Transcript: Scaling Sidewalk Autonomy: How Serve Robotics Is Reinventing Last-Mile Delivery

Executive Summary

In this episode of The Road to Autonomy Podcast, Grayson Brulte sits down with Ali Kashani, founder and CEO of Serve Robotics where they discussed the company’s journey from a project within Postmates to a publicly-traded company.

Ali details the massive market opportunity in last-mile delivery, explaining how their robots are designed to solve the inefficiency of using a two-ton car to deliver a two-pound burrito. He covers their expansion strategy, key partnerships with Uber and Magna, and the long-term vision of creating a diversified robotics platform for complex human environments.

Key The Road to Autonomy Episode Questions Answered

Key The Road to Autonomy Topics & Timestamps

[00:00] The Origin Story: Spinning Off from Uber/Postmates 

Serve Robotics originated as an internal project at Postmates in 2017 before being acquired by Uber. Founder Ali Kashani spun the company off in 2021 with the vision of creating an independent platform whose robots could serve every delivery service, thereby achieving the best unit economics.

[00:45] The Market Opportunity: Solving the Inefficiency of Last-Mile Delivery 

The core idea was that while autonomy wasn’t ready for roads, it was ready for smaller form-factor sidewalk robots. Ali highlights the inefficiency of using a “two ton car” for a “two pound burrito,” especially when over half of deliveries in a city like Los Angeles are for distances under 1.3 miles.

[05:30] Sidewalk Chaos: Programming Robots for Unpredictable Human Environments

Sidewalks are described as far more chaotic than roads, with unexpected obstacles like “goats on sidewalks of Los Angeles”. The robots’ strategy is to be deferential to people; if a robot is in a situation where it doesn’t know what to do, it will wait and let humans make the decision to avoid getting in the way.

[06:40] The Universal Language: How Robots Use “Body Language” to Communicate Intent 

To communicate with all pedestrians, including those who may be visually or hearing impaired, the robots use “body language”. For example, when crossing an intersection, a robot will gently “inch forward” to clearly signal its intention before accelerating, removing confusion for surrounding people and vehicles.

[07:33] Dealing with Fans and Bullies: How Robots Interact with Curious Children 

Kids are often big fans of the robots, but they can also bully them, especially in groups. The robot’s primary strategy is to “start playing dead” by turning off its lights and stopping, which typically makes the children believe they broke it, causing them to lose interest and leave.

[10:30] Market Expansion: Choosing New Cities like Atlanta and Dallas 

Serve Robotics chooses new markets based on data from partners like Uber, focusing on areas with high delivery demand and labor shortages. The overlap with markets chosen by other AV companies like Waymo is seen as a coincidence driven by these same underlying economic factors.

[11:30] Scaling the Fleet: Growing from 100 to 2,000 Robots 

The company is in the process of a massive 20x expansion of its fleet. It plans to grow from 100 robots at the end of last year to 2,000 robots by the end of the current year, supported by a large-scale agreement with Uber.

[12:30] The Power of Data: Using Postmates’ Historical Data to Design the Perfect Robot 

Before building the first robot, Serve leveraged historical Postmates delivery data to run simulations. This data was crucial for design decisions, revealing that a robot needs over 10-11 hours of battery life to capture 90% of daily demand and avoid the inefficiency of mid-day charging or battery swaps.

[15:30] Beyond Delivery: The Long-Term Vision of a Robotics Platform 

Serve is a “robotic company,” not just a delivery company. The primary goal is to build the definitive technology platform for robots operating in complex human environments. Last-mile delivery serves as the first “killer application” to develop this platform, which is now being licensed to partners like Magna and major automakers.

[19:30] Robots and Drones: Partnering with Wing for Multi-Modal Delivery 

Serve Robotics has partnered with drone delivery company Wing to create a combined logistics solution. In this system, a Serve robot picks up an order from a restaurant in a dense area, transports it to a less congested handoff point, and automatically transfers the package to a drone for longer-range delivery.

[23:15] The French Fry Test: Keeping Food Fresh During Delivery 

The robots are designed to maintain food quality. Their insulated cargo holds, combined with an average delivery time of about 18 minutes, ensure freshness. The company has successfully passed its own “pizza test” with Pizza Hut and an “ice cream test,” delivering ice cream in over 100-degree weather in Los Angeles.

[27:45] Gen 2 vs. Gen 3: The Evolution of Serve’s Robot Technology 

The new Gen 3 robot is a major upgrade, featuring more cargo space, a higher top speed of 11 mph, 14 hours of battery life, and five times more computing power than the Gen 2. Despite these significant improvements, the Gen 3 robot costs only one-third as much to manufacture.

[34:30] Diversifying Revenue: Building an Advertising and Data Licensing Business 

Serve Robotics is building multiple revenue streams beyond delivery fees. This includes an advertising business where brands place their logos on the robots, a software licensing business for its navigation platform, and future opportunities to monetize the valuable visual and lidar data the fleet collects.

[39:20] The Future of Mobility: How Sidewalk Robots Will “Unbundle the Car” 

Ali predicts that sidewalk robots will evolve from a novelty to a common piece of city infrastructure, like a mailbox. He views this technology as one of the first new mobility form factors of the 21st century, which will help “unbundle the car” by taking over short-distance errands, ultimately reducing traffic, saving lives, and lowering carbon emissions.

Full Episode Transcript

Grayson Brulte: Ali, you founded Serve Robotics in March of 2021 quite a while ago. Now you’re thriving, you’re public. What did you see in the market then? 

Ali Kashani: Yeah, thanks for having me. First of all. Um, you know, in, uh, 2021, we were actually a, a team inside Postmates, and, and at the time, uh, Uber had acquired Postmates. So really we were inside Uber. And, uh, I had this idea that we should actually spin off be an independent company because the robots for your building should be serving everybody, every, every partner, every, um, delivery platform, because that’s, you get the best unit economics. But if I go a little bit back further, 2017 when we actually started the effort inside Postmates. It was actually a simple idea. I have a PhD in robotics. I’ve worked on sensors like lidar. , We applied them in mining application. We were in DARPA Grand Challenge, and I didn’t believe in 2017 that the technology was quite ready for the roads. But it was ready for a smaller form factor. But what we have, these robots are, you know, about the size of a shopping cart. They have 3000 times less kinetic energy than a car. And when I joined Postmates, I could see that the, um, number of deliveries that happen, even in a fairly sparse country like the US that are just a couple of miles is actually the majority of them. So more than 50% in a city like la, half of the deliveries are only less than at 1.3 miles. So you can solve basically majority of last mile with a more appropriate form factor, as we like to say. Why move a two pound burrito in a two ton car that just doesn’t make sense anyway. So you can solve the majority of last mile in a smaller form factor with a technology that exists today and start to bring, , some level of automation to last mite that has lacked for so long. So it seems like the perfect place to get started with applying autonomy, uh, in a commercial way.

Grayson Brulte: In la if you look at the 1.3 mile, I’m gonna zone in here where I used to. Live and I co-chair the Autonomous Vehicle Task Force with the city of Beverly Hills. There’s a lot of restaurants in the Golden Triangle. There’s hotels there, there’s apartments, there’s condos. Is it delivery food that you’re seeing used the most? Is it, is it takeout? Is it perhaps a small delivery from Whole Foods? What are you seeing being delivered? 

Ali Kashani: Well, food delivery is the most popular because we eat three times a day. I think after that you have things like, , groceries and, and maybe pharmacy after that. , And then of course parcels. Uh, so, but, but the nice thing about food delivery, what makes this Nice for us as a first application, but also, uh, fairly inefficient in general is the fact that you need the food right away. So you don’t have a ton of opportunities to batch items together. Uh, you want your food in half an hour and, and that means that I. Most of the time there’s maybe one, maybe two items that someone can grab, , before they head. Also, restaurants are usually in the busiest parts of town where parking is often challenging and you know, there’s not a lot, a lot of real estate you can dedicate, for this sort of thing. And homes are usually. Kind of in surrounding areas, most of the density, that’s how we design our cities. And then you get into the suburb, et cetera. But most of the density, most of the demand is actually structured really nicely in places where you don’t want to be on the street because of congestion usually. , And you actually can access the demand by having a robot on a sidewalk moving at a few miles an hour.

Grayson Brulte: Have you ever seen a scenario of your a restaurant? Okay. You don’t want to pay for the valet. There’s nowhere to park. Oh, okay. I’ll park three, four blocks away and have the robot deliver to me where there’s no traffic, no congestion. Have you seen a situation like that occur yet? 

Ali Kashani: We have actually thought about this because again, we were inside Postmates and then Uber we, we had the opportunity to really think what the new system needs to look like. I. In a long enough time horizon, that’s an entire possible, uh, approach because, , you know, drivers. Can be actually supported. If you are going a few miles and a robot can’t make it, it might, it might make more sense to have one or maybe more than one robot picking up multiple items, bringing it to a central location where the drivers can wait and they don’t have to work, uh, wait for parking, et cetera. They can be maybe on a side road or an a, a dedicated parking. And now they have multiple items to deliver so they actually become more efficient. Uh, I think from a maturity point of view, it’s still early days. There’s still a lot of short distance deliveries for us to do, but over, over the longer time horizon, I would absolutely expect this kind of, uh, coordinated, uh, deliveries.

Grayson Brulte: Today you’re operating on the sidewalks. Do you eventually see a, perhaps where you’re going on the side, the side of the road, say. A foot off the curb, a half a foot off the curb.

Ali Kashani: Yeah, absolutely. I think it’s, uh, in a, in a, again, uh, in a longer time horizon. Those are all, uh, parts of the expansion plan. We call them kind of road margin applications. Uh, sometimes it’s because you wanna move faster. Sometimes it may be that there is no sidewalk where you’re going, even though. Most places where we care to operate right now have them, but sometimes they don’t. And the form factor we have right now, if you look at the robot, is actually designed to be able to handle that. So it can go like 11 miles an hour. , So if you are on the side of the road, you can actually operate there. And, uh, it’s big enough to be visible, but small enough that it can operate on a sidewalk.

Grayson Brulte: When you’re operating on a sidewalk, how are you managing the rules of a sidewalk? There’s a lot of custom rules that are not put in any book that you’ll find anywhere just customary that, you know, how do you program the robots for those, those habits, the way that individuals walk on sidewalks? 

Ali Kashani: These are, these are some of my favorite problems actually, because, um. You think about the roads, uh, it’s challenging in a way that you have a much bigger vehicle moving a lot faster. So a lot more, again, kinetic energy, low margin for error. Sidewalks are way more chaotic though, so there’s, there’s a lot more going on. Uh, a lot of unexpected things. We’ve dealt with things like goats on sidewalks of Los Angeles. So, um, with that, I think part of it is just having, um, a lot of difference to people. Because people generally know how to navigate. If we are in a situation where we don’t know what to do, we should just make sure that we are not getting in people’s ways and we let them make the decision. So, uh, that has actually been a very good strategy in general. Uh, another one is you need language to communicate with other road users . and in this case it can’t be sound. People may not hear it. It can’t be, uh, visuals. They may not be able to see it. Some people may be visually impaired or hearing impaired. What we found to be probably the most effective is the most universal language, which is body language. So the way the robots act and move can actually communicate the intentions and remove a lot of the confusion. I’ll give you one example. When a robot is, uh, trying to cross the intersection. It actually starts by almost stepping in very gently. Like, it, it, it kind of wants to imply that intention that, Hey guys, I’m about to start crossing. But it doesn’t go like full speed right away. It actually, uh, you know, inches forward so that everybody understands the intention and then it, it starts to accelerate and, and really commit to it. So those are the small things you do that, that are actually really important in not really damaging that harmony that already exists to the extent that it, it exists on, on sidewalks.

Grayson Brulte: On sidewalks, I could see a scenario where you have an 11-year-old child, this is cool. I wanna take a selfie with the robot. How do you deal with those types of traffic jams that that could occur? 

Ali Kashani: we deal with them all the time. Actually. Kids are, are big fans. I’ve seen people, uh, pull over in their cars and get out with their kids to go take a picture with a, with a robot. Uh, honestly, most of it is very harmless. They just take their picture. I love it. I see them on social media. It actually, I’ve, I’ve had cases where parents reached out to us and wanted to go see one of our depots, where, you know, their kids just were obsessed with the robots in, in all the right ways. And then there are times where actually there’s research on this where kids can bully robots, especially if they’re in groups. Um, so the, the basic, you know, SOP for us, which is based on that research is. At one point you start playing dead. So turn off the lights. Just, just stop. Generally, robots are more patient than humans, so you just let them kind of get their curiosity out of the way and they, they very quickly get bored. But especially with kids, if they feel like they broke it, they’re more likely to just, uh, go away. And then sometimes the other strategy is go close to an adult nearby. And that, that usually helps. But it’s really interesting. These are the kind of problems that we actually do, do have to deal with.

Grayson Brulte: The kids run away and hope the robot doesn’t follow ’em or an adult and seems they don’t get in trouble. It, it solves your problem there. I wanna zone in here in LA ’cause I lived there half my life in, in Beverly Hills. And you’re operating in la. Where are you operating in LA And can you cross sunset? Can you cross Wilshire? Can you cross olympic? 

Ali Kashani: Oh, absolutely. Yeah. I mean, look, most of those, uh, again, business districts have, uh, signals and, and, you know, uh, pedestrian lights. So we are, we are, we are able to basically be just like any other pedestrian there. We are in Hollywood, west Hollywood, Koreatown, Glendale, I think Long Beach, downtown la. It’s, it’s expanded a lot. And because I don’t personally live in la I don’t always know the, uh, every, the name of every neighborhood, but, there is a pretty extensive operation right now. Uh, you know, I think hundreds of robots now are in, are in LA alone, and then we have Miami. We have Dallas. You’re going to Atlanta soon. And yeah, the robots have been around for years. Actually. In fact, the most surprising thing about LA is people are so used to the robots that they don’t react. It’s like when I’m visiting town and to me it’s still like, oh my God, look, there’s, there’s a robot just moving by. And most people are just on their phones and passing by as if they just saw a mailbox. So that’s kind of the, uh, stages of maturity starts with that novelty, but eventually it gets to the mailbox stage.

Grayson Brulte: I guess that Rick Rubin wrote a great book last year. It was about Buddhism, and he said, if you just remove yourself from your surroundings and you actually sit in your backyard and you listen. You, you hear things and if you, if you walk around Los Angeles or San Francisco or Miami and you don’t stare at a phone, you don’t have headphones in, you actually see things and learn new things. And when I look at the markets that you’ve publicly announced that you’re either operating in or you’re expanding to, there’s a trend here. They’re all autonomous vehicle markets. Is that by coincidence, was that planned? was that coming about? 

Ali Kashani: Actually, you know, it is like we are not sitting there looking at anybody else. In fact, I think Atlanta was the last announcement that I have to say. I was a little bit, surprised by. So we announced it, I think a few weeks later Waymo announced it and I don’t think anyone is looking at anybody else. The fact is, I. We were working on it for a few weeks, and I’m sure they were working on it for, for, uh, you know, as long, if not longer. And we just happened to find the same markets. Uh, compelling. I think there are some underlying, um, patterns here. For example, we talked to our partners in both cases we are working with Uber, for example. We wanna know where they have a lot of demand. We also wanna know where they have more shortage of labor, their cost inefficiencies, like all, all that. , So I think those may be the driving factors. The truth is there’s just a lot of places you can put these robots for the foreseeable future, I believe, via robot constraint. And, and, uh, there’s more, more even we could just, uh, you know, put all of our robots in LA and there’s enough demand. Uh, we, I don’t think we have really touched on this much, but we are launching 2000 robots. By the end of this year, we had a hundred by the end of last year, so we are 20 x-ing our, our fleet size. , Today we are already at more than 300 robots. In the first quarter, we launched 250 additional. In Q3, we are launching 700 more, and we could have put all of them in Los Angeles. We are expanding because we wanna go to new places. We wanna learn, uh, you know, new markets. We wanna develop our playbook for growth. But the reality is there’s just a ton of demand for this. Uh, so it’s interesting when, when you see the companies end up in the same cities and it’s probably because of those underlying like labor and demands, factors.

Grayson Brulte: You’re taking the Waymo playbook, Waymo’s Con constrained in markets now ’cause they don’t have enough vehicles and you clearly don’t have enough robots yet, but you’re learning in multiple markets. When you go into markets. Uber’s a, an investor of yours, they’re also a partner through the Uber Eats division. What kind of data insight is Uber sharing with you for where the demand is for the best places to deploy your robots? 

Ali Kashani: Yeah, we are, we are very aligned in terms of our incentives to put the robots in the right place that gets the most utilization. Um, so, uh, we are very collaborative in that sense. I can tell you before we spun out, uh, when we were inside Postmates and Data Uber, I actually had access to historical data of like, say Postmates database around the country yesterday. One of the first things I did before the very first Serve robot was launched was actually build a tool that would download that information at any mo moment and simulate, uh, robots at different geography. So I could literally say, Hey, put 20 robots in this location, and then feed the historical data into it and see what the robots would do if, if they were actually operating there yesterday. This informs so many of our decisions. We’ve actually been unique in the sense that we knew what delivery looked like better than anybody. Before we even built the first robot, there were decisions like how much battery should the robot have? I think generally our industry decided, you know, a few hours would be fine. You can charge the BA robots in the middle and maybe you swap the batteries. We ran the numbers and immediately knew that that’s not very good for your unit economics. You don’t want a robot that’s sitting idle halfway through the day. You don’t want a human involved in that process of battery swapping. So we actually build robots that can go all day. Our robots can go like 14 hours in a day. Now the other question is, what is an all day robot? Do you need to go 14 or 20 hours? Or 10 is enough? Turns out anything more than 10, 11 hours. Captures 90% of the demand. Again, something we knew because we had the, you know, raw data to begin with. So we found that information is really, really helpful. So we’ve been able to like focus on the product and know what it needs to be and not have to change our mind and kind of iterate on some of the fundamentals.

Grayson Brulte: Is it pattern similar to a restaurant? You have your breakfast rush or your lunch rush or dinner rush, and then you have your. You’re after dinner rush, is it a very similar pattern to what you’re seeing? 

Ali Kashani: Pretty much, I mean the, you’ll have, you have restaurants that either are serving lunch, dinner or breakfast, and then, and then there are ones that just go throughout the day, like coffee shops and, and you know, desserts and other, other stuff, ice creams, that sort of thing. So. We, we try to have a diversity so that the robots are utilized, uh, you know, all, all, all, all, all the hours of the day basically.

Grayson Brulte: Then that raises a question, how do you deploy your depot? You strategically put them around the densest parts of where you know most of the are. And occur. How do you pick your depots? 

Ali Kashani: Usually each depot services multiple neighborhoods, so it’s kind of a. Off the beaten path, but not by far. So we don’t have to be on a major street that, where real estate is really expensive, but we need to be within a couple of miles of a few neighborhoods where, where there’s a high business density. And then in the morning the robots basically go out, uh, by themselves. They can go to, to the neighborhoods nearby, and then at night they can return home where they get charged, they get cleaned, they get maintained, uh, and, and get ready for the next day.

Grayson Brulte: From a depot perspective, I’m assuming you’re operating a very simple depot. You just, you just have a locked garage because the robots are valuable and you have charging infrastructure. Is that pretty much it? 

Ali Kashani: That’s right. That’s pretty much it. And, and some internet infrastructure. We get the data, of course, we use the data for training, et cetera, so when the robots come home, we can offload the data as well.

Grayson Brulte: So on the data standpoint, in your, in your Q1 earnings call yet, you published a, supplement and you, you have a licensing partnership with Magna, what are you building with Magna? 

Ali Kashani: Yeah, this, this actually speaks to a bigger picture, uh, question we haven’t touched on, which is we are not a delivery company. We are not a. Even a robotic delivery company. We are a robotic company, and what we are trying to do is build robots that can navigate complex human environments. So you have, you know, huge success in warehouse robotics. This has been going for a while, that those have scaled, but those environments are really controlled. So the technology is very different when you leave the warehouse. And then there’s the other end of the spectrum, which is the road. So the self-driving cars, again, really different, really unique environments. And we are not going after either of those two, but we are going. Basically for everything in between, which is places where humans and robots are gonna interact. They’re gonna coexist in the same spaces, which by the way, is most of the spaces, uh, now food delivery happens to be a really good place to get started. Uh, in fact, last Mile broadly is a really good place to get started because, , it’s very inefficient today. Just to give you a sense, if you order something from China right now to your front door, it costs about $2. But if you order it from Chinatown, it costs $10. So it’s a very, very, you know, ideal place to, to go bring efficiency. And again, starting with the food vertical and you can expand later. But once you solve last mile and you have a technology that works there. It can do a lot of other things. I’ll give you an example. Walmart asks us to do a pilot where we would pick up things from their back office and bring it to the parking lot. This is for people who are ordering pickup, uh, and they’re waiting in their car. Now, Walmart has pickers. They, they grab stuff and put it in the back. The same pickers had to spend more time walking than picking, so they just go to the parking, hand it over, come back. Well, a robot can’t do that. And we showed up there with the same robot that was working on sidewalks. But with a 99% autonomous in 48 hours on a Walmart parking lot, it’s a similar space. You have the cars, you have the kids, you have the pets. It, it is all very similar but in a, in a different application. So our goal is to build the best platform. For robots that can navigate complex human environments and we can’t make every form factor of those robots. We should be the one that basically has the killer application using which we build the platform itself, the underlying platform, and then we can give access to other folks. Magna is one of the first where they have access to our platform and they’re using various pieces of it. Uh, there’s actually, I mentioned in our recent quarter call that. We have a, another partner now, a European major automaker that’s, that’s doing the same. Uh, there are a couple others we haven’t announced or named them yet, but we would in the future. So that’s the long term play for us as we build a platform, open it up to more folks.

Grayson Brulte: As you build the. Platform. Now I’m gonna give you a real world example. Daughter goes to school, forgets, you know, a book, textbook, or binder. Oh, I gotta get in the car and go, it’s two miles down the road. I would love to have a personally owned serve robot that I can put the book in there, send it off to school . has world-class manufacturing capabilities. You’re building out a world class robotics data set . at some point, will you sell me or I can subscribe to a personal robot.

Ali Kashani: what you’re pointing out is right now most things are moving from a central location to the home. They’re not moving between homes as easily or even reverse logistic of sending something back is kind of a nightmare. So you bring that $10 cost that I mentioned down, you can do a lot of other interesting things. In fact, when people talk about the TAM for last month, I think it’s infinite. I. Because it’s 10 bucks right now and we are using it so much. If you bring that down to a dollar or honestly because of advertising we talked about, you may be able to bring that cost down to zero because advertising and other monetization sources could actually pay for the robots operating, uh, out there. So if you bring that cost down, I can imagine people sending each other stuff and you know, when you forget something. And if it’s close, the robots would do it if it’s longer. Maybe it’s a combination of robots and a car or a drone, which by the way, we haven’t talked to, we have a partnership on the drone side as well, so I absolutely see us doing things very differently as this cost comes down.

Grayson Brulte: It’ll lock so many opportunities. If you wanna send something to a neighbor’s house, you have to get in your car lot’s. A it’s a, it’s a waste of space. It’s a waste of time. You could put it in one of your robots and send there. Is that drone partnership, is that with Wing? 

Ali Kashani: That’s right,

Grayson Brulte: it was really cool. Uh, Mr. Perot, junior Ross Perot hosted me at his ranch at Hillwood, and I got to experience a wing drone in person. That was cool and that that works. So how does the integration with Wing work? Do you, do you take your robot ticket to the, let’s call it the launch platform and then away it goes and then it, then it eventually comes out to me.

Ali Kashani: Actually we, we’ve posted a video of this. It, I would tell you the only thing cooler than robots or drones is robots and drones. And, uh, we have a video of how this works. Basically, the, one of the challenges, which is true actually for self-driving cars and drones. Getting to the source in, in busy environments, like again, where, where restaurants usually are, you don’t have a ton of real estate you can dedicate to a drone or a car. In fact, sometimes you don’t own any of that real estate, uh, around you. And, and even if you did, there’s also the traffic. There is the, you know, people around. So they may not like drones coming and going. So what we actually do is we can go to any restaurant, like right now, we cover 1500 plus restaurants in areas we are, and we don’t need any changes to the environment. We don’t need anything special for us. We, we just incorporate, we, we just basically get added to the existing environment. And when we pick up the order from the restaurant, we can go a couple of blocks away to a dedicated parking location where there’s a, basically a handoff that happens. Between our robots and the drone, and you can see a video of that is fully automated and basically our robots gives the package to the drone, which can now go five miles in, in minutes and and deliver that to any home. So this is really cool because we can do the short distance s entirely, and then we can also be part of the long distance database and help this technology get to more homes faster.

Grayson Brulte: A restaurant, the individual in the kitchen, they package the food. Do they go to a certain door to put into the robot? Is there a robot loading zone? What? What does that look like from a practical standpoint? 

Ali Kashani: Yeah, we can get all the way to the front door. Uh, often restaurants tell us where they want the robot to wait. They actually really like the robots. They have negative experiences when, uh, you know, the most drivers and, and, and dashers are great people, but every now and then there could be one person who, you know, is, is less than pleasant and, and, and you know, that, that kind of unfortunately happens every day. So,, with robots, they can actually tell us what to do. We will go, wait where they ask us to. All they have to do is literally step out, open the lid, put the item in, and then the robot kind of secures that. It locks and it gets on, gets going.

Grayson Brulte: Now on the other end of that equation, there’s a customer that ordered that meal . do they select the the option to have it delivered by a robot instead of by a traditional driver? 

Ali Kashani: Well today actually, they, they get assigned to it. So if you are in, let’s say Miami right now in, in Brickell, and you order from one of the restaurants we are, uh, delivering for, which as I said like in total is more than 1500 today. You would receive a notification after the order is done that it’s been assigned to a robot. You, you do get a heads up. Before that, it might, uh, be the case that this restaurant is part of the program and you do have an option to opt out, but by default you are able to get assigned. And then when you, uh, receive your order, you press a button, you go downstairs to your robot, basically you press a button on the, uh, on the Uber apps, uh, screen, and it unlocks the lid of the robot and you can grab your item.

Grayson Brulte: Give you a simple analogy. Uber’s the car key essentially, that allows this. To function, unlock. But I gotta give a, a shout out to a very, a very, very dear friend of mine, Mr. Roger Webb. And he, he is a, he calls himself a student in the restaurant industry. His whole career worked in the restaurant industry and every time he evaluates a new technology in the restaurant or the the QSR or the quick service restaurant industry, he says, can it pass the french fry test? And Mr. Webb’s whole, the French fry test says if I’m gonna order a hamburger, french fries. The french fries get delivered. They have to be crispy. So can your robots pass Mr. Webb’s french fry test? Will they, will they come crispy? 

Ali Kashani: I believe so I, we have our own version of this, which is the, I would say there are two actually. The pizza test and the ice cream test. So we do ice cream deliveries in a hundred plus degree, you know, Los Angeles summers. , Our average delivery time from pickup to drop off is about 18 minutes. So the ice cream stays in in good shape by the time we are at the customer location. \ Keep in mind that the cargo is, is, uh, insulated so that that helps, uh, with the temperature. , The other one is pizza and there is actually, uh, quite an obsession with temperature when it comes to pizza delivery. You really wanna keep it in the bag and keep it in top shape by the time it gets there. So we actually did a pilot with Pizza Hut in, uh, Vancouver, uh, a couple years ago. And they, uh, would measure the temperature from the beginning to the end of the journey to make sure that the temperature is maintained. And again, we could meet, uh, if not exceed the, the human delivery quality. , But there are more opportunities here, actually, I think in a longer timeframe. Uh, we could even add cooling and heating to the robots. We right now, as I said, we have 14 hours worth of battery life and I think there’s still more optimizations to be done. Uh, that’s kind of more hours than we need to operate anyway. As I said, like 10 hours gets, gets you 90% of the demand. So eventually we can even put some of that battery capacity towards having maybe solid state heating or something inside the bin so that we can really exceed the experience you get by, uh, you know, today’s kind of standards.

Grayson Brulte: Indirectly answer correctly. You get it. It’s all about packaging. Packaging is what’s gonna allow the french fries to be crispy. The ice cream to be cold. The pizza to be warm, Yes. We don’t want the coffee to spill. So do you engineer a special thing for coffee? ’cause I remember years ago when you were at Postmates, I got to go in the prototype car that you worked on with Ford for delivery and there’s all these different ways to engineer and there was a pizza one to hold the packaging. If I order, say a Starbucks, does it come to what’s not spilled all over the place? 

Ali Kashani: Yeah, there’s two sides to this. So there’s the packaging on the merchant side, and then there’s the packaging on, uh, or, or the, the cargo design on our side. So one of the nice things about being part of Postmates when we started this, uh, project, I. We actually had access to all the merchants and, and partners. So we actually involved them in the process. I remember when we used to make cardboard box version of our cargo design and actually go to LA and put it in front of folks and see how they interact with it. So one of the lessons, which is I think the same thing people learned about minivans was put a lot of couplers in there. It always helps, uh, but yes, we basically have the design so that it can bring the items in as good. A shape as possible. It does make a big difference on what the container is to begin with. Uh, there are cases where the drink is not very secure and that generally makes it challenging, whether it’s a car or a bike or a robot. And there are times, and I’ve seen, uh, Starbucks recently make some upgrades at least, where, where I am. And man, the, the packaging is tight. So they have a sticker on top that’s really, really, uh, strong and they, the, the kind of, the size of the lid is really well, uh, sealed. So if you have that, in addition to having a cup holder and enough kind of thoughtful design in the robots, you’re gonna have really good quality.

Grayson Brulte: You have really good quality. Um, I’m sensing here a, a collaborative nature. Serve robotics right now. You are operating two models. You have the gen two and the gen three, which is now going into service. How much input do you take from your customers as you develop new generations of your robots? 

Ali Kashani: Oh, a lot. It’s, it’s, I mean, customers, merchants, and the delivery platforms are, are the three stakeholders here. Uh, and then the fourth one would be the pedestrians, folks around the robots. So we are constantly trying to understand where are the opportunities for us to make the robots better for? I’ll give you an example. We were able to fit extra large pizzas in the Gen two robots. But getting the, uh, pizza in and out was actually challenging. It was just your fingers getting in there was, was more challenging. So we actually expanded the cargo space. Now you can put an extra large Domino’s pizza in there and very comfortably getting in, get it in and out just as an, you know, one of the examples, uh, there’s a lot of, of, uh, you know, feedback that we try to, uh, incorporate into the every iteration of the robot. Now, we’ve been around since 2017, so we’ve learned a lot of things. From the early days that are already incorporated, but as you scale, you always learn new stuff.

Grayson Brulte: It’s a better customer experience. How about from a technical perspective? Are there any main differences between gen two and gen three? 

Ali Kashani: Oh, yes. Uh, we, it was a really big, uh, upgrade. We have, uh, more cargo capacity. We have a faster, uh, speeds. Uh, I think that all the robots could do like 6.7 miles an hour. Top speed. This one can do 11, uh, more battery capacity. That one was maybe less than 10 hours. This is 14 hours. Uh, but also it has more compute. I think five x more compute. It has better sensors. Uh, and then to top it all off, it actually has, , the cost of one third of the cost of the gen two robots. So for basically one gen two robot, we could make three gen three robots. Despite all these improvements, , that were made to the, to the specifications.

Grayson Brulte: Is that because of the, the quantity and the scale, or is there, or is there something else with a big breakthrough that? caused that? I would say dramatic . dramatic . price drop.

Ali Kashani: You know, there’s a lot of, uh, tailwinds, here. So first of all, technology gets cheaper. So compute, for example, we could basically pay the same cost but get five times more compute or, or sensors. So some things are just getting cheaper by themselves if we don’t do anything at all. Other things are related to manufacturing. The way we are manufacturing at a larger scale. We can reduce costs, of course buying, uh, items in, in higher order volume helps. Tightening or supply chain finding, vendors that are cheaper helps. Uh, and then of course design. We are doing all of this at the same time. So there are parts in the robot that we have redesigned. There are parts where we, uh, had a off the shelf component that we are now designing in-house to remove that. The, the extra margin basically. And, and by the way, every, there are gonna be other incremental iterations of the robots where we continue to bring the cost down. So initially I think we cut the cost in half, and then we came out and announced that we’ve done another 30% on top of that. So overall, it’s gonna come down to about one third now. So that process is gonna continue. It’s again, a combination of tech getting cheaper, uh, our scale getting larger, and our design getting better.

Grayson Brulte: Which is a positive what your investors. Want to hear, and from you as an executive standpoint, it’s gonna allow you to scale and to go new markets, then backfill where you need capacity. When you take possession of the robot for Magna, what type of testing has to be done before you can deploy? And I asked that because Magna came out in a press release. Last week and said it’s 30 minutes from the time they gave Waymo the vehicle to the time it’s picking up a passenger. I was like, wow. Okay. That’s impressive. Obviously you’re operating on a lot smaller scale, but you have a great partner as Waymo does in Magna what has to be done before your bots go into service.

Ali Kashani: Yeah. You know, I, I am, uh, really excited about that partnership because there are few companies that know how to build vehicles better than Magna. We don’t need to reinvent a wheel. We don’t need to build our own devices when there are others who can do it, uh, even better. So we do a lot of the engineering and design work to decide what the robot is and, and all that. But uh, when it comes to the manufacturing, we have the right partner who can scale this. So I don’t have any concerns around whether or not we can scale as a result of that. And one of the things that Magna is really good at is testing to make sure that everything is working as intended. Uh, so there’s testing that happens on their facilities. And then when we receive the robots. If there’s a new generation, if we made changes or even if the software has changed, we have a very thorough testing process ourselves. So every time we do a software upgrade, for example, we do test in simulations. We do test in historical data that we have. Then we do test in our lab where we have fake sidewalk areas. I. We basically simulate scenarios there. , And fun little story that I, that I wasn’t allowed to share publicly in the past, but I’m finally gonna say it. We actually once had a robot escape our testing facility a few years ago, and I have a security camera footage of a robot that just left because it, it was programmed to seek the sidewalk. It was, uh, on a fake sidewalk, and it was just keep driving because we wanted to put miles on it. And then it actually saw the actual sidewalk outside and, and took off, and we found out two blocks later that the robot is missing. So, uh, it, it would be a good social media post one day, but we do all that testing. And then once we are done with that, we go on the sidewalk with a small fleet and test it on a real sidewalk. We more human supervision. And then eventually when the statistics kind of come in and we look at the data and we are happy with the outcome, we roll it out to the rest of the robots.

Grayson Brulte: What you described, their runaway robot sounds like a Pixar movie. Maybe it’s a spinoff of cars.

Ali Kashani: It. It’s really good. It actually gets to the edge of the sidewalk, like at the parking gate, and it pauses. You feel like it’s looking around and then it takes off. It’s just just like picture perfect.

Grayson Brulte: which is great. And I’ve always thought that your robots have resemblance to Pixar, which I think is great ’cause it builds public trust. ’cause it has that that cuteness factor to it. When you’re considering bringing on a new market, let’s use exam Atlanta for an example. ’cause. That’s a new market coming on. How do you determine the initial ODD of where the robots are gonna be deployed? 

Ali Kashani: Good question. I, I, you know, look, a lot of it, uh, starts with the conversation we had with our partners, whether the city or the, you know, neighborhood in the city. We wanna see where the demand makes the most sense, where the restaurants are. And then of course we look at robot capabilities. We have our own ODD definitions. Uh, for example, we had generations of robot in the past that didn’t have, uh, mechanical breaks. They had electrical braking. What that basically meant is you apply current to the motors to stop the robot or to keep it stopped. That wasn’t ideal for ha neighborhoods because what if a robot runs outta battery, then you don’t have a brake. , So in the, you know, gen two robots, we started adding, uh, mechanical brake, and now in the gen three as well. If a robot runs outta battery, it breaks. So that means our ODD now includes hilly neighborhoods. Uh, so we look at those kind of specifications and decide if there’s anything that stops us from going to a neighborhood. , So we exclude that. But otherwise, generally the robots have pretty good range and, uh, most of the decisions end up being about where are the businesses, where the customers.

Grayson Brulte: how about advertising? the high rent district, the low rent district, does advertising play any. A defining factor of where you’re gonna initially deploy.

Ali Kashani: You know, I don’t believe we have ever, uh, , had that as a criteria in the ODD generally speaking, again, wherever we go, it, the busy environments with a lot of people around. And, so it hasn’t, it hasn’t been a factor yet. I. , But generally we see advertising as a really solid, uh, opportunity to bring the cost of delivery down because we don’t, we don’t just reduce the cost of, uh, delivery itself. We increase the revenue sources, which means the cost that’s passed on to the customers down the line can actually be reduced. Uh, but I haven’t seen that impact our ODD decisions.

Grayson Brulte: On the advertising front, advertising has become a very quiet and very big juggernaut inside of Amazon . over quarter when Uber reports. Their advertising business is growing dramatically. Lyft is now building an advertising business. Will you follow that same path of as you build your advertising business? 

Ali Kashani: I never thought I would get a PhD in robotics and sell ads, but, uh, I think the world just kind of takes you there. We, when we were inside Postmates, we had, uh, one of the most popular merchants in in La Pink dot reach out to us and say, Hey, uh, can I put my brand on the robots? And I. Honestly, that gave us a bit of a heart pain. At first. I, I, I, I wanted to try it, but I went to Postmates leadership and there was pushback. He’s like, what about the other merchants? Are they gonna be upset? And then we decided to experiment. Just, you know, put a few robots out. See, see what the reaction is. It turned out to be really positive. In fact, we had merchants that were maybe competitive to Pink Dot, which is a, you know, convenience goods, uh, store. And they reached out to us and actually wanted to be part of the program. So we found the reception to be really positive. Now, the whole industry has adopted this model. So sometimes you just get dragged there when you see the demand. And I, again, I, I found that in a low margin industry, low revenue, low margin, which is delivery historically, is really nice. If you can introduce new revenue sources. And advertising is one of the first ones we talked about our platform that would be a future opportunity. Also, the data that the robots collect, we collect more data per day. In fact, four times more data per images per day than, , GPT four’s vision model was trained on. So there’s tremendous value on that. And it’s not just images, it’s paired with lidar. It has the robots kind of information and actions, uh, included. So there’s tremendous amount of value in that. And if we can really take advantage of all this, again, we can bring the cost of delivery that’s passed to consumers over time, , to be a lot less from that $10 it is today to a dollar or maybe one day, you know, even less.

Grayson Brulte: You have a, a growing advertising business and let’s, let’s look at that. You got alphabet’s, Google and you, and you got meta with Facebook and Instagram advertising juggernauts . the data licensing becomes interesting. You hear it on the earnings call, earnings call after earnings call Reddit. What’s your licensing strategy? They’ve, they’ve got a juggernaut of data under there and you have something very similar. So what you’re telling me is that you could potentially have a deliver a robot delivery business, an advertising business, and a licensing business. And so now all of a suddenly, now you’re a highly diversified autonomy business.

Ali Kashani: Yeah, I mean if you look at our revenue right now, I think like last year maybe half was delivery fleets related. And even there, there was advertising and. And delivery fees that was separated. There was some licensing and, and software services there. And we now have a new recurring software revenue starting in Q2 because of some of these recent partnerships that I mentioned. So, uh, you know, it, it takes some effort, but we’ve been, uh, very deliberate trying to make sure that we have a diversified, uh, revenue base.

Grayson Brulte: As you’re diversifying your revenue bases, I do wanna highlight for our audience here in Q1 2025, your delivery volume. This is impressive, increased 75%. You complete over a hundred thousand deliveries. What led to that dramatic increase? Was it more robots on the platform? Was it more demand from customers? What led to that dramatic increase? 

Ali Kashani: Yeah, I think we shared Q1, which is 75% increase, and then in Q2 we are projecting another 60 to 75% increase again. So, uh, the reason is, as I mentioned, we started building our Gen three robots. , This year we wanna launch 2000 of them. We have a contract with Uber. I. We ended last year with a hundred. We’ve already built, 250 more robots that we, we added, uh, in Q1. And those robots, they didn’t get introduced on day one. They came over time. And also when they reach us, there’s, uh, you know, it takes a few weeks, , for them to actually get into utilization. So,, with all that in mind, the robots were gradually increasing in numbers. And at the same time, we were adding new neighborhoods. We were adding new cities. We were adding new merchants. Our merchants, uh, uh, partners increased by 50% last quarter. So we, we are, as I said, over 1500 now. Uh, so all those, uh, have had the impact of lifting the volume. And again, they’re gonna continue to do that in Q2. Even though we are not adding more robots anymore, we are just becoming more efficient. I’ll add one more thing. By the way, in the two year, three years before, uh, this year where we started adding more robots, we had the same a hundred robots. And those a hundred robots became, I think four times or more, in terms of the delivery volume growth, something like 20% months over months growth, I think is the, the exact number, uh, on average over three years. So that was, uh, you know, all by efficiency, just getting better at using the robots. So we have a lot of levers. To push, uh, uh, right now and a lot of, uh, upside in front of us, basically add robots, add cities, add merchants, and become, uh, more efficient and better at, uh, at our job.

Grayson Brulte: You have. Clearly growth upside, but you’ve hit the nail on the head efficiency. When you can unlock efficiency in business, you can really unlock a good things. Ali, over the next decade, how do you see the autonomous sidewalk business evolving, and what do You see it becoming? 

Ali Kashani: I think it’s gonna go from novelty in most cities to the mailbox stage, the utility stage where you don’t even notice it, and then it becomes this. Infrastructure that cities come to rely on. , I remember for example, when Covid happened and we were operating a small fleet in LA and there was this overnight, , change where people used to see our robots and say, what is that? To, they would see the robots and immediately know it’s contactless delivery. So there, there are certain moments where you see this, this transition and how people, uh, how receptive people are. I think we are gonna have this adoption. Of, uh, robots. And it really, truly is one of the first, , truly new form factors of mobility in the 21st century. When you think about these robots, I think, uh, rob delivery robots are what the scooters needed to be, which is autonomous. They are able to move around by themselves. They don’t need humans to deploy them and move them and charge them and all that. So this to me, is one of the first truly novel kind of move form factors that’s gonna start this process of what I call bumbling the car. Because we do too many things with the passenger car. That is way too inefficient. And by the way, it also costs lives. Like there’s 20 pedestrians dying every single day in the US from car accidents, and we seem okay with that. But if we can remove a quarter of car trips that are about last mile errands and shopping, for example, by having robots like this. Well, that, that’s gonna save lives, that’s gonna save congestion, that’s gonna save on CO2. So we, we are on this mission. I think we can bring that first truly 21st century, , form factor to market. And in the next decade they will become very commonplace and, you know, a fact of life.

Grayson Brulte: They’re gonna become effective life. The thing that you didn’t hit on that I’m gonna hit on will lower individual stress levels. Oh, I forgot the milk. Oh, I forgot the eggs. I gotta get back in the, and you’re stressed out. And let’s be honest, when you’re stressed out, you drive fast. You don’t necessarily pay attention. Boom. You hit a kid on the bike. Well you just, their dead that you ruined your life. Well, if you, you forgot the milk, you put it in a, in a sidewalk bot, that scenario never happened. And there’s a lot of positivity that that comes out of it. And then you said this with scooters. The biggest issues with scooters was the sidewalk clutter that over as they scaled, only got worse. Your robots don’t cause sidewalk clutter like the Pixars. They go home to sleep at night. They could be friends with Lightman McQueen, who knows? And if they talk at night night at the museum, you could have night at the depot, but, but they go back and they’re not there on the sidewalk. Ali, this has been a fascinatingly insightful conversation as we look to wrap up for today. What would you like our listeners and views to take away with them? I.

Ali Kashani: I think my message is robots are coming and uh, you know, they say future is here, but not evenly distributed. Is, this is one of those examples. If you go to LA it’s just the most normal thing. And, uh, you know, for, to most folks it might be news. And, uh, and I think, you know, if you, uh, really take a step back, , this is a really interesting moment because we’ve taught computers to. Understand us and to understand our, our world and also communicate with us. This is all the progress in the last couple of years. And those are the, the, they have been the biggest missing pieces for robots to enter our lives. So, uh, I’m obviously super optimistic and excited about this, this transition. I think it’s gonna make our lives better and, uh, you know, we are gonna keep pushing and bringing these robots to more and more cities over the coming months and years.

Grayson Brulte: Robots are scaling. Robotics is becoming a multi-billion dollar, soon to be a trillion dollar business. The future is bright. The future is autonomous. The future is serve robotics. Ali, Thank you . so much for coming on The Road to Autonomy today.

Ali Kashani: Thank you.