Name: Bots on the Block: Your Lunch is Learning
Uploaded: 2025-12-10
Duration: 10 min 2 s
Description: They’re tiny, quiet, and smarter than you think. MJ Burk Chun reveals how Serve Robotics’ autonomous fleet is mastering urban navigation and reshaping delivery through sensor fusion, real-time learning, and sleek (plus adorable) design. The session concludes with a live demo of the company’s latest robot in action.

They’re tiny, quiet, and smarter than you think. MJ Burk Chun reveals how Serve Robotics’ autonomous fleet is mastering urban navigation and reshaping delivery through sensor fusion, real-time learning, and sleek (plus adorable) design. The session concludes with a live demo of the company’s latest robot in action.

Bots on the Block: Your Lunch is Learning

Conferences

Alright, good afternoon, everyone. Thank you so much for sharing your afternoon with me. Um, I'm gonna talk a little bit about Serve Robotics and geranium is gonna help me here. Um, so just starting off, uh, a, a couple of slides. Um, I don't know if it's come. Oh, there it is. I totally missed it. So we are an autonomous, uh, delivery robot. Start off within the Postmates, uh, tech innovation lab with the question of why move two pound burritos in 2 ton cars. Really, there was inherently in this, uh, Postmates data, and we were, you know, born in, in SoMa in San Francisco, was this a concept of why it was so much cheaper to get something from Chinatown, or sorry, from China than from Chinatown, right? Like this is just like a known thing. Um, and whether or not you actually can find yourself a 2 pound burrito or they're a little bit lighter, it's still an inefficient means of transportation from a delivery perspective. And we looked at all of the delivery data, and it's a three-sided marketplace. You're getting drivers to agree to a job at a restaurant to a customer. And part of that driver pool, they're really interested in maximizing their time. And so what they weren't interested in were short distance deliveries with low tips, especially in dense urban areas where now. You have to find parking, maybe you're double parking, maybe you're risking that fine, um, and you're jumping in your food may or may not be ready, and that's eating into your earning potential for the, the much more lucrative job. So that's how we started dreaming and evolving robots. So we actually, uh, partnered with San Francisco, um, had a pilot program. And I think week two, we had that first little iteration, um, we had someone's phone. On RC car and drove it up and down Market Street and said, alright, what are we gonna learn about what it means to be on the sidewalk? Um, then we had another iteration where someone was like, oh, to Hurt Locker. And so, you know, we, we were like, oh, OK, this is where design really started to matter. We're talking about technology and embodied AI in our world. Um, that's, that's not gonna win anyone favors, um, and so we had what we affectionately called truck, and then we evolved to more of this robot, uh, form factor based on a shopping cart because it was familiar and we really had this idea that you wanted to be welcome on the sidewalk. And so, um, really, what are we trying to do? Safe, socially aware, and this is where the people part really matters, robots in a dynamic world, because people are unpredictable. Our cities are very unpredictable and diverse. And the theory, uh, this is where the, the AI portion really comes into play. We have the concept of like identifying people, uh, very traditional method, um, understanding what's the, what's the wall, what's the ground. Uh, what's the pedestrian light signal? What's the status? Can the robot cross the street? Not a joke, a real thing. We're like, oh, let's let's spend lots of time talking about how do we know that a robot is safe enough to cross the street. Um, robots have nightmares about cars. Cars are also very scary for robots. The intersection is actually one of the most dynamic places in our cities. You've got high velocity moving vehicles. And you've got a robot that's trying to share the public space as it crosses the street. Um, so it's a very, uh, you know, autonomy and hardware bound problems. Truly like where AI in 2017 was all about. And then we, we deployed in Los Angeles. We're doing deliveries for the very first time with robots, and the reality was actually even more dynamic than we expected. Um, that's one of my favorite photos. Uh, the robot was thoroughly confused about the pet baby goat. It's just like unidentified does not you know, like see the person, but like what is this, um, that was, that was one of the moments where like, oh this is this is gonna be an interesting long tail problem um sidewalks very different from simulation. The discontinuous tree roots, uh, the infrastructure is not quite there, uh, and, and so we started learning a lot more and evolving the hardware to be more reliable, um, and, and really to really traverse the neighborhoods and so one thing that's really interesting about this like sidewalk experience for us. And looking at this room and some of the conversations that's been happening really to me is all about how the future really is already here and I think many of you are involved in that particular version of the future. It's just not evenly distributed. William Gibson, uh, he has a great series of of books about the near future in terms of, of speculative fiction. Um, we're adding new cities all the time and so we're adding our, our. You know, flavor to these cities, um, in terms of what it means to, to live in a city. What's really, really critical about being physical AI in these cities is we have to be welcome. Robots don't have rights to be on sidewalks. People do. And so most of our design decisions, engineering decisions, and safety decisions really are around how to be conservative, but effective in a dynamic environment. And it is also quite amazing how each city expresses itself in the way people walk, might be a little different. Not just the sidewalk infrastructure, but also how people drive. And so one of the ways that it's a delivery business, it's all about being on time and reliable, one of our things is we've invested in routing and we learned that the conservative routing that we've optimized for in Los Angeles in, uh, very high traffic, high speed intersections doesn't quite hold up in Miami Beach. Folks are gonna cruise down Miami Beach. They're not racing to make that right hook and from a routing perspective. Might not make sense. And so we're learning things as we scale and really, once again, it's back to that human element. What makes cities dynamic, vital, interesting, is the people, what's really specific geospatially, it's locale, it's community. Um, and so we're learning a lot about how we can build better experiences for people, leveraging that learning, and it wouldn't be possible without some of the advances in sensors, sensor technology, um, and autonomy, specifically around end to end AI, uh, we're able to take. LiDAR data, camera data, movement, map it to GPS coordinates in the cloud. We've actually taken, um, we're taking all our millions of miles. We're on, on our way to have one of the largest AV fleets in the world. Um, and these are images, uh, and really maybe not the most exciting images for some of us. Um, we care very much about the sidewalk, the sidewalk quality, whether or not an intersection is missing curb cutouts, uh, so it's, it's a lot of data that we're collecting about the status of the city, the state of the city, and then we're leveraging that power not only to train. The um algorithm so that the robots are behaving better but we're actually sharing the data with cities so that they can address things like that little yellow circle it's a little hidden pothole and we're changing the city infrastructure by improving it with the data that we're collecting by doing deliveries on the sidewalks. Um, and I think the last piece, go back to it, is that embodied AI perspective of always delivering delight. Um, I think this is where I wanna maybe be humble and say more people will walk next to the robot, drive a car through an intersection, than we ever get a delivery from a robot. So if you're, if the robot's not delighting someone. Not providing some value to the community, then it's a missed opportunity to really build that future that I would hope we get to all live in. So, with that, uh, thank you for allowing me to share our story. Um, geranium is also very excited to participate in. So, um, with that, I think it's it's Oh, Exit to your left. Thank you again.