Full Transcript

Dhruv Sharma: Hi there listeners, and I hope your week is off to a great start. Look, I'm very excited about today's episode because if you, like me, grew up in the 90s watching Bob the Builder, and I do have something in my hand, it's like a scale model of a JCB Earthmover. So, if you grew up watching Bob the Builder, you'll remember that, you know, there used to be machines that would talk to each other at, you know, Bob would give them instructions, they would talk to each other at the construction site, and then the machines would just go off on their own and bring, you know, projects into existence. And our guest today, his name is Manish, his company's name is Flow Mobility, he's actually making that the cartoon that we, a lot of us grew up watching a reality. So with that, we're going to bring Manish on to the show, Manish, it's great to have you on the Offline Network.

Manesh Jain (Flo Mobility): Pleasure to be here, Dhruv, and yeah, really looking forward to discuss about that today.

Dhruv Sharma: So Manish, I, of course, gave my version of what the company does, and thank you for bearing with it. But why don't you give us a more serious, more credible version of what you guys are doing at Flow Mobility?

Manesh Jain (Flo Mobility): Yeah, so we are building robots for construction industry, because we realized that there's too much construction happening around, but availability of labor and productivity of people who are working on the site is a big challenge today. And that's why you see almost every project that is going on right now is running delayed. Some by month, some by week, some by even years, right, some of the residential projects in India are running late by years. That is what we are trying to solve to automate the activities that happen on a construction site. And interesting thing is the construction site is a very haphazard environment, you have a bunch of activities, each of those activities have a very different type of skill set required, have a very different type of workflow, which is defined for them. And hence, traditionally, the automation in this industry is not really taken on, right, you wouldn't see a lot of automation, unlike warehouses or the production manufacturing facility. This is the reason it is a very disarray arrangement, not very structured environment to automate.

Dhruv Sharma: The thing is, it's also a very dynamic environment, it changes and evolves pretty unlike a warehouse. And so it's difficult to navigate for autonomous machines and robots as well.

Utsav Somani: You want to follow the rules of like safety and I mean, I mean, just the global standards of construction, right? I think it's unsafe. Also, I think more so in India, I believe.

Manesh Jain (Flo Mobility): Absolutely. Absolutely. So if you see all the elements that we talked about, right, there's a productivity angle, there is a safety angle, there is a carbon emission angle, construction is one of the biggest carbon emitter globally. But we can't help it because we need to build as humanity, we need to build to grow progress, right? We cannot stop building, right? But at the same time, is there a way for us to address all these three things, right? Be more productive in building, build better structures, build structures which are more carbon efficient, which are more safer during construction. So if you bring all these together, that is where the value of what a robot fixer automation product can create. And that is what we are trying to solve for the world.

Utsav Somani: And your first product is FlowHauler. What does it do? What does it, I mean, what kind of data does it rely on? Like how does it do? I mean, sensing of its environment?

Manesh Jain (Flo Mobility): Sure. So before I talk about the hauler, I'll probably take a minute to say on how we actually landed to this particular product, right? So we realized that construction sites have, as I said, multiple activities, right? And we cannot automate all of them very easily. But the material movement was something that we realized is happening across the board, right? From the day you dig the first hole in the ground all the way to the last brick or last bolt is put on the door, right? So we realized that material movement is something which is what we should start with. It's a good way to understand what is happening at the construction site and what type of different activities are happening. So that's how we decided to start with material movement, even though this is not really, I would say, one of the biggest pain points. The bigger pain points are, let's say, a wall finishing or a painting or so many different more skilled activities, so to say, right? So our journey is that we start with a horizontal activity first, because that gives you a very good view of end-to-end construction stages. And then you can go deeper and start automating other pieces which are more skilled, but applicable only for a certain...

Utsav Somani: It's like a wedge that earns trust for you, basically.

Manesh Jain (Flo Mobility): Absolutely. I think that's a very good way to put it, right? That's an entry point. And then by virtue of that being a horizontal process, we also got a very good view of what type of scenarios that we can expect, right? So when the earth is being dug, what type of environment it is versus when it is 15 floors that are already built and 16 floors where you are working, what are the type of modus operandi, right? It's very different. You need to collect data for both of them. A robot needs to understand both these scenarios and also learn through this evolution when ground is dug all the way till when the last penthouse is being built, right? So this was the reason why we started with the material movement as a use case first. And our first product flow holler, as we call it, is basically to move the material, specifically for the last mile. Once the material reaches the site, it has to reach every nook and corner of the site, right? And it is highly variable. The workflows and the pathways are not really well-defined there. So that is the problem that we are going after, that something which a tractor cannot do or a DCD cannot do or a crane cannot do, which are the current methods to move material, is where our machine would come in. This process traditionally was being done by the manual wheelbarrows. You would have seen those three-wheel small equipment that people used to haul or take the material through. That is what we are implementing. So it's basically a motorized wheelbarrow flow holler, which can take half a ton of payload, navigate autonomously to different nooks and corners of a construction site, go and deliver the material and come back without any human intervention. That is the primary use case that we are trying to address using flow hollers.

Dhruv Sharma: Manish, can you help us understand what makes it particularly challenging for an autonomous vehicle to navigate a construction site? It's not the same as road traffic. It's not the same as being inside a warehouse, obstacles, ditches. So give us a sense of that. And for any engineers tuning in, we'd also love to understand the sensor suite on board the holler and then your subsequent line of products.

Manesh Jain (Flo Mobility): Sure. So firstly, I think, as I said, the nature of the environment itself. It's not a structured environment, which means you have to provision for more corner cases. You have to provision for more unintended consequences. We generally call it an autonomous language parlance. So that is what makes it difficult. But let's come to the sensor piece first, and then I'll tell you what are the other difficulties. So we use a mix of LiDAR, DTS, and camera. And why we use this mix, once I explain, you'll probably be able to understand what are the challenges. So generally, LiDAR is a laser-based sensor, which is used for making a 3D image of the environment. Generally, all autonomous cars will have LiDAR as a base sensor, except Tesla, of course, which decided to go vision first, or a camera, which is another way to perceive the world. Now, in an outdoor environment, LiDAR doesn't work because it needs features, it needs things around for it to give you a very good localization. So LiDAR doesn't work. So in an outdoor environment, DTS works well. But the moment you come indoors, DTS doesn't work, LiDAR works beautifully. So generally, other use cases you talk about, like a manufacturing location or a warehouse, they would be either indoors or outdoors. Versus construction is one such use case where you have to do both. You have to be very well running in a very confined space indoor, at the same time running with a much higher speed and bigger payload in an outdoor environment. The same machine is supposed to do that. So that, I think, is the underlying reason why it's becoming very difficult. And we have been able to track it through this redundancy in sensors. We have three different sensors. All of them are actually doing the same thing, but in a very, very different environment. LiDAR, a camera, and a DTS. Each one of them has their own territory where they are best. And then supporting when they are, let's say DTS is indoors, they're supporting. And LiDAR is outdoors, it is supporting. DTS is traveling outdoors. That is one of the big, major reasons. Another quick input there also on the challenges is that for an AI model to work on a construction site, you need data. Generally, construction globally don't have very standard data sets. If you go to a manufacturing warehouse, even road autonomous cars, a lot of standard data sets are there. Especially like Indian construction data.

Utsav Somani: I'm guessing that's probably the rarest of rarest.

Manesh Jain (Flo Mobility): Yeah, yeah. So even globally, we don't have a lot of construction sites. Indian sites are very, very unpredictable. You can expect a cow dung or a small animal moving, or you can expect a cardboard on the road, which otherwise you wouldn't see at any place. And a robot gets confused. What should I do? It's a cardboard. You can just go over it. But it doesn't know, right? So you have to provision for each of these small nuances for it to run on a unstructured site. And Indian sites that ways are the best benchmark. Once you have run it here, I think we can run it anywhere.

Utsav Somani: Can you maybe talk to us a little bit about that? Because physically, I think Dhruv and I were discussing about other startups that have come and spoken to us about physically. We've not had many on the show. But how do you build this unique data set? There was a clip that I saw on the internet where somebody's installed this camera on the workers. Like, I think thousands of workers in that factory have that camera on their head in India to collect that unique data set for, like, say, making garments in an industry which might be used later and sold off also. So how was that for you? And you started off with trying to do autonomous installations for e- scooter, right?

Manesh Jain (Flo Mobility): Right, right. So that's a different journey.

Utsav Somani: How did you land on the data and this problem? And how did you go about gathering the data and building a model for this?

Manesh Jain (Flo Mobility): Got it. So let me answer the first, second question first. And then I'll come to how we solve the data problem. So we started as an autonomous mobility company. That's why the word mobility in the name. Then we said we will build autonomous e-scooters. Very soon we realized that the market is not ready. This was early 2021, late 2021, early 2022. We realized, no, this is not something which is going to work because the market will take time to evolve. The micro mobility or the last mile mobility itself was going through a transition that time post COVID. So we kind of said, no, this is not really working. And as a startup, we need to find something where the need is a bit more urgent. So we moved to agriculture from there and spent almost a year to solve for autonomous e- scooters in agriculture. And we realized, no, again, India is not a place. So either we decide to stop everything in India and go to a different market or find a use case in India. That's how from agriculture we moved to construction. And then the moment we entered into construction, we realized that it's a massive, massive opportunity because there are so many things to be done. And the technology penetration is actually very, very big. And eventually it's a global problem. So we were very clear that we don't want to solve for them. We could have technical expansion, as they say, that I should be able to do more use cases and geographical expansion also that we should be able to go to more places. This was all my previous startup experience was playing a role in this decision. So that's how we actually ended up to construction. We were not really from construction background. In our quest to find out which industry is right for the system through robotics is where we landed into construction. And two years since we took that decision, I think we are very, very, the conviction is increasing every single day that we are at the right place at the right time.

Dhruv Sharma: Manish, before we switch to the Rita question, which is an important one, staying with construction, what's the reception from the industry been like? Who are the developers you're working with? What is typically the pitch to them? What are some things they instantly get? What are some things that are relatively hard sell?

Manesh Jain (Flo Mobility): Sure. So traditionally, the industry is a very value conscious. So you have this builder, but most of the builders will probably outsource to a general contractor. Contractors are basically our customers because they are the ones who are doing this actual work of building something. So contractors is our primary target market. And we have people like L&T, Shobha, Godrej, PST, Aluvalia, Embassy, Sriram, some 15-20 large customers in India who are now using our products. But it has been a journey, right? Because as you said, this is a very value conscious market. The biggest thing for them is efficiency from day one. It should be able to save the time.

Utsav Somani: And what kind of promises do you make there or claims do you make there?

Manesh Jain (Flo Mobility): So initially, our kind of pitch was, it has refined a lot now, but initially our pitch was, we'll help you move the material faster and we'll move the more material. So today we move 100-120 kgs in one go versus we will move 500 kgs on day one. It's like four times more material we are trying to move in. And because it's motorized, their speed also will increase. We will move more material in a lesser time. Now, after we have all the case studies and data from the market, we are able to very confidently say that it is bringing around 40-45% efficiency on cost. Whatever they were doing earlier in X amount, now they can do that in 0.5-0.55 X amount. This is a function of all the three things. This is a function of more material, more time, less material breakage, lesser number of people required to actually move around. All the things together come together in almost 40-45% efficiency.

Utsav Somani: And you are live in the UAE as well. So, I mean, how has your pitch changed or is it the same across the world, wherever you go?

Manesh Jain (Flo Mobility): Or are they looking for different things? But the pitch has definitely changed because the processes there are a bit more well-defined. So, for us, the baseline itself there is better and now we are improving on top of it. The pitch there is more on the safety aspects, the ability to do things at the right time because there the pace of construction is faster. Generally, we have seen that the speed at which UAE builds is faster than the speed that they build in India. And that is how our pitch is that because the speed of construction is faster, your downstream activity moves more material faster. Let me help you move this material faster. Versus in India, it is more on within this activity, how much are you helping in saving time. So, the pitch is slightly wider when you go to Middle East market where we are talking about end-to-end process improvement rather than just material help. Versus in India, it was very focused.

Utsav Somani: So, I mean, because you are able to move, I mean, construction industry is very unique like all the challenges that you identified. But what about warehouse robotics? I think Orange also does this very well. Have you looked at that? I mean, that is much simpler than the industry that you are working in and given the payloads that you said like 500 kgs, I think that also is pretty impressive in size.

Manesh Jain (Flo Mobility): So, two things. One is that there are quite a few players already in that industry. So, you have Grey Orange, Akebo there, Unboxed. There are quite a few Indian companies. Edward is doing good there. So, there are quite a few companies in India as well as globally. Amazon has been doing warehouse robotics for almost 10 years now. The biggest one. So, that was we couldn't really see, okay, what is it that we will do differently? Why should somebody choose slow mobility versus anybody else? It could be maybe marginal improvement in efficiency or marginal improvement in cost. That's not what we want to go after. Even though part of technology is actually transferable to warehouse use case, but we didn't really want that to be our primary use case. It could be more of sometime later we will do a later entry into the segment. But core focus should be in the industry which has, where we are the first movers basically. That was the thought, number one. Number two, because warehouses are very structured environments, the terrain is very, very flat. But in construction environment, the terrain, the environment, dust, the outdoor, all weather conditions are very different. We cannot like build for this or that. We have to build for both. In construction, because we ended up building the full stack all-terrain solution, we can go into warehouse terrain whenever we want. The other way around is not possible. If I build for warehouse today, very difficult to bring it to a agriculture or a mining or a construction. Any outdoor terrain. Fundamentally, there is a difference in terms of the way you approach. And that's one kind of caution when we talk about physical AI. Physical AI is where you are interacting with the world. You have to be very, very careful about the environment. Because one variable goes here and there, it might push you back by three months, and that could have been the time cycle. So that's the reason why we said we'll do construction, we'll do deep entry, we'll do a good job there, we'll do multiple products there, rather than trying to be another player in a warehouse market.

Dhruv Sharma: Mahesh, I have a question for you. But before that, I'm going to say you should absolutely talk to Elon, because I think he can use all the help he can get to build cities and masts, but we're not going to be able to send enough people. So in the first Starship, the first Starship should be packed full of flow mobility robots so that the robots can go and construct the first.

Manesh Jain (Flo Mobility): Absolutely. We'll go to Mars, but let's start from a slum in India first, and then we'll reach Mars.

Dhruv Sharma: The question for you though was going to be, can you paint us a picture of what a construction site, even in India, will look like in 2050?

Manesh Jain (Flo Mobility): 2050? Okay.

Utsav Somani: Maybe give us some hints about flow mobility's line-up as well, of products, wall finishing, painting, what else are you looking to take over in the next few years?

Manesh Jain (Flo Mobility): Sure, sure. So one is the 2050 is too far off. The way things are changing, we are talking about very different types of materials. So 15 years from now, buildings probably will be 3D printed. You will build them brick by brick. You will have very different types of architectural things. Your materials will be very different. So that also feels very different because this industry has not really innovated a lot on all those aspects of the material or the way we construct. We're still building brick by brick. Instead of brick, now we are plastic. But still, we are doing the same thing, right? So that is one aspect of your question, where we say that how does this industry would change? I think there are fundamental shifts that will happen in this industry, right from the way a building is architected. You will probably start architecting in a 5D image. Today, if you're already a 3D, 4D building, BIM and other CAD software, you'll start doing it in 5D. You'll start getting design based on human experience rather than the construct that we make today.

Utsav Somani: Sorry, what is...

Manesh Jain (Flo Mobility): As a human, that has to...

Utsav Somani: For my understanding, what is 5D and for our listeners also.

Manesh Jain (Flo Mobility): Yeah, for example, let's say today when you build a building, you see the building. You don't really try to assume, okay, how will the peripherals, how will the electricity wires, how will the other wires will run into that. If you try to now bring that as a 4D element in terms of how this would look, how things would move in that environment. That's a 4D element. But 5D is when you also start visualizing how would humans move in this. Today, you don't see 3D drawings of humans. You have just simple designs which are built. So more immersive designs you will be able to do when you do 5D designing. And then people will start architecting around this because the AI can now handle all that thinking and it will be able to visualize on how people will move in the structure, let's say an architecting function. How will people move into this? You can start visualizing in a much better fashion. That is what I mean by 5D. But to the other question of what will happen from now till next 15, 20 years is if we continue building the way we are, then there are a lot of other activities. So construction is probably three, four major phases. The biggest phase is when you are actually bringing the structure up. That is the biggest phase which is where the maximum number of labor is required, which is where material consumption is more, everything happens during that duration. Land rotation and finishing are towards the end, maybe 15% on either side. Bulk of work happens in the middle. There we foresee that technology will start making interventions to the extent that all of that will start getting automated. So for example, today when a slab is constructed, which is the floor is built, you have a team of 5-10 people who are pouring the cement, spreading the concrete. We are, for example, working on a product which will help builders and contractors cast the slab better by using autonomous machines, which can actually take care of leveling and take care of all the operations which happen in the time of making the floor and pouring the concrete. So we are in the process of building or automating those elements, casting the slab, wall finishing, painting. Each of these issues are very, very different. We cannot do 50 things. We have to stick to 3 or 4. But then those 3 or 4 will be the very high-valuing activities which are really making an impact on the efficiency and flexibility of the structure. While the world will keep innovating towards 3D printing or a house which is made from other elements, all that will keep happening. But for the current industry, the legacy will continue for a while and we will keep building products to automate more finishing things. Someday even laying the cables, laying the, let's say, symmetry wires, mechanical, electric, plumbing, all that would also be done by robots, which is a very skilled job today. A human needs to really visualize and see that before it is done. So that's what we hope. Industry will go through a radical shift. The way we think about structures will also go through a radical shift. A lot of structures wouldn't be rectangular anymore. You will see more round structures. They are difficult to build, but with technology and with robotics, you can build structures in any shape. The rooms need not be square or rectangular anymore. So those are the shifts that we hope to see.

Utsav Somani: Manish, this has been super nice, but I think we missed the data question as well. For us building in physical AI, I think it's very interesting to learn from you given that you're operating in a very hard industry as well, construction. That vertical seems extremely, extremely tough and kudos to you for doing that. So how can founders choose these niche white spaces, or not niche, but white spaces which have not been captured before and build unique AI data sets? Is there a technique that you've used that founders can learn from?

Manesh Jain (Flo Mobility): I think there is a lot of inter-domain dynamics in this industry. You have to find it and it's very difficult to find. You can't be ahead of it. But generally, there are a few key factors that somebody can look at. One is whether the industry is ready for that. You have to see that the industry will respond to the same question with a different question number. When you ask a question, in the case of let's just bring it to this. Let's say 10 years ago, somebody would have told a construction company that I'm working on wall conditioning and they would have said, what are you talking about? I don't know. You didn't even listen to it. You didn't even understand the question. How important is it for your customer to solve it?

Utsav Somani: I think the audio, I'm unable to gather your response properly.

Manesh Jain (Flo Mobility): Is it better now?

Utsav Somani: Much better, yeah.

Manesh Jain (Flo Mobility): Oh, okay. I don't know something happened maybe. Yes, I'll just repeat. I think for a founder to narrow down on a niche, one of the first and foremost things is how ready my customer is. And is he really asking for this? Is it a pain point for him which he wants to solve today? Or is this something which is good to have for tomorrow? So marry that pain point with a maturity of underlying technology. Whatever technology disruption that you're building, it should be mature enough to meet that use case to a certain extent. So when these two things come together, app tech maturity and a customer's adoption maturity when they come together is when the magic happens, when a startup can build a winning product. That is generally how I would put it together. Once you have these two pieces in place is when you come about, okay, what are the remaining things in terms of my underlying structure? Should I build it first? Should I take it from somewhere else and put my technology on top of this first? Those are all the nuances that you figure out. Specifically to also answer the data question that we are talking about, I think it's kind of becoming a norm now where whenever you have to automate something, the first thing to do is put sensors on people who are doing it today and try to understand how it is being done. It could be folding a cloth or putting a brick on a wall. You just understand by putting those sensors on how it is being today. In our case, it was like a different variation of that. What we did was we said, okay, let's run our machines through a remote control operation for a while. So for first five, six months, the machines are not autonomous. We had an operator who was actually running these machines and he was going along with the machine everywhere. And in that process, we collected a lot of data. Customer wouldn't pay you for collecting data. So customer is only paying you because his work is getting done. And in that process, we are able to collect data and then use that data to train our models, which eventually can now make the machine run in that same environment or fully autonomously without operation.

Utsav Somani: Awesome. Thank you so much for coming on our show. Wishing you all the best. Construction industry could use a redo.

Manesh Jain (Flo Mobility): Thanks a lot. Thanks a lot. Really looking forward and great chatting with you.

Utsav Somani: Awesome, Manish. Thank you. All right, listeners. That's it. Short and sweet from us this Monday. Wednesday with Ruben, I'm taking a break. We'll see you on Friday. Have a wonderful, wonderful week ahead. Thank you. Bye-bye.