off. And he decided to actually give it to us, to my brother, because my brother was also reading about computers and such and magazines and newspapers and getting excited about, like, having a computer. And my uncle said, You know, I have one sitting, collecting dust; why don’t you actually pick it up and start using it? And that’s when I actually got the chance to touch the computer keyboard. We had a few games and such back in the day we used, like cassettes, magnetic tapes, essentially. We started playing those games and such. But over time, they started appearing as boring things, like I needed to be doing something more interesting with those computer. So how can we actually start programming them? And that process actually gradually evolved into like, how can you program computers to be intelligent? How can we actually make them not give us canned responses but have their own minds and like, perceive the world and, I don’t know, interpret the commands and all that stuff. So that’s how we kind of progressed into AI and eventually robotics and intelligent robotics.
Eric Hornung 10:01
When you’re kind of starting one of these autonomous projects, how, how much of it is already, like built in is foundational, and how much of it like has to be built from scratch?
Tekin Meriçli 10:11
It actually depends a lot on the project itself. It is very difficult to put things under a single umbrella, if you will. I mean, if you if you’re looking at it from a software development perspective, yes, you can actually divide it into layers. There is, I don’t know, the operating system there, there is what is called the middleware layer. And on top of that, there is the application layer. So you may be able to reuse some of those–of course, like operating system is there, so it’s reused. Even the middleware nowadays, for instance, if you are doing any kind of robotics applications, there is this popular open source middleware called Robot Operating System, ROS. So that is mostly reused and it provides reusable components. And on top of that, you build your application layer. But when it comes to application layer, it is very difficult to define a single umbrella even though you could, from us systems engineering perspective, you could kind of define your building blocks as, okay, for instance, if I’m building an autonomous car, it has to perceive the environment, so there has to be a perception module, it has to be able to I don’t know, localize, know where it is in the environment, that has to be a localization position estimation module. And it has to be able to plan routes from point A to point B, so that has to be a planning module, etc, etc, and controls, so user interface. So there are those building blocks as well, maybe even that kind of abstract concept could be reused. But the details, that bits and pieces will be different from a particular application to another one.
Jay Clouse 11:36
When did you start thinking about autonomous vehicles in particular?
Tekin Meriçli 11:40
Oh, let me see my journey in autonomous vehicles actually goes back to the DARPA challenge days. So we were talking about those things back in like 2005, even prior to 2005, and that’s when the first DARPA challenge came out.
Jay Clouse 11:57
What is that?
Tekin Meriçli 11:58
So, DARPA be Military kind of agency, they like pushing the limits. So, in that particular challenge, the goal was to have a driverless vehicle drive 460 miles in a desert environment with no human intervention. So whoever could actually drive that route fastest and safest would win the challenge. So it started that way. Two years after that, in 2007, DARPA, this time said, you know, you guys seem to have solved this desert driving problem. But what if you make it more interesting and define it as a, an autonomous driving in urban environments kind of problem. And then they announced the DARPA Urban Challenge, the famous DARPA Urban Challenge, in which the aim was for the vehicles to actually obey the traffic laws and deal with pedestrians, deal with other traffic elements of their vehicles, etc, etc, and drive in a urban traffic environment essentially, simulated on, of course, in real world. As a matter of fact, Carnegie Ellen’s team was the winner of that challenge. And that pretty much sparked all the autonomous vehicle development that we are seeing nowadays. So pretty much you can trace any autonomous vehicle developer, major autonomous vehicle developer team, back to Urban Challenge, and even back to Carnegie Mellon and, and NREC. All of those folks have roots in that challenge. So that’s whenI got involved. Back then I was not at Carnegie Mellon, I was at UT Austin doing my masters. But we also had a DARPA Urban Challenge team. I was also coteaching and assisting a class on autonomous driving, actually. So we had a team of undergrad students as well as some grad students and some postdocs and such focusing on that problem. So we are doing project management, coding implementation. Back then ROS did not exist, for instance; we had to write device drivers, implement device drivers ourselves, we had to write all those building blocks I mentioned a few minutes ago ourselves. So there was no like plug and play kind of setup or open source stuff back then. But yeah, so that was my entry to autonomous vehicles more than 10 years now, close to 15 years.
Jay Clouse 14:07
When DARPA puts out a challenge like this, how long do the teams have to build a solution?
Tekin Meriçli 14:13
Not too long, actually, tops like two years, even under that. You have to be very knowledgeable to be able to put something together in that kind of a short period of time.
Jay Clouse 14:23
Yeah.
Eric Hornung 14:24
Before we get into Locomation, we hear so much about Carnegie Mellon with autonomous vehicles and Pittsburgh being a hub, and you hear about professors getting scooped up by Uber and whatever else is happening out there. Can you just talk to, like, what the culture of autonomous vehicles in Pittsburgh is like right now and what’s happening at Carnegie Mellon?
Tekin Meriçli 14:45
I mean, Carnegie Mellon has been the pioneer in intelligent robotics, autonomous vehicles, and computer science in general, like one of the top schools in the nation and in the globe as well. So autonomous driving at Carnegie Mellon is not something new. It goes back literally decades, in the 80s. Carnegie Mellon demonstrated hands-free of cross country autonomous driving. Back then that project was based on a very simple neural network essentially detecting the lane markings and associating the lane markings positions with steering commands.
Eric Hornung 15:19
To a non-technical person, very simple neural network is a oxymoron.
Tekin Meriçli 15:26
Maybe I should say this. And nowadays, the hype, one of the hype items is deep networks, deep neural networks, deep learning etc, which is composed of millions of neurons, a lot of layers, tens of layers, etc. By simple neural network, I’m, what I mean is a lot less number of layers, a lot less number of neurons to process that information, and the image resolution was much smaller that is being fed into that neural network. And it was almost like learning from demonstration kind of application. So a human driver was driving during the training period and that neural net work was being trained on whatever the camera sees and whatever commands the driver, the human driver, actually provides to the vehicle. So and then the neural network in the background through the, throughout the training session is figuring out the the correlation between the two. And then you will just kind of, invert it and then provide whatever input comes into the neural network as the output to the vehicle to be able to command and keep it in its lane. So literally, like 30 years ago that was demonstrated. It was much much earlier than, I don’t know, way more Uber whatever, all those other companies came up, and in multiple domains, again. I guess I briefly touched on that in my introduction, so this particular example was an on road driving example. But Carnegie Mellon, through its Robotics Institute and National Robotics Engineering Center, has been doing a lot of projects with the military and the industry in multiple application domains: single vehicle, multi vehicle, convoys to be deployed in like unmaintained road, kind of environments, or stone mines, in rural Australia, or underground operations, or agricultural fields, whatever you can imagine. So we have done it all. We’ve seen it also, we know the challenges in autonomy. That kind of culture and that kind of practice actually gave us the confidence slash, I’ll have to say, kind of caution to define our problem very well and make sure that we are not immediately shooting for the holy grail, if you will. And then it comes to our implementation at Locomation, because this problem is tough.
Eric Hornung 17:35
So let’s talk about Locomation. How did that idea and narrowly defined problem come about? And how did it kind of grip you and become something so exciting that you were willing to leave and go private sector on it?
Tekin Meriçli 17:46
Yeah, so we’ve been talking about and building autonomous vehicles for a while now. We’ve implemented similar technologies for heavy vehicles as well in different domains. We are constantly, or we have been, I should say, constantly looking at commercial opportunities as well. Back when we decided to form Locomation, the ridesharing space was already too crowded, it would be hard to differentiate ourselves. Besides the technology part, it’s still actually very difficult because it still requires some scientific breakthroughs. So, whatever all of those companies are chasing after is not a solved problem, it cannot be solved with today’s and known scientific techniques. So, essentially, when you are talking about level four, level five autonomy, especially in urban kind of environments, you are looking at the problem of what is called artificial general intelligence. That is not only perception, but also cognition. It is not a solved problem. Artificial general intelligence has been an open research problem for the past 70 years. When humans are driving, they are doing a lot of magic when it comes to processing the scene and attaching some semantic labels to objects and predicting their behaviors that’s…
Jay Clouse 18:57
Like this is a child, or this is a ball, or this is a…
Tekin Meriçli 19:00
It is very, It is very easy to actually fool machines. So if you guys are familiar with the latest like deep learning techniques and such, there is this concept…
Jay Clouse 19:09
Which we’re not.
Tekin Meriçli 19:10
Okay, maybe I should say, again, it is essentially given an image, what are the objects in this image, a stop sign, I don’t know, lane and car, human, etc, etc. But the machine actually does not look at it from a holistic perspective, it’s essentially like looking at all those pixels, those pixel values, and trying to make sense of those things and attach labels to that. But it can be easily fooled in that sense. You can change a few pixels here and there, and all of a sudden it is recognizing a stop sign as, I don’t know, an exercise ball with 99% confidence. Or even worse, sometimes, with those kind of fooling techniques, you can make a machine recognize a stop sign as a speed limit sign, like 60 miles an hour. Imagine that. I mean, instead of stopping, you’re actually zooming through an intersection. So that’s really dangerous, right? But as I said, it’s not just like assigning values to pixels, it’s also the cognition stuff. We are, as humans, when we are driving, we are paying attention to a lot of different cues. So for instance, when you see a car in distance with a dent on the side of it, you all of a sudden attach a probability of accident, high profile of accident, right? So you kind of try to stay away from that vehicle. Or I don’t know, when you look through your mirror and see the car behind, the driver of the car behind you looking at his phone, you often say okay, this guy is not paying attention to the traffic, so I, it’s very likely that he is going to do something stupid, so I should better stay. So we are still not there when it comes to machine perception and cognition. All of it is, at the bottom of it is actually what is called uncertainty. So uncertainty in general is the killer of intelligence robotics, if you will, when you have zero uncertain, let’s say factory automation problem, right? So everything is hundred percent under control. You have robots, let’s say, working at Amazon warehouses, they have their pads, even lines on the ground or barcodes or whatever, there is nothing jumping in front of them, there’s nothing trying to fool them. So with that kind of setup, they work perfectly. And hundreds and thousands of them are working 24/7 without any kind of fault. But when it comes to autonomous driving, we are dealing with this hybrid environment. There are some robot vehicles, and there are some human driven vehicles, and there are some other actors again, pedestrians, cyclists, etc, etc, that we have no control over. So we, there is a huge amount of uncertainty in the world. So how do you deal with that uncertainty is the key question. When you hear people talking about this like a very long tail problem, that long tail is this, like dealing with the uncertainty part. So going back to the Locomation story and how we decided to actually pursue trucking and, and particularly autonomous comboing, if you’ve looked at all those problems, okay, there’s still like some scientific breakthroughs needed to solve for level four, five urban driving kind of scenario. We’ve built similar systems in the past, so we know how difficult they are. So maybe I should also make the distinction between demonstrating autonomy and turning it into a product. Anyone can demonstrate autonomy and give it a ride in a very controlled environment. But when it comes to turning that into a product that has 99.9 more 9 reliability, it is a whole different game. Right? So taking all those kind of factors into account, we said, okay, so there is this technology perspective of it, it still requires some scientific breakthroughs. So it’s not a good idea to bet our business on some research to be made and done. And there’s the business aspect of it as well. So looking at the trucking industry, it is really a pain point nowadays. So trucking industry is facing driver shortages. So as of today, it is about a hundred thousand drivers short. And the demand is constantly increasing. Everyone is purchasing stuff on Amazon, and everyone wants those purchases to be delivered within a day. So who’s going to deliver that, right? And also a combination of the two, from the technical and the business aspects of it, most of the operation, tracking operation takes place on highways and interstates. And highways and interstates are relatively simpler to handle compared to really chaotic urban environments. You don’t expect lunatics, I don’t know, waving a traffic cone and waving a stop sign in front of you in a highway kind of scenario. Of course, the stakes are a little different than higher on certain aspects, for instance, the speeds are higher, you have to be traveling at 70 miles an hour with a 40 pound vehicle instead of a smaller passenger vehicle at 25 miles an hour. Those kind of risk factors being shifted. But otherwise, it seems to be more doable, if you will. But even further than that, so that is, let’s say, from a perspective of an individual, autonomous truck handling highway scenario is easier compared to an urban center, as they say. But even further than that, what we did was to combine humans cognition, which is missing in machines, with the precision control and acceleration that machines are capable of doing. We essentially bundled the two tracks together, with the front truck driven by a human operator acting as a cognitive filter of the environment, and the robotic truck, the fully autonomous follower, closely following that human driven leader as a result, significantly reducing uncertainty in the world because all the uncertainty needs to deal with is whatever a few meters it has between itself and the leader. And that’s it, just like precisely follows towards the leader in front of you and try to stay in your lane. That’s the only responsible to our autonomous vehicle in our autonomous car voice and at Locomation.
Eric Hornung 24:43
Is it only two trucks in a convoy or can you make like a train?
Tekin Meriçli 24:47
So that’s a good question. And to start with, we are actually doing two trucks, two truck convoyes, and there are a few reasons behind that. From a technology detailed perspectiv there is this concept in control theory called string stability. So imagine a chain, long chain with many links. So the first link in the chain is your leader. And the second chain is the follower and so on. So there is going to be some error propagation between each vehicle along that chain, right? So let’s say the leader wiggles in its lane a little bit, and the follower kind of perceives it and add on top of that the the error in the the followers processing of that data, and it wiggles a bit more as a result. And the third vehicle observes the second vehicle as its leader and sees its wiggling, and then it becomes a bit more because of that amplify that and propagated error. And by the time you arrive at the end of the chain, your last vehicle is all over the place. It’s not even in the lane anymore. So the only way to deal with that problem is to inform all the vehicles, all the follower vehicles in the chain, about the absolute position of the the ultimate leader. So that’s kind of a technical complication, wedid not want to deal with that from a technology development perspective. Aside from that, when you actually put more than two or three trucks in a convoy, all of a sudden you are looking at several hundred feet long road trains. That will be very frustrating for people in traffic, right. So there’s the social acceptance aspect of this as well. So nobody wants to miss their exits because of a half a mile long road train. And the third thing is, some infrastructure elements are not ready for that kind of rate density in a very small footprint. So some overpasses, some bridges cannot actually take that kind of weight density, because we are talking about 40 ton trucks. And if you just bundle them, like multiple of them, and going over a bridge, because of the resonance they will create and the rate density they will create, some of those road segments may actually collapse. So we don’t want to do that either. That’s why starting with two vehicles seemed to be a sweet spot.
Jay Clouse 26:58
When you guys were thinking about all these factors, you said I think autonomous trucking makes the most sense. Were you thinking from the perspective of, we want to commercialize a business here, because you guys are at CMU, were you thinking we want to eventually commercialize a business? Or was this a research project first that you’re curious on solving?
Tekin Meriçli 27:17
It was actually about commercializing a, a business. So again, in our past lives, we got involved in a lot of commercialization processes. My title actually used to be Commercialization Specialist, but back at NREC. So you oversee the entire lifecycle of a project, start with the the proposal, propose it to the funding agency, that could be government agents, that could be industry, whoever needs a solution to their problems. Then you form a team, you devise a technical execution plan, you execute it, you demonstrate the technology, the advanced prototype, hand it over to whoever is your sponsor, for them to actually harden it and turn it into a product, and then go back to square zero, right another proposal, and propose it, etc. So it was like a groundhog day, if you, if you look at it from that perspective. So we kept doing the same thing over and over. Not that I did not enjoy it, I loved it. But over time I started thinking, you know what? I mean, we are very familiar with this cycle. And many of our colleagues by then had actually started their own startups doing stuff, I don’t know, that the autonomous vehicles or some other applications, etc. And you could see the the commercial and potential financial upside of things. I mean, at the, at the end of the day, this is a business and everybody hopes to make some money.
Eric Hornung 28:36
Just a little bit?
Tekin Meriçli 28:36
Just a little bit, and advance their personal agendas, right? The intellectual aspect of staying at academia was great. I got great satisfaction, but I felt like okay, now it’s time to turn all that knowledge and experience into something that I can’t see in the real world operating and, as a result, making some money. So that that was kind of the decision behind that, that move.
Eric Hornung 29:00
How far away are you guys from actually deploying one of these in the real world?
Tekin Meriçli 29:05
So our prototypes are actually functioning as of today. A couple of months back, we completed our first successful close truck tests, demonstrating fully autonomous convoys. Towards the end of first quarter this year, we will start our pilot operations with certain fleets. And our commercialization schedule is rather aggressive. And we are looking at end of 2021 as our first commercial deployments in small batches, essentially.
Eric Hornung 29:32
When you say fleets, who are the customers whom you’re selling to?
Tekin Meriçli 29:35
We are starting with the large fleets, carriers. And then once we actually implement our solution in their system and demonstrate that they kind of see the value of implementing that technology, they will likely have the power to influence the manufacturers as well. And we are talking about really large fleets, they have the power to ask, okay, I want this kind of feature, that kind of feature in our next generation products. So our goal is to be able to actually make Locomation technology a checkbox in the purchasing form. So then you go to a truck dealer. And it’s also really interesting when it comes to trucks, it’s, it’s very different than passenger cars. You cannot just walk into a dealer and walk out with a vehicle. You actually have to specify all those specs, okay, like this kind, I want this kind of breaks, I want this kind of steering, I want this kind of item as features. And then based on that, that specification, your track is scheduled to be billed that way, and you receive your track several months after that. So what we would like to achieve is to have Locomation technology as an additional checkbox in those forms. So people will be able to say, Okay, I want I autonomous convoying feature coming with my trucks in the next iteration.
Eric Hornung 30:48
What is the hardware component of that? Is there one?
Tekin Meriçli 30:52
Yeah, so our technology is composed of hardware and software. So we have our sensors, we have our computing platforms, and we have autonomy stack on top of that, the software stack. So our retrofit kit, aftermarket retrofit kit is a combination of all these components, hardware and software.
Eric Hornung 31:09
And when you look inside of a truck, there’s hundreds of components already. So do you have to play nicely with those? Or do you have a layer that kind of sits on top of them
Tekin Meriçli 31:17
Even advanced prototypes right now, when you look at them from a distance, you do not actually notice anything different. It may be, I mean, the, I guess the most salient features are the sensor pods on the sides of the vehicle; but otherwise, it looks like a regular truck. And then you go inside, you do not see anything built in and that’s because our engineers actually nicely designed and tucked all the components in, all the hardware components. And so from an outsider’s perspective, the only salient features are the sensors from outside. But from inside, it is not even visible. So we were able to integrate our technology very, very nicely into the existing package.
Eric Hornung 31:56
You mentioned there’s a shortage of drivers in the trucking space, and you mentioned two convoys. Is there still a driver in that second truck or…?
Tekin Meriçli 32:04
Yeah, that’s it. That’s a great question. Maybe I should actually…We talked about Locomation a lot, but I did not really tell what we are exactly doing. So we are essentially doing what we call autonomous relay convoy. So the two trucks start with two drivers in them, they drive through urban areas and such, which are difficult for machines to handle. They go on a highway, they get into that convoy formation, and once the system says okay, all my sensors are working properly, my software is working properly, etc. etc., so you are ready to go. And then the follower vehicles driver presses a green button saying about autonomy engaged, and then releases the control. And at that point, the driver does not have to actually be on duty. They can leave the driver’s seat, go back to the sleeper berth and start resting while the convoys actually moving. Doing that, we essentially reduce the operating costs by half because the most expensive item in truck operations is the operator cost, the driver cost. So, in our system, effectively only one driver is controlling two trucks at any given time. So we are reducing the operating cost by half. And because of the convey formation, the platoon formation, we achieve reduced air drag around the entire convoy; and as a result, the vehicles burn less fuel, so we’ve saved some fuel there as well. And all combined, we are actually looking at a 30% cost reduction in overall operation, which is huge when it comes to an industry that operates with razor thin margins like 3% to 5% profit margin. All of a sudden we are giving them that magical 10x. So they travel in that code formation for a couple hours, however many hours the drivers are allowed to drive, for instance, and then the follower vehicle driver, all well rested, wakes up and sits in the driver’s seat, takes control of the vehicle, and then goes and passes the former leader, becomes a new leader, and the former leader, now the follower, all like driven for several hours and tired, presses the button, says okay autonomy engaged and then releases controls and goes back and sleeps. So that way, the convoy keeps going without the tracks having to stop, because traditionally when you have one driver per one vehicle, when the driver goes out of their hours of service, driving hours, and they need to rest, the truck also rests with them, it just sits there. So there is huge problem of low asset utilization. So, traditionally, trucks are utilized only for like 30% of their times, and they are super expensive assets. And you don’t want them just sitting and collecting dust, I don’t know, 70% of the time. So with our approach, we increase our asset utilization close to 90%, reduce operating costs, reduce fuel costs, and increase delivery speed because the vehicles keep going, so we deliver goods twice as fast, twice the distance in one shot. So it’s all like win, win, win, win all over.
Eric Hornung 34:58
So the truck shops love you, the truck stops.
Tekin Meriçli 35:00
Yeah.
Jay Clouse 35:01
What about the drivers? So I understand you know, drivers get paid by the mile, like when they’re driving, they’re getting paid when they’re resting this, this truck is now still going. Is there any benefits the driver where they say I really liked this technology?
Tekin Meriçli 35:13
There are multiple benefits actually from the drivers perspective. And maybe I should also say that our vehicles are identical in terms of hardware and software, so they can actually do that swapping thing, they can swap between a follower and leaders. But having that capability also enables us like running our autonomous tech in the leader truck as well, but not controlling the vehicle as an even more advantageous, if you will. So our autonomous stack essentially improves safety of the entire convey and the driver. On top of that, because of the increased delivery times, are reduced, I should say, delivery times, the drivers can get back to their homes much faster, because especially long haul drivers leave their homes and come back, I don’t know, weeks later because of all the long driving and rest they need to take. But with our technology, they can be back home in a few days or even within the same day, depending on how you define your delivery segments. So that’s another plus. The third plus is essentially what we are doing right now with two tracks and two drivers, but one of them being kind of autonomously driven, is essential implementing the well known concept of team driving. So in team driving, usually a single track is deployed the two drivers in it. So the drivers take turns driving, and it’s actually a familiar concept. One of them sleeps in the back while one of them is thriving, but that kind of violates personal space and, and privacy as well because I mean, even though you may be traveling with your best buddy, it is still a little uncomfortable having to live in the same cab, right? With our system. We are still doing team driving, but with two trucks. So each driver has their own personal space, has their own privacy. Only one driver is actually controlling the vehicles. It’s essentially the same thing, kind of distributed teams driving, if you will. So from those perspectives, actually, the drivers are also very warm to this possibility.
Jay Clouse 37:07
So does, does the customer, the fleet manager or the owner of the fleet, do they pay you when they purchase the software? Or is it like a SAS agreement ongoing for the software?
Tekin Meriçli 37:18
So the model we have in mind is some sort of a subscription model, monthly payment over three years. The reason we picked that three years as our term is that, so trucks accumulate a lot of miles, and normally in the traditional operation, the fleets actually end up changing their tracks every five years. But since our system increases asset utilization, they are going to churn through those fleets or those inventory, if you will, their trucks much faster than five years. So that’s why we said three years would be a good time to actually upgrade the system to change into like the next version of our technology. So, but to answer your question, yes, it’s going to be subscription model over three years.
Eric Hornung 38:01
Is that common in this space? I feel like this is a large cap back space. And now this is more of a operating expense. What is, what’s been the feedback you’ve got from customers for that model?
Tekin Meriçli 38:11
So far, it’s been very positive. I think that’s, that’s aligning with their existing operations that the way they operate today. So it’s going to be a very kind of seamless integration to their pipeline.
Jay Clouse 38:24
I see a lot of news stories of like, fully autonomous trucks making their first trip all the way across the country, right? And I’m sure that’s like kind of a demonstration phase. That’s not a commercialized effort. Do you expect Locomation to be the first commercially deployed autonomous trucking technology?
Tekin Meriçli 38:41
We believe we will, because as I said, since we are taking a very cautious and very well-scoped approach, we will be able to validate the safety of the system a lot easier and sooner than full autonomous vehicles. All it boils down to, I mentioned like the differences Between demonstrating the technology versus turning it into a product. So that difference is safety and reliability. So if you actually keep your problem very open ended, you cannot guarantee that you actually touch all the potential edge cases and you can actually handle everything that the vehicle will be exposed to in the real world essentially. So that why that long tail we reveal really, really long for people who are trying to achieve full level four, level five autonomy from the getgo for individual vehicles. And by cutting that kind of scope and limiting the autonomy responsibilities to only following a leader and staying within its lane in our system, we believe that we will be able to check all those safety boxes much sooner and much faster than our competitors and we’ll be able to deploy this as a product much much earlier.
Jay Clouse 39:50
I’m trying to wrap my head around this aspect of, if I’m if I am a long haul driver and I have a 600 mile trip and I can drive–hese are made up numbers–but I can drive 200 miles in a sitting or something, you know, now my earning potential is 400 miles versus 600 miles even if I do get home faster. Do they like that trade off
Tekin Meriçli 40:10
Depends on the person I guess. But there are only the certain types of operations like this relay stuff is not something new that we are introducing, we are just blending it with our technology. But some fleets already do those kind of segment definition along a real long graph. So they have relay points, the driver actually takes it to that first relay point, meets another driver coming from the opposite direction, exchanges, loads and then goes back to their origin points. So this is already a concept that is being implemented. So when the entire route is really long, it does not necessarily mean that that entire route is dedicated to a single drive.
Jay Clouse 40:49
I guess a better question would have been, there is a shortage of drivers. So does this actually like on an annual basis? Would a driver who is full time employed trucking earn less if he’s in a Locomation…?
Tekin Meriçli 41:01
No, actually, it’s going to be the opposite because as I said, we will essentially be saving almost 30% operating costs for the fleets. And since they have, they already have driver shortage problem and retention problem through using that 30% saving a portion of that 30% saving to increase the salaries and living conditions of the drivers, they will actually be able to attract more talent and retain their existing talent, right, existing driver pool. So that’s gonna be paying even better than our technologies implemented to the drivers.
Eric Hornung 41:33
How do you decide how much to charge for something like this?
Tekin Meriçli 41:35
You mean the technology part or the overall operation?
Eric Hornung 41:39
The technology part like what are you charging customers for this?
Tekin Meriçli 41:42
Of course, we need to offset the retrofit costs. But on top of that, we are providing certain amount of savings. So our business development folks are crunching those numbers to make it appealing to the customers. We are essentially saying Okay, you know what, we make you save this much money, so give this percentage of that saving to us. So that’s pretty much any kind of subscription model follows, right? There are also well established pricing practices, I meam trucking is a very, very well established industry. So, we are also trying to follow the the norms there. But even then, because of our technology, we are able to actually provide all of value and ask for a little bit of return,.
Eric Hornung 42:24
Do you see a future where your technology is licensed directly to OEMs rather than going through the fleets?
Tekin Meriçli 42:30
That is one possibility. As I I said OEMs and and fleets are actually working very closely together because especially large fleets when they purchase new trucks, they purchase in thousands or 10s of thousands. That also gives them a lot of leverage over suppliers and manufacturers. And they are already, as I said, working very closely getting feedback from the customers, okay, I really like this feature, so make this part of your next generation release or whatever. So hopefully, Locomation is going to be one of those features available on the on the purchasing form.
Eric Hornung 43:00
It feels like every new Freightliner or Kenworth or Volvo or whatever that comes out all has like some aspect of Driver Assist. How does this technology work with or usurp that?
Tekin Meriçli 43:12
As I said, I mean, our autonomous stack actually runs on both vehicles controlling one vehicle and providing assistance to the driver of the other vehicle. So it’s going to be complementing the existing a-das technologies. It’s going to be an even more advanced version, of course. The reason that fleets are actually adopting those kind of technologies, assistive technologies, is that they want to actually reduce their accident related costs, insurance costs, etc, etc. Right? There are also very interesting cases where people in traffic actually aim for trucks of big fleets knowing that okay, those guys are making a lot of money, if I get into an accident with those guys, I’ll be able to actually mute them.
Jay Clouse 43:51
Oh, wow.
Tekin Meriçli 43:52
And they, they do that. And because nowadays most vehicles are equipped with dashboard cameras and all those adas systems, they have the evidence now. You know, what this guy actually did that on purpose, it was not a genuine accident. And here’s the evidence. So with our technology, are sensing, we will be able to provide even more detailed scenery, construction and evidence for those kind of purposes. So that will help reduce insurance costs. And since our autonomy, technology will be inherently safer than a human driver, again, that will reflect on insurance premiums and such.
Eric Hornung 44:26
Who owns the data that comes off of all of this?
Tekin Meriçli 44:28
Most likely, Locomation is going to own it; but of course, it depends on, I guess, the agreement between us and the customer. So we may be flexible there.
Jay Clouse 44:38
What is the 5 or 10 year vision for Locomation? Lke what’s the what’s the big future goal of the company?
Tekin Meriçli 44:44
The future goal is, I mean, this is inevitable, this automation stuff is happening. Coming from an autonomy background and knowing how difficult this problem is, as I said, we are very careful about crafting this, this as a progression as a gradual process. And autonomous convey is our entry point. And while we are progressing along the timeline, the autonomy capability of our vehicles are going to increase gradually, and we will be expanding the operational design domain. And at a certain point, throughout our operations, like dense operations in the freight network, will be able to identify, okay, like, these segments actually seem to be suitable for fully autonomous operations, and our vehicles are now capable of operating fully autonomously, so, we will start deploying fully autonomous vehicles, trucks, in those select segments. And we will essentially be gradually progressing along that timeline. And in the, I don’t know, 5 to 10 years, we will start seeing some, I guess, a mix of autonomous convoys and individual vehicles operating maybe in smaller kind of segments. This is the area, this is happening, and we will be a part of it, and we will be most likely the first implemented deployer of it.
Eric Hornung 45:57
Love that. If people want to find out more about you are Locomation, where should they go?
Tekin Meriçli 46:03
We have a beautiful website, Locomation.ai. Info about the company, our technology, our founders and team, and other contact information can be found there. People also could reach out to me over LinkedIn, Tekin Meriçli is my name. I guess I am privileged to have a very unique name, so when you actually type my name on Google, only one…
Jay Clouse 46:23
You’re gonna find it.
Tekin Meriçli 46:24
Only one person comes up. So yeah, either through a personal channel or through our website, they can reach out to us, and we’ll be happy to talk to them about our technology.
Jay Clouse 46:38
Alright, Eric, we just spoke with Tekin, the CTO of Locomation here at CES. Hot takes on our first autonomous company?
Eric Hornung 46:47
I’m dumb.
Jay Clouse 46:51
Yeah, I felt that too.
Eric Hornung 46:54
Yeah, I felt really dumb.
Jay Clouse 46:56
Tekin definitely has spent a lot more time tackling more complex issues than either of us have. Was a really eloquent speaker actually, though, in the way that he describe pretty complex topics, you know. He really went to on some, some tears, but they’re really well structured, and they helped me really go from basically nothing to understanding a lot more about what goes into this and why his team may be particularly suited to do it well.
Eric Hornung 47:21
Yeah, he did a really nice job explaining it to a couple of dummies over here, Jay. But I think one thing that can’t be overstated is his just like mastery of the space in general and being able to explain to us where the problems in the space lie.
Jay Clouse 47:38
Yeah, it felt like because they are taking this stepwise approach of having humans in both vehicles, he had what felt like the license and the freedom to speak to the shortcomings of autonomous driving, and be realistic and talk about the timeframe in which we might see these things on the road n a worl where there still are human drivers. It seems like they’re really taking a different approach to commercializing, this getting this on the road, in practice with real commercial customers and not just demonstrating the technology.
Eric Hornung 48:10
We heard a really cool question that we’ve never heard before on upside here at CES this year, which is a question from a VC to a founder of, why don’t you just go take a $250,000 job? And in a very direct parallel here, Tekin was offered probably at least a $250,000 job with Uber. A lot of his colleagues took it, 40 of them, and instead he decided this was the better approach. So that is, I think that speaks to, one, his belief that autonomous isn’t there yet. There’s definitely an autonomous future, as he says, but it also speaks to this is the now and this is what we’re building on the road to the future myself. And I like that.
Jay Clouse 48:55
Yeah. He talked about kind of the pattern of when you’re doing research, you solve a very specific specific problem, you kind of see that one very specific result. And then you start that process over and over again. It seemed obvious that he really wanted to sink his teeth into something that was a little bit more long term. We didn’t have time to dig into this. But I’m really impressed that, at this point, you know, they’ve raised only five and a half million dollars. It seems very research intensive, time intensive and potentially capital intensive for some of the things we’re trying to do. And something else you told us off air was that they started Locomation from scratch. Leaving CMU they started this from scratch. They didn’t license any IP there. It just seems like an expensive endeavor that they’ve really gotten a lot of mileage–pun intended–with five and a half million dollars, which I think bodes well to a really smart, resourceful founding team.
Eric Hornung 49:46
So do you think it’s worth it? Do you think there’s opportunities big enough?
Jay Clouse 49:49
The hardest part for me to think about this is some themes we’ve heard on the show talking to WorkHound, for example. There’s clearly a driver shortage and a driver retention problem in the trucking industry. Huge industry, it’s going to go towards autonomous. You know, this might be the first commercial application of it, but it’s going to go towards autonomous, and drivers going to be out of jobs, or maybe the jobs that they can’t fill right now, you know, are just automated out anyway. But the drivers who are still sticking around, there’s this phrase, keep the left door closed, you get paid when the truck is moving. And in this world with Locomation, the truck is moving and you’re not getting paid. And so what I brought up to, to Tekin was okay, but doesn’t that mean that ultimately you’re monetizing less of the trip as a driver? It seems like there might be an incentive misalignment there, but the point he made being drivers can get a higher salary because the company’s saving more on operations. If that’s true, then that probably takes care of that. But I think we’re too early to see if that will remain to be true, or if those companies will, you know, just take those savings and put them elsewhere.
Eric Hornung 50:53
Do drivers get paid when they sleep on the side of the road?
Jay Clouse 50:57
No.
Eric Hornung 50:57
So really, it’s no different than their current situation.
Jay Clouse 51:00
It’s very different because when they sleep on the side of the road, the trucks not going anywhere. They still get paid for the entire length of the trip. If the trip is 600 miles, they’re going to personally drive 600 miles. But if the truck is moving 200 miles of that while they sleep, they’re not getting paid for that time that the truck is driving itself.
Eric Hornung 51:20
I guess my assumption is that there’s no downtime in between jobs, which may be a bad assumption.
Jay Clouse 51:26
Yeah, I mean, especially on long hauls. And he said, some of them, some of them work in convoys, and they switch off anyway. But yeah, when you’re, when you’re off the clock, you’re off the clock, but you still have the entire trip where you will get paid. It’s just kind of like, how much distance can I cram into this period of time. And in this world, you’re not getting paid for some of that distance that’s going on. Hopefully, that money gets just pushed into salaries, like he said, that it will increase retention and better attract new drivers to those trucking companies. But anytime there’s an incentive misalignment with somebody in the value chain, I get a little bit worried. But in addition where you think it’s probably going to go totally towards autonomous anyway, I think a driver would rather be assisted in the autonomous vehicle than wiped out by the autonomous vehicle.
Eric Hornung 52:09
Yeah, I think my shadows come along the lines of their go-to-market, what’s the price point look like, where they’re selling the fleets right now. Obviously, I think it’s better selling to OEMs, because then it’s kind of baked into that purchase order that he’s talking about when you go to Freightliner or you go to Kenworth, or you go to Volvo, or whoever it is, and you say I want convoy, autonomous convoy, then it’s this is our autonomous convoy solution. It works perfectly with every other instrument and everything in the truck. Right now it’s more of an aftermarket solution for the fleets.
Jay Clouse 52:43
Totally. And if you’re running a fleet, you want this in all of your trucks because the important functionality is you can switch off with those two drivers, you know, and sure if you have 100 trucks and 30 of them have this on there, you could control for that and only have two Locomation enabled trucks going together at any given time, but that naturally will limit the routes that you can put together and who is working with each other if 3 out of your, or 30 out of your hundred, you know, have it.
Eric Hornung 53:10
I think when you think about autonomous trucking and our kind of bucket sizes problem, this is an interesting one to look at. Because if you look at autonomous driving and how big that can be when it’s fully autonomous, that’s a massive market, that’s easily over our 20 billion kind of top tier two big care bucket. When you look at this, it’s almost like, you said, stepwise earlier, and I think that’s a good phrasology here, because the near term market is probably pretty large. We don’t have the numbers for it. But then there’s this question of how soon is fully autonomous? The longer fully autonomous is out. The bigger this opportunity is in the near term.
Jay Clouse 53:49
Right, it’s a window,
Eric Hornung 53:50
Right, It’s a window.
Jay Clouse 53:51
Unless, unless local nation can get to a fully autonomous future themselves and become the driver of choice.
Eric Hornung 53:57
Yeah. And then they already have the first foot in the door with all these business relationships. And I think that’s the goal, but I think there’s a difference between that happening in three years and that happening in 15 years.
Jay Clouse 54:07
Some of the value created here, you have the obvious asset utilization increase for the fleet owners, you have 30% cost reduction. Those are awesome numbers that make it obvious that there’s real value for the fleet owner. But to your point about OEMs, if asset utilization is increasing 90%, and now, instead of changing over every five years, fleet owners are buying trucks every three years, sensibly, that’s more value created for the OEM as well, which you would think there’d be incentive then for the OEM to really want this in their vehicles.
Eric Hornung 54:38
Is that possible to do without a continued increase in demand, though? I was thinking about that question, and I don’t think I have a good answer. Just because your technology can increase utilization doesn’t mean that if the underlying demand isn’t growing at a fast enough rate, it will.
Jay Clouse 54:53
I would imagine that demand is increasing at a faster rate with the, you mentioned at the very top of the interview, consumer expectations of how quickly you get freight to you. Now, is that serviced by more trucks on the road? I don’t know.
Eric Hornung 55:07
Or do fleets just need less trucks at a faster clip?
Jay Clouse 55:11
I don’t know.
Eric Hornung 55:12
Well, there’s a bunch of things we don’t have answers to.
Jay Clouse 55:13
Yeah.
Eric Hornung 55:14
But what we do have an answer to Jay is what you want to see from Locomation in the next 6 to 18 months.
Jay Clouse 55:21
I want to know more about who their first customers are, how much they’re paying.
Eric Hornung 55:27
Are you kidding me?
Jay Clouse 55:29
What, is this what you wanted?
Eric Hornung 55:30
That’s what I just said I was gonna say.
Jay Clouse 55:31
Oh really? It’s probably, I probably thought it was my idea. Do wanna know who the first customers are and how much you’re paying. And if that’s your answer, then I’m sorry.
Eric Hornung 55:40
That’s okay. My answer is I want to get some clarity around the numbers, the business model, what they decide, is that fair cut, because if there’s going to save 30% on operations, is it a win-win at 15%, and what is 15%? Right? What is that number? Is this a large dollar figure every month or is this a couple hundred bucks? I just don’t have any sense of the scale of what that is. And I think we could probably dive into a lot of research and try to figure out what that is. But it’s just a bit of a blind spot for me right now.
Jay Clouse 56:15
All right, well, we’d love to hear your thoughts. You can tweet at us as always at up @upsidefm, or you can email us hello@upside.fm. Otherwise, we’ll talk next week.
Interview begins: 5:25
Debrief begins: 46:38
Tekin Meriçli is the cofounder and CTO of Locomation, a company focused on developing autonomous trucking. Founded in 2018, Locomation hopes to implement autonomous driving technology into freight trucks to create convoys of semi-autonomous, highway-travelling trucks.
Tekin became interested in computer technology from a young age while growing up in Istanbul. An alumni of UT Austin, Bogaziçi, and Carnegie Mellon, he has a multitude of engineering experience in the field, which he brings to Locomation.
We discuss:
- Ad: Finding experienced employees for your new business with Integrity Power Search (3:48)
- Tech community in Istanbul (7:37)
- DARPA challenge and Carnegie Mellon’s advances in autonomous AI (11:36)
- The idea for Locomation and the Artificial General Intelligence problem (17:35)
- Locomotion truck convoys (23:56)
- Creating a business from autonomous driving (26:58)
- Locomation’s progress to date (29:00)
- Truck drivers of Locomation (31:56)
- Implementing Locomation (37:07, 41:33)
- Will Locomation be the first? (38:24)
- Decreased driver’s pay vs. improved driving conditions (39:50)
- 5-10 year goals (44:38)
Learn more about Locomation: https://locomation.ai/
Follow upside on Twitter: https://twitter.com/upsidefm
Advertise with an upside classified: https://upside.fm/classifieds
—
This episode is sponsored by Integrity Power Search, the #1 full stack high growth startup recruiting firm between the coasts. They partner with venture capitalists, private equity groups and CEOs to build amazing teams for the world’s most disrupting companies.
Learn more about or get in touch with Integrity Power Search: https://upside.fm/integrity