EP 115 - Mohammad Ashouri | Driving Autonomy: Unveiling the Tech

EP 115 - Mohammad Ashouri | Driving Autonomy: Unveiling the Tech

Show Video

Driving a car, stepping on the gas and changing gears can become a luxury in the future. You are replacing the services of transportation. Tesla's main product is not Tesla's car. Tesla's product is a megafactory. The driver's decision-making is ultimately skill-based.

When you talk about a self-driving car, everything is knowledge-based. If you want to build a robot that can To be a generalist and be able to do a lot of things, you have to have a human form. Because we created our world for humans. The main challenge for you to raise your level of autonomy is to be able to generate edge cases and test them in a similar environment. This is another model, another simulator that I modified. We called it Driving Cube.

My project is not a better mode now. My project is a team that is building a better mode. Working in a teamwork requires a special attitude.

If there is no competition, there is no skill to be gained. It's all about how. In LinkedIn, Mohamed Reza, you're in the UK. Yeah, yeah. You're in the UK? Yeah, I'm in the UK.

In San Francisco, I don't know. We haven't seen each other in London either. Yeah. You can count on me. I was in Fana, and I sent you a message, and you replied, and we connected.

This is my new case. Yeah. Well, it's not working, but it's fine.

Let's see. Are you okay? So I didn't see you anywhere. We only talked online.

Yeah, we only had one screen and we talked for about half an hour. It was our first meeting. Yeah, in short, for a long time, I actually started listening to your podcasts from time to time. Because you know, people come from abroad, especially if It's a European country, and the Iranian community is not very strong.

People don't get bored, especially my age, who travels a lot. The environment we live in, because we live in our own village, there aren't a lot of Iranian people in the environment, like in London and Manchester. And it's very stressful to be alone. I was lonely, so I started a podcast. Actually, the podcast came into existence during the COVID era. And there's another point, it's possible that at the end of the week, you'd like to hang out with your friends, go to a bar, go home, and so on.

But generally, people like us prefer to... When you are sitting together, you are talking about topics that you are interested in. The probability of having this with old or young friends is very, very low. Maybe even zero.

And you have to find a new network. I personally feel that I was lucky to have this experience in the world of startups in Iran and Toronto, where a lot of people have come. I tried to collect this network and I kept it.

And I realized that the main issue is that, It's a matter of how brave we are to sit here and there. Or say, how brave we are to sit here and talk about our own topics. I can understand these two worlds. One world is this way, the other is that way.

This challenge you're talking about, Sohail, I've had it since I was in college. I wasn't one of those nerds who only focused on books and classes. On the one hand, I was doing the right thing. I mean, I didn't have any interest in universities or classes, but... I didn't know how to read, I didn't know how to read books, I didn't know how to do everything in life. I wasn't a single-minded person. In short, we had some friends in terms of gambling and gambling.

Some of my friends, who are now partners, or I don't know, people who are in the same university. It's very difficult when you want to talk to people who are close to you and have a lot in common. Especially when you enter the world of work, tech, and I don't know, this kind of stuff.

You see your friends, for example, their whole phase is on the other side. To some extent, you can have fun and entertainment with them. But if you want to talk about topics of interest, you can't do much.

It doesn't work. I mean, it's very cumbersome. It's something that I think is missing a lot. When we were in college, people would always say, who is this guy? I don't know, he shaved his head like this. I don't know, he doesn't look like this at all, he doesn't talk like this. When you went to college, you had other challenges.

You had to do two things for your community. Son, the reason I wanted to talk to you is because you are in the field of machines, Oh, Pharmoon and Dando. Autonomous driving, maybe. I think you also had a job in Iran. You worked for Saipa in Iran, and you were in the world of car stories.

Now you're in the UK. in the story of cars and all that. The main thing I remember talking to you about is that your story was very cool.

I mean, it was like your passion as a kid was that you loved Range Rovers and classic cars and that kind of thing, and now you were able to reach that dream in some way. Is that right? Do I remember what you said? No, it's definitely right. Now, from my interest as a kid, I can say that yes, I was a car lover.

When I was a kid, my dad was the beginning of the revolution. I mean, the beginning of the post-revolution. He would take five Range Rovers and change them with different colors. So, since I was a kid, I've always been interested in cars. My cousin, who I chose in college, I would take him to mechanics classes.

Later on, we didn't go to mechanics classes at all. I didn't even think about it. At that time, the economic situation was such that if there wasn't a lot of money in the car, they wanted to keep the shares.

They would register the car and sell it. They would register the car every time a new one came in. My goal was to go to the catalogue, get the car, sit in it and read it. I would sit in the car and listen to Zabtar, and on the first day, I had already read the entire catalogue.

It was time for me to go to sleep in the car. As I was saying, I was very young. And a lot of these stories and experiences came to me later. I got interested in going to university because of this. I tried to go into the field of mechanics, towards the field of hydraulics and hydraulics, which was in a mechatronic state. Those who know what dynamic control is in the field of mechanics, it has a lot of overlap with electricity, electronics, mechatronics and so on.

We went into those areas, creating driving simulators, and I can do research on these autonomous driving areas. It can be said that almost everything I worked on from the beginning was related to this area. I mean, I tried not to have a deviation in my resume or in the old code, for example, to jump from this area to that area.

This interest of mine led to the fact that after the time I was in college, we made video simulators, and we did diagnostic tests, and I don't know, it was a big hit at that time. We did a lot of interviews on TV, radio, newspapers, and things like that. We used to make fun of each other. For a while, it had become a trend on TV that whatever I said, for example, if someone had invented something, they would say, it's the first time I'm hearing about it in Iran. In short, we started these simulators about 5-6 times, now with different levels, and each time the same repetitive words and the same conversations. In short, this issue also arose, after the story of Pagio and Shito in the university, I went to I wanted to become an architect.

My teacher had a position in the Faculty of Architecture in Saipan. They created a new department called Advanced Wake Technology in Saipan. It was a very short period of time that I worked in one of Iran's large suppliers, called Kourouz, which is the largest self-employed branch. Many of Iran's branches are self-employed, and a part of Saipan.

Then we went to the invitation of my co-founder in Saipa, Department of Accuracy. He is a co-founder of advanced technology in the Department of Accuracy, not a startup. Another topic was electrification, which was about electricity.

Another topic was autonomous driving and assistance systems, which is called advanced driver assistance systems. I was leading that area. We gathered the forces, we got the equipment, we put it abroad during the training, and we broke it. It was the same game, the same thing, where the sanctions had already been imposed.

I think we went forward with a lot of enthusiasm. I was able to get good budgets, and we spent about 10 million euros to get the equipment. and the training period, and so on.

So we worked for a while, and we closed the deal with the suppliers. The platform that Shahin is on right now, came out of the site, and other teams were working on it. There were supposed to be a series of features Not in the form of a code run, but with lower levels of autonomy, like a self-driving car, a blind spot, and things like that.

It had a series of placeholders, a series of buttons, and things like that, on the code runs that were displayed on it. Then, shortly after the ban, they cancelled all the suppliers that were supposed to be there. The project went through a hold and we were left with a series of equipment and research projects, which I decided to move to England after 5 years.

Now, as you can see, my microphone is a mess. Yeah, it's a mess. Yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, First of all, the state of self-organization in Iran.

Because when it comes to, I don't know, Iran Khodro and Saipa, we always talk about everything in a big way in Iran. Everything is macro. You know, personally, I'm interested in the broader points. I mean, is Iran that advanced? Because on the one hand, you know that companies like Iran Khodro, I've heard of them, These are funny stories.

But on the other hand, this is the situation in our country. Last night I was listening to Elon Musk's podcast with Joe Rogan, which was released a few months ago. all of them emphasized, a thousand times, that it's not a matter of design, it's a matter of manufacturing. Manufacturing and the fact that you can build a car to a higher number, production line, that's the whole point. The point is not that the Iranian companies are probably not able to design cars, the designers can probably be upset, but you can't build it to a higher number.

The production line can't support it. The design of the car has nothing to do with it. Even in the story of self-driving cars like Tesla, we had a lot of startups that came and made autonomous cars.

You know, they made self-driving cars, self-driving cars, cars that they can drive themselves. We have a lot of startups that all raced. The point is not the design. Not this.

Everyone can design. If you give me 2 million dollars, I can build my own car with all these softwares. Even Tesla is self-driving. I think it's open source.

Everyone can use these algorithms. These cameras now have all the image recognition softwares. Oracle and Amazon have all these clouds that you can do all these calculations on the cloud.

Software is also open source. You can really connect two or three Legos to build a car. It's art, but how can you make thousands of products out of it? It's a factory. Tesla's main product is not Tesla's car. It's a megafactory, Tesla's product.

It's a factory, it's a production line. And I think that's one of Iran's problems. Because in Iran, there are a lot of engineers. You say you work there.

It's strange to me that when we... the ABS and the brakes don't work, the steering wheel doesn't work, you honk and honk. You're saying that we had in Iran, I don't know, the sluggishness of the driver and all that, and that's why there's such a contradiction between what we're seeing and the research that's really being done. Where do you think this is coming from in Iran? Let me ask you, the thing you're saying, It's true, but when it comes to... We call it OEM, Original Equipment Manufacturer.

The OEMs that are market leaders, their quality issues and their conventional systems are completely solved. They are talking about scaling up an advanced technology, which is a challenge in itself. we are a few steps behind. It is not a problem to escalate production. Let's move on to sanctions, which have caused a lot of problems in the past.

The challenge in Iran is that they are very weak in terms of production quality and equipment. For example, in 206, A French car that is good in the first half of the series, and then as time goes by, when the design starts, or not design, let's say the interior construction that starts, get it from the same order until things are more advanced. What do they start doing? They copy it and say, we know how to do this, we're going to spend less, we're going to make the original, we're going to make it ourselves. Yes. For example, when they start assembling the car, there is a gradual process of interior design. For example, in the first 20,000 to 30,000 units, we bring a complete stick-aiding.

All the parts come from abroad, and only the assembly is done here. Then they start to discuss the internal production. Now here, for internal construction, self-construction can never enter directly, that is, the main thing is the construction. For example, if a company wants to buy a chair from a foreign supplier, they have to meet the standards of the supplier, and they have to meet the standards of the supplier. In terms of the transfer of technology and design, Iran can get along with some companies in some areas. But some companies, under the conditions of sanctions, do not allow anyone to do interior design in Iran.

Now, the pressure that they bring in terms of the regulations that they want to close, they say, if we insist on interior design, we can, for example, give you permission to do interior design, but we do not give you the plans, we do not give you the standards. The result is that, for example, the manufacturer of the chair or lamp 206, which can based on the standards of the company, for example, Ascon Valeo, which produces for Peugeot. It's the same thing that happens in Turkey.

Turkey is a very professional company, even though they don't have their own brand. This is their concept. And because they don't have the permission or the quality, they don't even have export permission. They say, this platform is yours, you can bring whatever you want on it, but only in Iran. For example, the D2C6 that you see after 3 years of in-house manufacturing, the quality is getting worse and worse. The production lines are not well-equipped, the designs are not well-equipped, the validation is not well-equipped, the test-pillars are not well-equipped.

The power plant itself does not have what it needs. Based on these requirements, I will be able to implement them. For example, you may see yourself as a random, some of them have a good design, the quality is okay, some of them are not. I mean, if a requirement is correct, you have to be able to implement one thing to two different suppliers. But since the requirements are so wide and don't constrain the manufacturer, you basically give one thing and get two or three different things. This is a quality challenge that is in the internal manufacturing and self-manufacturing sectors, which is mostly carried out by the manufacturer itself.

But in Iran, we don't even talk about advanced technologies in the conventional sectors. foreign foreign I mean, the measurement of the size or the KPIs that they have for their own growth is a bid. Not like, let's say, what quality we are modifying. Because they know that people will sell them, people will even buy them.

Even if you have to withdraw money to buy the car. But when it comes to advanced technology, this is exactly for OEMs and market leaders, this issue is solved. They say, we have solved the issue of quality and commercial systems. Now we have advanced technology on the issue of self-employment or electrification.

Now we want to push ramps and scale-ups. There is a challenge. What we had a lot of problems with in Iran, which was the beginning of our department, was the quality issues, because we always had this argument that we still don't know the situation, the system of our switches is this, the owners don't talk about it, that ZEH is on the doorstep, the price of a piece should go down, or no, you should talk about radar and remote control and advanced technology. These two issues are separate. Even if you want to get to that point today, a curve has a lot of smoothness that wants to get to the point where you have something to add in a technological basket. These two should not be mixed with each other.

This has always been our challenge. For example, you would go and hire a manager, you would hire a team manager, you would go and get a certain budget, you would bring in the equipment, you would create a sub-building, a team of experts, because if they don't want to, they won't work with you. Even in a large organization, if not everyone contributes to a specific issue, you won't be successful. I mean, half an hour or the first hour of every session, if not all of our sessions, we will get to the point where we will explain ourselves again for what we are doing this work. Because the door of entry opens, Chinese gods come, Chinese gods didn't have any of these systems back then.

Now look, they all have the same equipment and technology. At that time, they said, for example, we benchmarked, compared, they said, I don't know at all, don't go to European manufacturers, don't go to Korean manufacturers, just compare yourself to the Chinese. When we were monitoring the roadmap, we saw that China's situation is like this now. With this growth rate, in 3-4 years, they will have the same technologies that Europeans or Koreans have. I have been here for 4 years now, and we have reached this point.

Now, the Chinese companies all have the same equipment, radar systems, and their level of autonomy is somewhat different. There was this challenge. In addition, you have to be very sensitive in order to be able to have such things. For example, you want to have advanced systems, you want to have electrification. I mean, I remember, for example, we used to send RFPs to various foreign companies, even in the game of sanctions.

For example, from each of the 20 creators, we had to agree with their insistence that in order for them to have a business with Iran, to sell a piece, they have to give us a proposal. For example, you take one, you go to the commission of deals, I don't know if it's a decision about certain contracts, they say no, it has to be three proposals. You say, well, it's not.

I mean, no one actually comes to participate in this interview. With all these challenges, we were able to get the producers, but unfortunately, the production did not reach the end of the story. I know it's a bit of an obsession, when you scale up an old system with low quality, and you want to insist on having a distribution, or when you reach a suitable quality, then you want to maintain that quality and then scale it up. That's the challenge that the developers are facing right now. Look, I accept that there are two topics, but they are completely related to each other. Because if it is decided that you only focus on innovation, and you don't think about innovation, and you don't think about innovation, yes, there is a department, but they don't pay you to go and do research, because it is not part of the company's priorities.

Look, for example, a company like... First of all, the automotive industry is a very... I think other than Tesla and Ford, all other American companies have gone bankrupt. Now, Europe is also making some efforts.

And in my opinion, what is happening is that Tesla is innovating in every field. From locking the door, to taking control, to taking the battery of the car. Of course, they are not a car company at all. They are a robotic company. Now they have made a robot with four wheels.

But... Or you can even say they are a battery company. What can you say about it? It's a tech company. The transformation that is happening in the manufacturing industry is due to the innovation that Tesla brought. Power plants, traditionally, have a very long history. For example, if you see a new model that wants to enter the market, or a platform that wants to be changed, or a new model or a car, it takes at least four years.

It's a long history even in today's world. In general, Waterfall. But it's not like that in the tech world.

The agile system that is more prevalent in startups and travel companies. You try to create very, very short prototypes with very, very short iterations. You start with MVP, then you start to take maturity higher and enter the market. We are experiencing this transformation in Jaguar Land Rover, which is where I am right now. It's been about two years since they said that they wanted to completely change the venue.

Instead of being a traditional power plant, they wanted to turn it into a tech company. The customers have to be completely open-minded. I mean, as I was saying, there are a lot of changes happening in the field of software engineering. We can say that we are following the same technical systems in the field of software engineering. I don't know if the topic of Scrum is strange or not. But traditional architects, who are now followers, have the same traditional approach.

It takes a lot of time to introduce a new product to the market. Right now you're working for JLR, right? It's a company that... Jack Warhol and Land Rover... It's a British company, Jack Warhol and Land Rover. It's a holding company.

It's not a holding company. It's more like a... It's written there, I was just thinking about it. It's a holding company, actually. The owner of the company is Tata India.

That holding is one of the companies that we have purchased. The Jaguar Land Rover company, which has been sold before. For those who don't know, JLR is the Jaguar Land Rover company. Jaguar is a well-known brand.

Land Rover is more famous in Iran. They are similar to G-Points. Off-road racing.

Well done. The most famous and most luxurious car is the Range Rover, which is a part of Producta, which is produced under the Land Rover group. These two companies have teamed up, and their new brand is Jaguar Land Rover, which is known as JLR.

They teamed up for a reason. MashaAllah, their revenue this year It was $22.8 billion. It's a GDP. It's not an American dollar. Pound.

Pound. Pound. Pound. Pound.

Pound. Pound. Pound. Pound. Pound. Pound. Pound. Pound.

Pound. Pound. Pound. Pound. Pound. Pound. Pound. Pound.

Pound. Pound. Pound. Pound. Pound. Pound. Pound. Pound. Pound.

Pound. Pound. Pound. Pound. Pound.

Pound. Pound. Pound. Pound. Pound. Pound. Pound. Pound.

Pound. Pound. Pound. Pound. Pound. Pound. Pound. Pound. Pound. Pound.

Pound. Pound. Pound. Pound. Pound.

Pound. Pound. Pound. Pound. Pound. Pound. Pound. Pound. Pound.

Pound. Pound. Pound. Pound. Pound. Pound.

Pound. Pound And all of this was due to the fact that the financial situation was not very good. In the last two or three years, practically, with financial support and, I don't know, changing their vision, they have started to overcome poverty and loss of value. Their latest strategy is called Re-imagination. Their target is to take the luxury of Khodra to a higher level, to have a new concept of luxury, at least to the level of Bentley, which is also an English brand. but they are trying to keep the sales.

The product that the market is going to improve will be released in 2025. We are working on the same project. which is an autonomous platform in partnership with NVIDIA, which is a new partnership.

This is a very interesting point, maybe it's a question for a lot of people. I've been a fan of NVIDIA since I was a kid, because of the gaming, the graphics cards, the gamepads, the GeForce. I remember when I was a kid, the computer world was a lot more fun. From going to AMD, one would connect to AMD, one would connect to Intel, one would connect to a graphics card. I mean, our guys were more into this world at that time.

One would connect to a cooler case, one would connect to a desktop, layer by layer. I'd like to have a PC now, because I've been on Macs a lot. But now everyone knows that NVIDIA is a huge company. And it may be a question for many that NVIDIA, because it's a chip maker, and it's a company that most of us know through games and graphics cards.

What happened was that all the computing power was on the CPU. And the CPU, to a certain extent, could no longer have that efficiency, both in terms of computing and in terms of electricity. Then, the graphics cards, because they didn't want to rely too much on the CPU, came with GPUs, a certain amount of their own computation power, which later became much more efficient. I mean, in a way, instead of going and changing RAM and CPU, you would even put a full-external GPU next to your computer, and your computer would become a complete mess. The game started from there, and to mine, They started using NVIDIA's GPUs, the era of Bitcoin and mining, they were very advanced.

There was a lot of fresh blood in NVIDIA's veins. They started again, a new market was created for them, in the gaming world, the market grew a bit. AI and the story of the Hi-Page, since they are all computation-heavy, and the reason we are talking about AI is not because its technology has just arrived or we have just discovered it, but since the 1950s and 1960s, all algorithms were ready, we all knew how it was, but the cost of this computation and the cost of these servers and the cost of this GPUs and chips have come down to the point where you can run those algorithms in scale. Otherwise, it takes a lot of time. And now these machines are also looking at the image with a camera, but you need a very high processing power in the moment that this is a human, this is a human, this is moving, this is green, this is a light, and all of this is being done in real time.

Just show that we are recording a video podcast, This is why this video has to be rendered. I'm like, why do we have to do this in real time? Just do it. But they need a very high CPU for this to happen. And all the companies, or NVIDIA, because I look at the CEO a lot, I'm a very smart and clever person. With the knowledge that the world is moving towards AI, he positioned himself as the backbone of the space that if you are an AI company, you should come to NVIDIA and our GPUs. And now the companies, the machines, like what happened to the mobile world, like Nokia, fell to Apple, not because of the hardware, but because Apple came and made a software phone.

Software eats the world. Now they say AI is eating software. But AI is a software. So still, software is eating the world. And a company like Nokia, which had 90% of the market share of smartphones, has taken the game to the next level.

Because they were working on Apple and Java, and now they have come from another world. Now there is also a change in the world of cars. Cars, apart from these four wheels, I don't know, door and chair and all that, nothing but softwares and softwares. It gives you an over-the-air update.

The car goes up and down. There was even an incident that was similar to Tesla's prototype. The chassis was a bit lower, and when it went over these things, it would take a spot under it, and with a software update, The car's suspension went up. You see, he wants to recall everyone to go to his workshop and change it.

Long story short, I just wanted to say in conclusion, that Mashallah himself was a new place. In conclusion, let me say that what happened was that now all the companies went to Narmafzor and companies like Apple Make yourself comfortable if you want to fix your position. Sahil, the camera fell and hit my foot. Yeah, yeah, I'm careful, don't worry.

Is it okay now? Yeah, yeah, great. No, it's good. Come a little more to the middle. Well done. Should I bring it myself? No, no, it's good, it's good. Keep going.

Yeah. So, what happened is that the companies that are making machines now, because they can stay competitive and they can keep their position, they all went to Narmafsar. And for those who want to go to the software side, this video is a behind-the-scenes video. Now, what you just said was exactly what happened to us. A company that used to be the main supplier of these systems, because you know, self-employed systems have an autonomy level.

You can share the slide later. These are levels of autonomy that start from level 3 to level 5. What is known as 2n2 now is level 2 or level 3. The driver still has to be in the loop and is responsible for the driver. It didn't require a lot of computational power.

It didn't require a lot of components. It didn't require a lot of strange sensors. It was also possible to run it on a DSP.

Then there were old cutting-edge companies that put their own ECUs like Bush, Continental, etc. These were the old suppliers. And if you want to raise the level of autonomy, you need to be able to increase the number of sensors, so that you can have a 360-degree view of yourself.

The range of sensors should be increased, the number of sensors should be increased, their frequency should be increased, because any kind of sensors can detect certain things, but can't detect certain things. This means that you need more computational power. As you said, you need GPUs, and the market leader in this area is NVIDIA. This is why our company is based on this new product, thanks to the partnership. It has a platform, and in this difficult situation, it is very important for us to build it ourselves.

This partnership actually happened because of this. Now, the data can be turned around like this. The war on Taiwan is the same, because Taiwan is the biggest chipmaker. Now even Foxconn, which is the largest company for Apple and all of them, All the assembly is done there.

With NVIDIA, in Taiwan, they're building an AI factory to produce chips. And there's a war going on. Is it the Chinese party? Is it the Taiwan party? China? And this war between Taiwan and the U.S.

is behind all of this. They want all the chips to be there. I mean, you produce Taiwan, you beat Tesla, you beat Jack Ma, you beat everyone.

No, no, that's exactly it. As I told you, the GPU and NVIDIA, and not only the sensors, but also the simulation. NVIDIA is a partnership, so it's not an ordinary supplier for us.

We are developing Safara jointly with them. The test of these systems requires very diverse simulators, and the main challenge for you to raise the level of autonomy is not the technology, but to be able to generate and test the H-cases and all the different states that occur in the world, and to be able to test them in the simulator environment. And in order to get more realistic results in the construction industry, the quality of the construction industry needs to go up.

What does this mean? We are going to talk about the game itself. You have to have a high quality environment, and the properties of the materials used in it have to be perfect, so that the virtual sensors can replicate what is happening in reality. I think a cool example of this is Microsoft Flight Simulator. They practically made a simulation of the real world. The data is also real-time.

So if you're a pilot and you want to fly out, it's not even a game, it's a simulator. The data is interesting because it syncs with real-world data. If it's raining in a certain area in the morning, if it's windy... I mean, you go on YouTube and play Flight Simulator, which I've been watching for hours, for 3-4 hours a video.

Guys, when it's morning, when we want to go... We want to go to Japan. It's checking the Google Maps.

It says it's raining now, it's windy. And everything is like this. They even came to train the pilots in the H-Case. Even if it's sunny, in this weather, in this rain, the color will be like this. Here, it is possible to hit the pilot's eye from this side, from this angle.

They have made the real world and the earth completely similar. That's exactly how it is. I mean, the game design systems, especially in the world of Godro, are a little more complicated. Because in the space you're talking about, the number of objects and items is a little less. On the road you're going, the complexity of the environment increases a lot. And you absolutely need a very, very powerful game engine to be able to do those designs much more accurately.

What I want to tell you is that the platform we are working on is the same platform that is used in the gaming world. And NVIDIA has made a lot of efforts to integrate this virtual environment seamlessly with everything else it has. And more than the gaming aspect, you can have a virtual world and have all your items look real.

We are working on this at a very high speed. You enter the GPU and see that you need to have a good simulation for your system. You need to have good simulation. You need to improve the quality. You need to improve your game engine. NVIDIA is connecting all of this data.

You see that NVIDIA wants to connect all of this virtual world. Even the formats we use for our running simulations have been changed to something that is used in the gaming world as well. The 3D assets that are used in them, the properties of their materials and so on. And it is very important for oil refining to be able to extract those H cases in our operations and to know what the machine has to do in the real world if this happens.

Now, this area that I work on, oil refining, is exactly my specialty, what I work on in sharing. So, when we started this discussion in Iran, my main specialty is oil refining and simulation. which actually brings us back to the university days. When we were in the university, we worked in a virtual reality lab, where we made driver simulators.

I don't know about the other topics, but we worked on real-life puzzles, real-life magazines, and so on. It was a very complicated and interrelated world. Can you tell us a few use cases? For example, what were you simulating? Look, in the simulation, for example, You have a complex urban environment, and the autonomous system that operates in that complex urban environment has an autonomous driving system. And in the urbanization that you are doing, you are completely urbanizing the traffic environment.

Imagine that a pedestrian wants to cross the street and wants to see if he can make a collision avoidance based on the design that his passenger has made. Or you can play with light conditions, or play with conditions. If you want to fade the lane marking, change the quality, see if the sensors detect these or not. In the world, because the levels of simulation are very complicated. What I am telling you now is pure simulation, that is, the self-driving model is simulation, the traffic model is simulation, the driving model is simulation. And the sensors that you have designed physically in the simulation environment, for example, the GPU reads the graphic information, detects it, then decides on it, finds its position, decides on algorithms, everything is virtual.

And you create different scenarios in the city environment. And objects can move at different speeds and conditions in that environment. And the closer you can bring them closer to the real world, the more realistic the answers you get. But the challenge here is that you design a series of scenarios.

I'll get back to the topic of generative AI. In traditional methods, a simulation engineer produces these use cases. For example, imagine a pedestrian walking at a certain speed in a certain position in the middle of the street and moving at a certain speed.

These are all written based on those requirements and a person writes these scenarios manually. Now it's up to you to write the scenario in that language. Then, we call them knowledge-based scenarios. That is, you have a knowledge about a series of scenarios and use cases, and then you want to implement them.

But it doesn't cover all the events that happen in the world. And it is very important that you can create a series of scenarios by yourself and generate them as a batch. You can create a series of scenarios on a massive scale that can simulate all the situations that you have seen or not seen without touching them.

And this is what we've been working on lately, to be able to create scenarios with generative AI that may not even reach the human mind. These are artificial scenarios, generated by AI, and we don't know how it will behave in those conditions. Can I ask a question? Yes, go ahead.

This is a tricky question. The final answer is clear, but it's a question that has always been in everyone's mind. These simulation projects, where you want to find all kinds of edge cases, are so big and complex that it's usually not a single company's job. And it's usually one of these projects that the winner takes all. It's like building an operator. For example, the reason there's only one Windows and one Mac OS One is that this is such a big and complicated thing that a startup can't just come and get a share market.

Another reason is that there is only Android and iOS now. This is such a big and complicated thing that a company can't just come and say, there are a few other players. Now, it's true that we have Linux and all that, but you get the idea. Even Sam Altman, who has started the OpenAI program, went to India. I was watching his interview.

They asked him, for example, how many people can come to the OpenAI war and become opponents of OpenAI. Sam Altman said, look, you can go and train your language model and bring it here, but the cost of servers and the scale of this is so high that no small or large company can afford to build this infrastructure that we have invested time on. For example, either Google or Microsoft, which has its own open areas, or Google can go to war. So some of these projects are so large-scale that if you want to start with a small team or a relatively large team, sometimes you think the project won't work.

You have to pay attention to this. For example, for me now, the algorithm Autonomous driving also has the same rule, you know? And Tesla, Elon Musk, for example, had a phase in which we, when we have something to believe in, to show our loyalty, for example, they open-sourced Tesla's autonomous driving software. So now it's open-source. And you, as someone who can do this, you have to write this scenario, or you can go and see what they did. You want to be one by one, you have your own use cases, you have to write the scenario, you have your own use cases, when they did this. This is not a re-do, this is not called reinventing the wheel.

You have pointed out a very good point. You are absolutely right, companies, because this is a very strategic discussion, and now there is a lot of competition on who will be the first to introduce a product that is really fully autonomous, for example, has a level 5 and who wants to introduce it to the market. And this is exactly the need for a database of all the HKC scenarios that are happening in the world. And this is the main key. The discussion of sensors has been resolved, the discussion of technologies has been resolved, the discussion of actions is a database of all these scenarios that have all these things. I myself am the product owner of exactly this product in the position I am in now.

My belief is that no company can create a database on its own. But since it's a strategic approach, at least among entrepreneurs, none of them want to share it. They say, I have my own use case, I have my own scenario, I have my own database, I don't want to share it. What's the IP? It's the IP, exactly.

We even had an AI ban, because even if we wanted to work on a large-language platform of another company, like OpenAI, they imposed restrictions on us. They weren't even ready for the data to be uploaded there. Even if I claim that we don't export or share data.

But in my opinion, when we talk about edge cases, it should cover a lot of different use cases. And besides being open source and everyone can contribute to it, I don't think there is a need for it. I know a number of companies, startups have started to create a database, a scenario catalog to be open source and others can contribute to it.

We are one of their members, but we never upload a scenario to them. To confirm what you said, OpenAI was supposed to be open source, but it's not. But there are Facebook LAMAs, and there's a high chance that OpenAI will switch to open source. Although they might claim that it's too IP-heavy. But to confirm what you said, they've reached the conclusion that If it wasn't for open source, the speed of development and creation of projects would have been very slow. I mean, if everyone worked on an open source project, it would have been much faster for us to get to where we are now, so that everyone can do their own thing.

No, that's exactly the right thing to say. The truth is, ever since I started your podcast and listened to it, I always had in mind what a good start-up idea could be. And this is exactly what I had in mind for a very long time. You need to have a database of scenarios, open source, and be able to get public contribution. Training data.

Exactly. Not just training, but how can you encourage others to contribute to this data? No, let me correct what I said. Training data, which is already out there. Training data that has been tagged by humans. Well done. Well done.

Well done. I don't want to send a series of test engineers to collect a series of scenarios. It's like Google, for example, how does it get traffic information? It gets it from the public.

And unfortunately, this technology still needs to be in a way that, in order for you to be able to collect these scenarios from all the resources that are available, the sensors that are needed for such a thing, More than a simple test that collects these scenarios in a smart way, it has very expensive equipment. A prototype that we make to be able to do this collection data is worth millions of pounds. And you can generate it with a minimum number of sensors. This is the main challenge. And if you want to extend your hand, for example, to build your own companies, for them to have a contribution, to come and put their data on your database, this is a no-go.

In general, I don't think it's going to happen. It's an interesting idea. I think it's one of the best ideas that Y Combinator and many other companies are pursuing. In general, the world is pursuing it. Not that you can create an economy for everyone, a gig economy for everyone.

One of the reasons why companies like Uber, Airbnb, Fiverr, I don't know, these projects were created, this gig economy that anyone can plug in at their own time, whenever they want to plug out, that was it. I mean, you can generate money from dead space, from dead time, and even solve a person's need. For example, Airbnb, Uber solved the problem of not having a place for hotels to fully cover. Uber solved the problem of not having a credit union. They even say that the next billion dollar idea is an idea that can help people make money. One of them could be this, because everyone has a camera, everyone has a headset, and if they can somehow capture this information, annotate it, and, for example, make the edge cases automatically, then something interesting can happen.

It's cool, but a strange thing is happening. Sam Altman himself, I would say, wrote a statement with Pena, Because what's happening is that OpenAI itself is... I mean, everyone went to OpenAI and created a dependency. They relied on them and took their projects up with them. Like the previous podcast that I had with a bunch of guys, which was in OpenAI, I mean, it was in a previous CopyData AI.

I was talking to him about copywriting, and he was talking about Jasper, one of these two products. And before he released the chat GPT, they were using OpenAI for copywriting and blogging. And when the chat GPT came out and said, you can do this with me, they went crazy.

Today, in fact, I think today is a public day, but yesterday and the day before, You can now upload a PDF file to ChatGPT. And another thing you can do is that before you had to choose only one plugin, you couldn't choose multiple plugins at the same time. So you had to either search on the web, or go to Zapier, or use one of these plugins.

We had two enhancements today. One was that you could upload a PDF file, and I could read the PDF file and summarize it for you and ask you questions. And you could choose a few plugins. And I didn't want to tell you what to do. I would go and figure out which plugins to use.

Companies and start-ups have come to you to upload PDFs and summarize them for you. OpenAI has come to you today and said, I don't think I can do it, maybe I can do it better. No, no, it's great that you said that. I really had... Yesterday and the day before yesterday, I was testing something.

The scenarios that we provide are based on an open standard. It has an XML format. Chachipiti knows these schemes.

Then I tried to tell him that I wanted a scenario. I mean, I tried to describe my scenario through those prompts. I wanted a scenario in which Khudro is in a street, let's say Fasqoun Street, there is a motorway, the traffic is like this, the condition of the water and air is like this, there is a crosswalk, the speed is like this, and let's say a crosswalk, the distance between Khudro and him is close, and he jumps in the middle of the street. You can generate this for me in this format, and it will generate it for you. I'm tearing up right now.

A lot. It's so annoying that you can't believe it. And even the opposite. You can even reverse it. Say, I'm giving you this scenario file.

For me, you can generate an English text in the form of a scenario description that is readable for a human. Because this is a challenge for us. For example, you can create a structured template that is a scenario description that is commonalized between a team of 100 people. If you give them a scenario file, they'll describe it in a certain way.

This doesn't happen. If you give them a structure and format it, it doesn't work. But if you give them a file, they'll tell you to write a scenario description for them. And that's what they do. The quality isn't ideal right now, but I see the world in that direction. Instead of having someone write an XML and create a simulation file, create a scenario file, and start explaining it.

The large language model itself will structure it. It's very consistent. For example, an Interpreter has a lot of value. Yes, exactly. You must know Andrej Karpathy, who was the Senior Director of AI at Tesla. He had a very cool podcast with Lex Friedman, and he talked about sensors, data, and computation.

We've reached the point where I'd like to talk to you about it. No, I don't know them personally, but I know what their ideas are. Their belief is that you can do a lot of things with your own binoculars, which is true. But it is very dependent on the fact that there is data.

It is very important to know how to process it and export it. There are a number of restrictions on sensors, which require a lot of technical work. The argument is based on a general way of thinking. For example, what Tesla has started, what Humanoid has built, a robot that looks like a human. Because if you want to build a robot, being human, having hands and feet, and standing on your feet is not the best form factor.

You can build the best form factor for anything. For example, if you want to wash dishes, The appearance is different. For example, if you want to make a car, it's a four-wheel drive car. For whatever you want to do, if you want to make a pizza machine, the appearance is not like a human being. Optimize it for your own application.

Well done. Optimize it for what you're going to do, and the design will probably change. If your car is autonomous, if your car doesn't have a driver, the design of the car will change. You know, the way you sit on a chair changes, the way you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit, how you sit If you want to wash the dishes, Sinkit is built for humans.

It's a place that has a height on the cabinet that you can wash the dishes. And if you want to make a robot that is about... to be a generalist, to be able to do a lot of things, you have to have a human form. Because we created the world for humans. The machine that is driving, if you want to compare it to a car, this machine, these roads, and this infrastructure, is made for humans.

Humans only use their eyes. So, if you want to use LIDAR and a number of other sensors, it may end up being harmful. Because this infrastructure is made for humans, and it only works with the eyes.

And the camera is a cheap sensor in terms of the data it gives you. LiDAR and MiDAR might give you some data, but they are expensive sensors and have advanced technology, but it's like a mirror that can see the light of the road, green, yellow, red, snow, color, distance, dimension. And you can even hear the sound.

I saw it on Facebook. We talked about it in the podcast. You can even hear the sound from the picture. For example, you can hear the sound of the train. It can simulate everything. This was one of the reasons.

Another reason was that when you are creating a production line for a big company, if you want to put all these sensors, you are actually For a company, you have to make debts. You have to have customer support, you have to have a production line, you have to have an engineering team. If you can solve all of these, you can make it more minimal, more simplified, and you have a more efficient production line. If something goes wrong, everyone is responsible. I learned a lot from this conversation.

I recommend everyone to watch the podcast. You're absolutely right. Let me tell you something. Let me give you a very simple example.

Let's say you're a driver. What kind of a situation, what kind of a scenario do you write as a dangerous scenario? It's not just a matter of perception. It's not just a matter of vision. It's a matter of sound, I don't know. Even the mimic of the driver next to you, who is driving sideways, you can understand that this side has an aggressive form, it's alert, it's not... I don't know.

Experiences that you had before, compared to the situation that you are in right now. Preview of the weather. Exactly. And all of these have an impact on the overall perception of a driver.

But when it comes to yourself, what you just said is absolutely correct. I think it is very long-term when you can have all of these experiences and all of this reality. In the current situation, because it does not exist, There are no previous experiences, you don't understand the traffic patterns and the special areas that are being explored. You can't predict the intention of that side based on the facial mimicry and the action it takes. A person can, with the information they have, be it their vision or their feelings, have a precise perception. It can be said that the decision-making of a director is ultimately skill-based, not just knowledge-based.

When you talk about a self-directed director, everything is knowledge-based. And because I don't have that skill or emotion, I try to fill the gaps with adding a lot of these sensors. But I think in the long term, this is what will happen. Because as you said, a person has these two eyes. With these two eyes, he can detect the speed of light. He can understand the distance.

And he can work day and night. So when we say, where is the weakness of the camera? Well, at night, as much as a person can see an ordinary camera, the camera may not be able to do this. Or if it eats direct sunlight, for example, The camera may not have that high performance due to current technologies. And the radar can even detect half-speed in absolute darkness.

Or have night vision, or a LIDAR. They've added some of these things, but the more you add these things, the more partners you get in the group. Maintenance, I don't know, supply chain, all of these things will increase the challenge. Entropy goes up. Exactly, exactly.

You had some slides you wanted to show us. Yes. If you don't mind, I have a few questions for myself. Without the slides. Because I've actually been talking about this for a long time, with the person I was interviewing. their daily experiences, I don't know, from the ceremonies they have, I don't know how their stand-up sessions are, and what they experience in their working environment.

Now, I wanted to say a few things about them myself, to see what you think, and what you do in your company. Do you want to share the slides first? No, definitely. Your question is about how, or do you have a more specific question? Yeah, no. But let me put it this way, let's say in our group, we use the same toolchains, I don't know if it's Teams, I'm talking about daily work, I don't know if it's Teams, I'm talking about collaboration, we use Lucida. Yes, exactly.

We use Jira a lot, and we use Lucid for collaboration. It has a good integration with Jira. It's very effective for mood working. I don't know if you know Lucid or not.

Isn't Lucid Safer for making charts? We have a software that looks like Miro. Miro, that you can use. But let me see, which one do you prefer? That one, I think. You're right.

I think Miro is the one that I prefer. The one that you can use? Yes, it gives you infinite canvas. Yes, exactly, exactly, exactly. Yes, yes, let's work on it.

Alcante, we mostly use FigJam, and now the guys at Product use FigJam a lot. The rest of our team uses Miro a lot. We use all three. Instead of spending a lot of money, you use all three of them.

Because each team has a preference. You talk about planning, for example, for the Falcon Spring, and I don't know if you have planned for any of your QBRs. What is the contribution of the whole team? Do they come together, for example, collectively? For example, do they work on the Falcon board together and plan? Look, this topic you're talking about is a topic that... You actually asked how my day-to-day life is. The thing that has occupied me the most in my life right now is that the first time we all got interested in this, you started with a car and we got into this game. We designed our school's website.

At first, we were very obsessed, and still are, with building. We learned how to build ourselves, you know? And after you progress in this field, the next step is that you're not working alone anymore, you want to work with a team. And when you bring people into the equation, you're not alone anymore.

When you say, I'm me, and I know a few travelers, and I can make this and that. At that time, your mindset changes. In the past, when you wanted to create a website, I'll give you an example, because it's understandable to create a website. You have a person in your mind, but when the team comes, you have to wireframe and be low-fidelity and bring stakeholders in the middle and communicate with everyone and all that. The phase I'm in in my life, and I'm working on it, and it's very complicated, is that when you want to create a company, Everyone thinks about building a company.

The reason I deal with all companies, whether their work is good or bad, whether their product is good or bad, whether their work is useful or not, whether they like it or not, I deal with all of them. The reason is that they have done one thing, When you go into it, you see that it's very complicated. You know? Stealing is a difficult job, too. I mean, if you want, I can even tell you that.

You know? I mean, for example, these movies that we're watching, these guys... Stealing doesn't mean that you go and do a choplifting, or you sit down and plan a program, you set a bank, you set the time, you gathered your team, I don't know, you looked at the floor plan, I don't know. All of this means that you need to have a system-designed mind. And when you bring a few people into a team, how do these people work with me? You are building the team that builds the product.

You know? My project is not a better mode right now. My project is a team that is building a better mode. So it's a very difficult job. Look, I mean... I actually... For lack of a better word, I'm reading in it.

I don't know what else to say. And what you're asking, what you're doing, I've tested all the mode

2023-11-23 07:01

Show Video

Other news