Building the packet switched network for the physical world
In a previous post, I outlined why Uber is still very far from achieving its mission of disrupting car ownership through providing transportation as a service (rideshare in total is only capturing 1% of vehicle miles traveled in the USA).
In this post, I want talk about a larger mission that encapsulates that mission of disrupting car ownership while establishing an even grander ambition for what Uber is ultimately trying to do.
This mission is “building the packet switched network for the physical world.” (h/t to Jeff Holden, who to my knowledge first coined this phrase.)
Cities and the users of their transportation infrastructure network have historically existed as something more along the lines of a circuit switched network. E.g. when I want to go from A->B, I plan to drive my car along a predefined fixed route until I get there, largely ignorant of any realtime network topology conditions or other users of the network (until the advent of real time traffic in GPS systems, which afforded some ability to self optimize one’s plan for traffic control). Multiple cars can share a lane, but not in any way that optimizes for traffic shaping or resource contention.
Meanwhile, other planning of moving an object X from A to B is done myopically and independently. Some examples of ‘movement plan creation’:
- other passenger travelers plan their own closed circuit routes
- cities plan their closed circuit, fixed stop bus routes (typically on a slow refresh cycle) and train schedules (even slower, by physical definition)
- UPS/FedEx/Amazon plan their own package deliveries
- other shippers, e.g. alcohol distribution companies plan beer deliveries
- restaurants plan their own food deliveries
- and so on
All of this plan creation is happening largely myopically inside independent systems, some of which are still very analog. E.g. someone is still using a highlighted AAA fold up map to take a road trip :)
Uber on the other hand, is in the beginning stages of building a packet switched network for the physical world. When a rider wants to go from A->B and expresses that intent to the Uber network, the best plan (pickup spot, drop-off spot, car, and route) for that trip is chosen in joint optimization with other rider plans and awareness of that selection’s impact on the overall network. With uberPOOL, riders can be multiplexed in a sense into a shared medium. With the growth of UberEATS, we’re also in the beginning stages of extending that optimization to packages in addition to people (delivering people and packages across shared supply).
It’s an interesting thought exercise to think about what a physical world analogy of the classic OSI model of computer networking might look like for the physical world of logistics.
In the Uber Maps stack, we have a model of the physical layer of the road network (navigable routes, turn restrictions, closures, speed, traffic, etc.). We also have a model of addressable physical points on that network, e.g. potential pickup/drop-off hotspots for a given request’s start/end point and an ability to geocode & reverse geocode between address formats (e.g. POI vs lat/lng vs street address). All of this represents a base addressable physical network topology with nodes of potential pickup and drop-off points and edges of potential travel paths between them.
The Uber Marketplace stack operates on top of the Maps stack. Given a network topology of addressable pickup/drop-off nodes and potential travel edges between them (with initial weights of estimated travel time or ETA between them), you can enrich that graph with information about realtime demand and supply (so on the demand side, where people and packages are, where and when they want to move to, and on the supply side, where cars are moving from/to as well as their state, such as open, on trip, en route to pickup). Now we have a network topology where we can intelligently route riders from their iPhone location to a pickup node to a car to a drop-off node near their destination.
The Uber Matching system is effectively operating as a sort of transport layer, figuring out how best to map physical world data (the atoms that comprise people and pizzas) into car ‘packets’ along network optimized routes.
As the packet switched network for the physical world continues to evolve, I am increasingly interested in exploring four key questions:
1: What are all the primitives that will ultimately exist in the packet switched transportation network? e.g. what others refer to as transportation ‘modes’ like bus or car. I prefer the analogy of a primitive from computer science, since what we’re really talking about is a basic unit of computation inside our digitized transportation network. What are the smallest units of computation for relocating an object of parameters X from location A -> location B? Obviously some primitives will work for some parameters but not others (various SLAs including arrival time, capacity requirements, ability to walk, etc.)
A few short years ago, no one really predicted that e-bikes and e-scooters would become important primitives. Everyone was too focused on future primitives like autonomous cars and flying cars.
I highly suspect that we’re not done yet. All of the important primitives that will ultimately play a large role in the packet switched network are not yet imagined or widely deployed yet. Clearly others share this belief, as evidenced by Nuro’s focus on new small form factor delivery vehicles or Lime’s foray into people moving pods.
2: How will these primitives be combined in the future? E.g. once you have scooters, last mile becomes lower friction. Pods could further change the optimality of short range packet switching. Moving data into and out of multiple packets will become lower friction (e.g. mapping a train or Uber arrival to another waiting Uber or bike or pod), and conveniently a single wallet transaction. Lower friction multimodal could significantly alter how people move.
3: What will the gateways and routers in this physical world packet switched network look like? Historically, this has been largely dictated in slow time by urban planning around train lines, bus lines, stations, etc.
If you want to see how the packet switched network is already straining the existing physical world limitations, look no further than the SFO airport. The physical infrastructure of departures and arrivals levels were designed pre-Uber. Now coming into the airport, you see signs telling you that in order to avoid traffic, you should drop off your departing passengers at Arrivals, and meanwhile Ubers are required to pick up arriving passengers at Departures :) Uber requesters can still overwhelm a single door inside a terminal by myopically all requesting closest to their point of egress, and of course uberPOOL adds more complexity to the gateway design and routing problem.
Redesigning airport I/O is just one fractal inside the many fractals of city I/O design. As another example you can look at Uber and SF working to define rideshare curb space or Lime working with local merchants to create egress / charge points inside their stores.
Where will the points of ingress/egress between various primitives be? How permanent will those decisions need to be? Which leads to a fourth question:
4: How active of a participant will cities be in designing this packet switched network? There are huge opportunities here for cities to define the ‘nodes and edges of the future’, be it curbs, stations, parking spots, sidewalks, dedicated lanes, and so on. Cities could also play an active role in creating and standardizing metrics for utilization efficiency of the network. E.g. company X is moving this many people with this efficiency across one set of primitives while company Y is moving another set of people with a different primitive mix. This could create a standardized definition of success in which companies could compete across. If Uber is to be given a chance to run the bus network for example, as Dara has said publicly that he wants to do, there has to be some definition for how we measure success there.
It is still very much day one, and I’m excited to see how the packet switched transportation network will reshape the physical world in the years ahead.
