S2E7 | Nov 19, 2026 - Physical AI Part 2 - Where it happens?
Billy Riggs (00:32)
All right, welcome to Rewiring the American Edge. And In our last episode, we talked about kind of the we talked about what is physical AI and who builds it. And you know, I want to pivot in this, I guess my argument there was that
you know, that it's not just software, it's not going to be built by engineers so alone, but by this broader this broader workforce. everything from electricians and technicians, operators and public servants, city planners. but I wanna carry this a little further today in our second of this multi-part series on kind of on this topic of
Where does it happen? And when we use the term physical AI, many times I gravitated to this idea of the autonomous vehicle. but it doesn't have to be. It could be an autonomous robot, it could be a sidewalk delivery device, it could be terrestrial, not not just terrestrial, it could be urban aerial vehicle.
rural area vehicle. I think we're gonna see more and more iterations of what a physical manifestation of technology looks like in the real world, whether or not it's your Roborock or your is it the EcoVac that cleans windows now? All right. The point is it's not happening on a screen, it's happening in the the physical world.
And it faces physical limitations in our streets, in ports, warehouses, in our factories, in our transit systems. and I even have this vision of improvements in the in the labor force for construction. And we think about the way we build buildings and it could be on construction sites and in energy networks, et cetera. And
and So I think we can say is that what we can say is that physical AI is when a technology collides with physical form, the built environment. And for city planners it's really important and for policymakers really important because it deals with with the infrastructure of cities. And you know, I think that
is really important because it takes us from this software pivot to the where question; the software pivot to from from software to place. And yes, when we work on a computer, we get these dynamic, this dynamic feedback when we when we use a word processor, if you're if you're a coder when you write code, or if you're a statistician when you you analyze your data.
You get a you get dynamic feedback in the form of of electrons, right? It's it's it's it's moving electrons from one location to another and manifesting them on your screen so that your eyes can depict them. But the the difference with physic this this new form of AI that's really i existing outside of a digital environment is that that it needs to negotiate curbs.
It needs to merge in traffic or avoid pedestrians or interpret a stop sign. but it also needs to interact with law enforcement and the fire department, and in the case of automated vehicles, work in all different types of of weather: rain, fog, snow, even construction, you know, avoid construction sites. and that inherently brings us to the where it happens question, the physical world.
And the problem changes because a robot moving through a warehouse has to under warehouse and understand that a ware a robot moving through a warehouse has to understand that space, just like a robot moving through your house cleaning your house has to understand that space. There's LIDARs in most of your little delivery royal the cleaning robots that you have in your house. And that's why the FCC, for example, I don't know if folks saw, but the FCC now is put out some pretty comical and yet onerous.
regulation on many of these these house cleaning robots from RoboRock to to Roomba, that that they will have to abide by some pretty, pretty rigid regulations simply because most of those components are made in China. So you know I think when we think about kind of how some of this tech has to exist in the real world, understanding buildings, people, space, balancing
You know, for example, balancing supply, demand, equipment failure, if you're dealing with robotics in an energy environment. you know, I think ultimately what we say is the physical world is uncertain, it's hard to model, but it also many times it's not what we something we can simulate. there's a number of of
I think There are a number of companies that are coming out in the automated vehicle space right now that are trying to say we can simulate driving using AI generated so We can train our self-driving AI using AI generated renditions of the built environment. But you know, therein lies a problem because it really just gets into this this these issue of models training models. and
you know, I think the the real world's messy. And can it be can it be, can we simulate that mess? can we simulate that scenario that the the cruise vehicle saw a couple of years ago where we had a a pedestrian hit and run by a human vehicle that's projected in front of a cruise vehicle? Or the issue when we had a a Waymo vehicle, and I think it was Waymo that was tracking a truck that was carrying a tree and
couldn't wouldn't go around it because it looked like it was a street tree. You know, these are physical, they're they're economic, regulatory, they're, you know, they're the same, they are yes, they have a technological manifestation, but this technology has to negotiate this physical these physical barriers to space. So I say that very long-winded intro to kind of say these these places where we deploy physical AI.
are often where it succeeds or it fails. And in our case, and in in in my work at the Autonomous Vehicles in the City Initiative, this is where, for example, the street for any of this is the biggest test and you know, driving it's why driving has been such a complicated function to to solve. you know, We have a a strip of of pavement that has many times have has different types of markings everywhere you go.
Some have sidewalks, some have traffic signals, some have parking spaces, loading zones, bike lanes, bus stops, different types of bike lanes. You can see different interfaces from a crossing standpoint. And it's also a very public space and it carries many different kinds of cassengers, right? We we carry everything from buses and bikes and pedestrians to delivery vehicles and and private cars.
And you know, that's layering on you have roadway workers and and utility crews and you know people that are maintaining and building our city at the same time, you know, construction vehicles and things like that. So we have that was just just popping in my mind we have We have waste vehicles that are that are there, you know, picking up our trash every week. And I think increasingly
we can think of this not as much as a as a transportation corridor, but a representation of the urban sphere. In my book End of the Road, you know, I talk about these different things that a street can be, a social space, a cultural space, a place for urban greening, in addition to be a place for multimodal transport. And
the a tree a street can be much more than a road. And that's why I call it the end of the road, because a street can can house many of these different functions. And that's what autonomous vehicles and and automated robots and even smart traffic signals and eventually, you know, maybe even AI assisted emergency response vehicles or or AI assisted drone to drone delivery.
will have to navigate. And it's where the frailty of these systems, the frailty of the in the infrastructure, the small weaknesses, these chinks in the armor of the system become very visible. In many of my lectures, I talk about the three things, the three best things that cities can do are install signs. Yeah. The three things cities can do, or the thing three things cities can address, the three things that cities can address are signs,
lines and potholes. And I usually say that somewhat comically, but you know, poor l signage, inconsistent signage, matters. You know, a If you don't have lines on your road that are visible, or if there's not a delineation on the side of your roads, it can be difficult for a vehicle to navigate. an unclear curb, unclear curb matters.
And it's where And potholes are not good for any drivers. drivetrain. So it's it's really where protocols, data standards, infrastructure at its most basic core starts to matter. And that's where, you know, automated vehicles that's kind of like the first thing, one of the first use cases that we're seeing in the real world that of fully operational physical AI.
I maybe maybe, you know, house cleaning robots are another fully operational example. I mean, I I love my Roborock that that definitely has me avoiding doing tons of cleaning and keeping the dust levels very nice in my house for my household. but you know, I think this this deals with some of the the challenges. If we were go back to
⁓ if in that example, the the Roborock could be, bumping around when things in, it its understanding of the built environment inside your house and the obstacles it will encounter frames how it will clean, just like a vehicle shouldn't obstruct the bus stop, shouldn't can't needs to be able to navigate a crowded event and to coexist with
transit operators, to read, and this has been a big challenge, to read police officers' signals. to understand when there's a and there's some new regulation that came out of the California governance governor's office to try to address some of these new these corner cases. And this happened over the summer of 2026... Kind of the one of the first times when we've we've made deliberate requirements on companies to
To really make sure that they have emergency management plans in place.
So dealing with edge cases, dealing with infrastructure response is really important. In this case, you know, if we drill down to that example of of automated vehicles and focus on it, the curb becomes the critical seam. And it's you know, it's long been overlooked, but it's where
a lot of the activity happens. And this has been an emergent topic for years. with the with Rideshare coming to fruition in 2012 when when Lyft decided that it would implement peer-to-peer ride share and and Uber followed suit very quickly. The curb so in overnight became much more valuable of an asset because we were seeing a move to this pickup and drop-off ecosystem. And
you know, it's not just for vehicles for hire or or gigwork vehicles, it's for transit, it's for people getting out of private cars, and it's for automated vehicles. but it's where passengers get in and out and and people get out of the bus. It's where passengers ingress and egress. And there needs to be safe access. There needs to be space also for many other types of uses;
whether or not it's delivery trucks or people in wheelchair accessible vehicles or even for emergency vehicles to be able to get in and out. And of course, these days we have a lot of commercial activity that happens on our sidewalks at the seams of our curb too. so you know restaurants using that for outdoor space and and an opportunity for placemaking, for urban beautification, for creating softer cities. And you know, I think
as we move forward, there is an opportunity here. And yes, the curb is governed by cities and it's gotten by these really static rules and these opportunities that that it has to stay fixed and non-transactional, that the curb is a fixed asset. In in itself, I I've called it in a couple of lectures. I've said we take a very Euclidean approach to the curb. We we assume that.
In And when I use the term Euclidean, it's from a classic planning case called Euclid versus Ambler. If you if anyone has gotten their AICP certification, it's guaranteed to be on that I AICP American Institute of Certified Planners exam, because this case basically established this idea of use-based zoning and how zoning needs to be separate based on use. Well, what we've seen over time is that that was that Euclidean-based zone zoning was established in a time where we had to put
You know, had to separate noxious uses, highly industrial uses from residential uses, from softer uses. Well, in our modern society, Euclidean zoning isn't as useful. We don't see as much variation in the the type of land uses over space. you know, residential area like next to a small office area may not be such a big deal anymore. And so I use that.
parallel for automated vehicles because we're getting to the point where the division of space on the curb is becoming very has become very separated. We have the bus space, the parking space, the the white curve for pickup and drop off, the yellow curve for commercial loading. So you get all these different fixed locations. And what it does is it doesn't function as well for transactions, for high turnover.
And this is really important because an automated vehicle and data in general, AI, really changes the the the equation there. And it I think what it does is it makes curb space more valuable. It makes an environment where you can have perhaps a softer curb, a curb that thinks
A curb that is this multimodal marketplace. And this, these are two terms that we've been working on with a couple papers, to where there's an opportunity to to do the pickup and drop off in what was in a place that a certain time of day could be parking, a certain time of day could be could also be a transit area. and a certain time of day could be for
Loading, commercial loading and transit and flate and and freight. and that in that case the the curb really starts to be less of a passive piece of infrastructure and the physical AI makes it more of a managed interface. And this management can also be done digitally. You can you can almost program chaos into the curb. And I'm I'm stealing that from
Iskander Tange, who I've worked with in in Utrecht and Amsterdam on some projects with his company Coding the Curb. But this idea that you can you can use access and price to maximize turnover in a very short-term way, where the curb can be reserved, monitored, reallocated dynamically, is really powerful and is something that Molly Cohen D'Agustino and I just go explored as a part of a pay a paper we were looking at
You know, what is the future of the curb? And what we found, and this was published earlier in 2026, what we found was that, you know, curbs need policy and goals and they reliable, you know, we need to have data about about them, and clear enforcement, and all of that shapes utilization. But at the end of the day, if you're a city and you're prioritizing storing cars on your curb, then you're not.
achieving the highest and and best use of that physical real estate. And I think that that organizational capacity to manage this physical asset rather than actually as a this physical asset as a digital tool is one of the central lessons of how we prepare and and plan for and are and are ready for physical AI because the you know despite the
technology being dynamic and even rideshare in general and transportation being transactional in nature. that people are engaging with it and consuming it; many of the physical environment institutions remain static. And if they do, you know, I think they will, we will, they will underperform, but they we we aren't maximizing the use of that resource.
Sometimes we use the term it's an asset that's stranded, but in this case it's under utilization. We end up with red curb, for example, that's dedicated to transit uses, that's sitting there vacant 80% of the day and only you divide utilized for a bus pickup and drop off 20%. How could that curb be actualized by integrating?
By integrating other services, whether not it's whether or not it's taxis or rideshare or automated vehicles. Are there ways to price in or create more flexibility with that space? And I think those are important questions that cities need to be starting to grapple with because you know, physical AI is is coming. Is it coming? How is it coming? you know, it's coming in waves.
It's not all overnight. I mean, think many times people ask me and they they they assume that they're the city of San Francisco is crawling with millions and millions of maybe even thousands of vehicle. Even today at a conference I said they said, there's a thousand AVs in San Francisco. And actually, from what we do know, I think there's a thousand automated vehicles in the entire Bay Area, which means at any given time I'm not
exactly sure that all thousand are in the city of San Francisco. So let's keep it in let's keep it in context. But also if you think about the number, sheer number of trips that go in and out of the city of San Francisco every day, for example, we estimate it's less than 1% of traffic volume. So it's it's when I say you know this is happening, this is happening in waves, it's happening in is a part of pilots, as a part of business models that take time.
To integrate with the system. And that's where, you know, I think the time is now for us to think about how they integrate with the physical systems we already have. And while we said the curb is one thing, one of the biggest opportunities we have in cities is to consider transit. Now, I think automated vehicles are sometimes framed as being competitors to buses and rail. I
Sometimes people may choose not to take a bus or rail for their for a ride. And in my mind, that's the market and that's their choice. But what the literature has shown is that what you know that market share that has been taken by automated vehicles is largely the market share that has already been attributed and already being attributed to rideshare. And what we s do show, and I shared this paper as a part of the last episode.
We have are already seeing that both rideshare, whether or not it's automated rideshare or traditional rideshare, fills last mile gaps. And we're now seeing examples, and we'll put a link to Waymo's credit transit credit pilot partnership in the show notes. we're seeing that that many of these things, these services formalized to fill these gaps and to bring people to public transit.
So typically From a cost effectiveness standpoint, RIP ride chair can serve lower density areas, provide service in late night hours, and also serve people with limited mobility. Particularly, there's been a lot of success with blind and low vision communities, and doing so in a way that's actually more cost effective per ride than public transit. Because what we see in terms of the public transit for fare, what you see in terms of public transit fare.
when you pay your fare is only about usually 15 to 25 percent of what it costs to provide that ride. Now none of this happens. none of that vision of bringing and connecting people with public transit or integrating automation, integrating physical eyes AI with public transportation systems happens just by waving waving a magic wand. It takes
clear and assertive visionary policy to not only regulate what happens with automated vehicles on streets, but also to focus on how it can be integrate. It can be integrated with our traditional transit services. And I make the case that there is an opportunity now to increase rides and revenue for public transit agencies.
And you can build complementarity into the system so that you don't have competing services. And some of the way you do that is exactly what's being piloted here in the Bay Area right now with some of the programs that are going forward by looking at issues of integration. And that's fair integration, service integration, providing geo-fenced station access. It's
making your curb data transparent and available and doing appropriate management so you have safe pickup and drop-off. And also making clear what your service, what city you if you're a city, you should make clear what your service goals are. Are they framed around equity? Are they framed around different hubs? And building on that, and we could cite some of what's happening in Hamburg with Hamburger Hochbahn or Oslo with Router.
there's an opportunity for public-private partnerships. There's an opportunity for seeing the synergies between new and innovative modes of public transportation, new and innovative modes of transportation, and what has traditionally been our core high capacity routes. And at the end of the day, you know, I think I would underscore that all of that is.
deeply, deeply spatial in nature. It's physical in nature. And that's why planning and design really really matters. It's why know physical design sh you know is not an obstacle to innovation. It's a connector and a, I believe, a facilitator of the innovation, particularly if we're if we
as a s society start to invest in the infrastructure that undergirds the performance of this technology. And you know, that can be as simple as making sure that your your energy grid can support the new charging demands. It can be as simple as making sure that 5G service is equitably is is equitably accessible across your city. And that actually is something that that I think really fits into a larger
Vision for public good. And it gets us to starting to talk about really pragmatic things, not these fictitious visions of what may happen or what may not happen, of a of a grand heaven and a horrible hell. What practically can we do on the ground to have a best have the best outcome? Because I think companies and cities want the same thing. They want functional
service in cities. And if we looked at like why we have robotics in society, it's to actually make our world a better place. Many countries around the world are using robotics in a future where we don't have enough workers, for example, to do certain tasks. And our infrastructure investments going into the future
are really going to facilitate that. Our infrastructure sent decisions are practical and not neutral. They don't, they don't are practical, but they're they they shouldn't be our our our infrastructure decisions are are should be pragmatic, but they also should be, you know, nonpartisan. they They simply provide this this this foundational piece
to seeing the innovation happen in the physical world. And yes, signs, lines and potholes, but that's also thinking about bridges, bridges, and power lines, fiber optics. thinking about how distribution centers and data centers fit into this equation. And those kind of that type of infrastructure can be deployed in a way that really benefits residents at large.
Infrastructure creates opportunity, it shapes and it unlocks or locks patterns of development. And you know, it's it's also something that helps our economy grow. you know, For example, our roads do make movement easier. Rail lines do are complementary when you install them or maintain them. you know.
A bike network is isn't truly a network. you know, a so, you know, I think what I would say is when you build this kind of infrastructure, we're not just making decisions about technology. What when you build smart a smarter grid or you invest in
digital architecture or your energy grid or data centers or any any of this stuff, we're making decisions about the future of our economic prosperity and the decisions, particularly when cities make them, to see that prosperity proliferate down the economic ladder. And you know, I think this is where when we, you know, if we really are honest with ourselves about these policy decisions, they aren't they're
They they do become things that perhaps the technology doesn't necessarily answer the the policy or the political questions. you know, I think what we have to do in terms of of a societal decision, so we have to decide do we want our cities to thrive and to innovate? Do we want to our cities to just be
passive recipients of new technology to be non-responsive, or do we want to co-develop them? I have long felt that that the public sector couldn't is best situated and the the best outcomes occur when they co-develop solutions. And I I feel like when we look at kind of the again, the autonomous
Vehicle example a lot of what happened with the really the 10 to 12 year policy development process, for example, in California, has been a very public and deliberate process where the rulemaking was open for public comment. Any citizen, any city could participate in that. And as a result, we've seen, I believe, in not only in the US but in the world, one of the most significant
policy and regulatory ecosystem for making sure that
cities and governments, cities, regions, states, are not just recipients of the technology. The technology has to go through multiple rounds of of testing and deployment and approvals, going from, for example, in California to from you know a drivered permit to a driver.
a driverless permit to a deployment permit. And that's all just to get a digital driver's license from the Department of Motor Vehicles. And at the same time, there's this whole separate permitting process for carrying passengers. And this is an opportunity not just for innovation. It's for convening and it's for universities and businesses and community groups and to work with.
utilities and public agencies and create innovation ecosystems and the architecture for everything we've been talking about. for the next year. Physical AI presents an opportunity for us to reinvigorate and rewire the American edge. For us to kind of really not just shape the market, but to deliver public value that increases economic
competitiveness. it allows us to kind of establish that, hey, yes, we would like to set a public goal that automated vehicles should connect to transit, that we can establish that AI can be used for traffic management, or that we want digital twins, for example, to support infrastructure maintenance and our construction planning and robotics to help with our inspection and waste management and
Hazmat work or you know, radical solution that I pulled out. You know, we want robots to help us in disasters. And you know, this is where if we this is where, you know.
This is where, and this is where, you know, one of the greatest barriers is fragmentation. We have a fragmentation problem in government that prevails. And it's something we found in this report is that you have a report that Molly Coasting Cohen D'Augustino and I did last year between University of San Francisco and UC Davis: is that you have multiple agencies, for example, managing one.
one thing. You have, you know, maybe traffic signals by one organization, utilities by another, and then of course, emergency response and enforcement done by another. And then, of course, your planning function, both transportation and land use, are done by another. And of course, that all abuts all your private mobility providers, your AV providers, your trash companies. And at that point, every
agent in this system may be acting rationally with their own policies and mandates, but when you put them all together, you could still end up with system failure. It's a classic coordination problem. And this is physical AI challenges us to cross these organizational boundaries and to break down the silos. And I will tell you, even academia, yeah, even academia has the biggest silos. So we're part of the problem too.
But when we look at where, for example, an automated vehicle interacts, it does interact with transportation planning, police, fire, public works, public transit, but also insurers and macroey systems and telecom and operational systems and logistics and even land use for charging. So as an example of physical AI, it's a challenge of integrated. The technology does not solve the integration. problem.
This is the role of government. And that's what we're going to talk about. This is a role of government. And that's what we're going to talk about in our last little segment: who governs and how do they govern. Governance is important because if we don't get it right.
Technology if if we if we don't address issues that that threaten coordination, if we don't s if we don't address issues of fragmentation, failures don't just happen, they get repeated. And when failures are repeated, when failures are repeated, we start to lose public trust. we start to see an erosion of our democratic processes.
And so, yes, we don't have, ⁓ I don't have all the answers. America doesn't have all the answers; There's examples from Europe in terms of governance. but at the end of the day, taking on these coordination challenges may be the biggest thing we can do to have the take on taking on these fragment at the end of the day, taking on these fragmentation challenges may be the biggest thing we can do to stimulate.
You know, to stimulate not just the building that we talked about in our first episode, and not just the
Not just the building that we talked about in the last episode, and not just the where it's deployed, but to stimulate the economic growth, the experimentation, the entrepreneurship so we can continue to see these innovation sectors flourish throughout our economy. So that's enough on this for now. We'll talk more about the governance function in our next episode. I'm Billy Riggs, and this has been Rewiring the American Edge.