The Advent of Decentralized Transportation Systems
Self Driving Cars, not Hyperloops or High Speed Rail
Summary
- Dealing with centralized systems like airplanes or trains is not a great experience (scheduling, getting to and from stations, waiting in lines, security, delays, packing as many people as possible into the smallest possible space). We give up some of our personal autonomy, and personal time, in exchange for the efficiency of the system.
- Decentralized modes — walking, biking, cars — where you have more control and/or more free time (in the case of self driving cars) is a much better experience — less hassle, more personal autonomy.
- So, when given the choice between two modes, where time, cost, and safety are not disparate enough to differentiate, most of us will choose to take the more decentralized mode; it is a better overall experience.
- Self driving cars will provide both experience and efficiency across a wide range of distances. This is going to be highly disruptive to existing, and planned, modes of transport. These fully electric ‘pods’ will be safe, quick, and inexpensive. (Assuming we invest in self driving infrastructure — dedicated lanes, high speed roadways, etc.)
- Therefore, we should stop investing in centralized systems and accelerate our investment in self driving infrastructure. Appropriate self driving infrastructure would allow level 3 autonomy to be widely applied, speeding adoption dramatically.
- This opens up a lot of opportunities for both incumbents and startups, and creates lots of challenges for cities, states, nations, the oil and gas industry, and entrenched interests. It may also accelerate the displacement of many jobs. It is incumbent on us to build the future of the self driving ecosystem to help mitigate this.
- This impending change is a good example of our more general thesis, around Systems of Controls, that more decentralized systems are going to disrupt our current centralized ones: they will be more efficient and provide better experiences to users.
There are a lot of ideas being thrown around about the future of personal transportation — from hyperloops to point-to-point rockets to self driving cars. How can we make sense of all of them? Where should startups put their efforts, and where should governments and municipalities invest?
Strategy frameworks often allow us to ‘step back’ and look at an industry through a different lens. In this case we need one that can compare and contrast different modes of personal transport at a high level — a holistic view of the space. In particular we need to capture an individual’s personal experience as they travel. How broadly any given mode is used depends on a complex interaction of time, cost, distance, speed, and hassles — perhaps something close to what Hipmunk terms ‘agony.’
At the extremes, consider the difference between your experience going for walk versus the experience of taking a flight. Of course radically different distances are involved here, but let’s ignore that for a moment. Taking a walk is easy, completely under your control, and has very few overheads — simply strap on some shoes and head out. Taking a flight is a lot more complicated: searching for and booking a reasonable route at the right time for the right price, the trip to the airport, going through security, waiting for the plane to load and taxi, allowing people to step over you to get to the bathroom, turning off your electronics for takeoff and landing, waiting for others to exit, and managing travel at the far end from the airport to your actual destination. In the terms we will use, walking is a highly decentralized activity — one where you rely very little on others and have lots of personal autonomy. Taking a flight is a highly centralized activity where you must interact with a myriad of services controlled by others. Of course, you are not going to walk from Los Angeles to New York, so the overheads associated with air travel are well worth it when the distances are large. But in many cases we will have a choice of how we travel — for example a 100 mile trip could well be done by bicycle, car, bus, train, small aircraft, or a combination of these.
The simple idea of mapping personal transport modes from decentralized to centralized is the basis of our framework. We call it Systems of Controls in parallel to other frameworks that help to define and differentiate businesses: systems of engagement and systems of intelligence, for example. Systems of Controls spans from the extreme where everything is under your control to the extreme where you have ceded control to others in exchange for efficiency, cost, or safety.
Using this axis, and based on the previous discussion, we can roughly map our current major modes of transport as follows. We will refer to this as the Controls axis.
Owning your own car gives you a lot of freedom — go where you want, when you want. Taking a taxi is a hassle (although with their apps they are improving), versus using a modern ride sharing service where you have more control and freedom (on-demand ordering, rating drivers, etc.). Buses, trains, and air travel all have significant centralized components: stations, scheduling, security, etc.
This mapping is approximate, and that is fine for our purposes. Obviously there are short distance trains (LRT) that have more stations and less hassles than long distance trains. Likewise for buses and planes. A more complete analysis could break all of those out.
Innovation in personal transport is going to occur in a couple of ways: (1) an existing mode, through some new attribute or capability, captures the efficiencies and experiences of adjacent modes and takes over market share and/or (2) a brand new mode of transport inserts itself into a gap in the axis, and displaces the adjacent modes. Both of these arise as new technologies are brought to bear that change the existing equation: GPS, self-driving capabilities, solar plus battery, vehicle-to-vehicle and vehicle-to-infrastructure communications, security improvements, infrastructure changes, etc.
Personal transport modes are optimized for distance
To explore this in more detail we need to inspect some attributes that lie along this axis. Let’s start with the obvious one which we have already discussed above — distance.
In some ways this is surprising — why would distance correlate so well with the Systems of Controls axis? With a little inspection, however, it is obvious: more centralized infrastructure and systems are required to travel larger distances efficiently. As we move to the right on the axis more people are put into one vehicle — so the vehicles get bigger and the infrastructure to manage them grows as well. If more people are moved at the same time, we need more scheduling and security and food services and…on and on.
There are many other variables which can be overlaid on this axis — for example, time. If the distance is short (0 to 2 miles, say), walking, biking and a car ride make sense, but a bus, train or plane may not. Why? Because the overhead associated with a bus or train would take too much time. For medium-small distances (say 2 to 20 miles) we might add a local bus or local train (light rail transit) to the list, but remove walking, as walking would now take too long. For medium distances (20 to several hundred miles) many modes make sense, and when we get to long distances (more than several hundred miles) we may even consider rockets or long distance hyperloops. We will not analyze all these cases here, but let’s start with medium distances.
Medium length trips
There are quite a few attributes that correlate well at this length of trip.
Efficiency (measured in cost per mile) bears some comments as it seems a little counterintuitive. However, it is true — walking costs approximately $1/mile in food (average of 100 calories per mile) whereas long haul flights are probably around $0.20 per mile in fully loaded cost to the end user (the price of the flight plus getting to and from the airport). Thus, in comparative terms, air travel is much more efficient per mile. Other modes fit the curve pretty well, for example owning a car is about $0.55-$.65/mile including all costs (lease, gas, maintenance, depreciation, license, registration, insurance).
Safety is hard to measure for walking, but biking is generally seen to be 3x to 11x as many deaths per mile compared to cars. For the other major modes we have:
- 7.30 deaths per billion miles for cars (assume 1.2 people per vehicle)
- 0.11 deaths per billion miles for buses
- 0.15 deaths per billion miles for passengers on long haul trains
- 0.07 deaths per billion miles for airplane travel.
Again, this follows our curve quite closely, and reinforces that more centralized systems are associated with safety.
In all of the above we are not trying to be exact — that is very difficult as there are so many types of cars and drivers, different terrains, varying prices of gas, etc. The overall point here is that modes of transport generally follow these curves with respect to Controls.
Let’s dig a little deeper on the length of the trip. Breaking it out, it looks like this:
As you would expect, as the system becomes more centralized, those central services add significantly to your overall trip time. Even with a car there are some overheads (filling up with gas, pro-rating mechanical maintenance, etc.) but these overheads accelerate as we engage with buses, trains and air travel.
If we look at time against distance, as shown here, you can see the central overheads in action. This is based on some assumptions (see appendix) around travel times and speeds to and from freeways and/or stations. This fits pretty well with experience. For trips of less than 150 or so miles, driving your own car tends to be the default. Above that you will take a plane or high speed rail; the time difference between the two is not big, so individuals will pick one over the other based on cost, schedule, and overall hassle.
The evolution and tuning of travel modes over the last 50–100 years has optimized our choices. We know how to choose the right mode to balance cost, time, and hassle. So, to date, these centralized overheads have added value in many scenarios — dealing with the overheads is worth it to lower travel time, cost, and hassle.
Here come the self driving cars!
Now let’s add some disruption. Assume for a moment that a self-driving car can drive safely at higher speeds than today’s cars on a medium length trip. Getting in and out of the city may be similar (although time there should be reduced as well) , but once on dedicated self-driving freeways that are properly banked and configured it is not unreasonable to expect that 2X current average speeds are possible, with some pundits saying they will get to 200 mph. That is a significant change, as represented below — the ‘self driving’ car time shrinks considerably. So, while the experience is better than that of using centralized modes the time for a mid distance trip may be similar to, or even shorter than, the total time to take the same trip by bus, train, HSR, hyperloop, or short distance aircraft.
Assuming ‘time’ and ‘experience’ are the two most important drivers of which mode of transport you choose, then the self driving car can disrupt everything above and to the right of its position on this graph. Consider a trip from San Francisco to Los Angeles of approximately 400 miles. You can take a plane, a train, a bus, HSR or a hyperloop (of the future). Or, you can get direct end-to-end service with no hassles, and approximately the same total travel time, from a self-driving car. Not only would HSR or hyperloop infrastructure be insanely expensive, it would unnecessary. The self driving car makes them moot, and further, may dramatically impact air travel between the two cities, as HSR has done in some corridors.
Let’s add Self Driving cars and Hyperloop to our time / distance graph.
In the assumptions we assume that dedicated self driving infrastructure is deployed. That allows self driving cars to travel slightly faster in-city, and significantly faster on freeways and between cities. On many routes it is tough to take a flight that takes less than 3.5 to 4 hours from home to final destination. In that same timeframe, a self driving car can go 400 to 500 miles! That changes the game.
Our assumptions around Hyperloop is that overheads are similar to an airport. You need to get to the central station, you need to leave time to check in, go through security, etc. The impact of an incident on a hyperloop will be similar to that of an airplane, so similar security protocols will be required. In fact, with hyperloop infrastructure accessible across more distance, and potentially running near high density areas, there may be more security for hyperloop than for air travel. So, taking a hyperloop, HSR, or plane over these distances is a toss up. Given there is more plane infrastructure already deployed, that would be the logical choice.
Hyperloop, from this perspective, is nowhere near as impactful as self driving cars. It will be relegated to niche applications, if it makes sense at all.
Let’s switch and look at cost per mile? Self driving cars are estimated to be 10% to 50% the cost of today’s car, per mile. Many are pegging it at $0.20 or lower. For this length of trip that is a lower cost than air travel! Some people claim that they will never give up driving, but if self-driving is four times cheaper than driving your own car, many will. Hundreds of years ago you built your own home; only hobbyists do that today. In twenty years hobbyists will still drive, but most of us will spend that time doing other productive things. (For a deeper analysis on costs, see the RethinkX report).
The cost difference between Self Driving Car, HSR, Hyperloop and Plane are very similar across these distances. It will probably not be the deciding factor.
Next, safety. We don’t know the impact yet, but the fact that self driving cars are always alert, respect traffic laws more robustly, have 360 degree vision across multiple spectrums, and can talk to infrastructure means that they will be significantly safer than existing human driving. However, we saw above that buses, trains, and airplanes are significantly safer than existing cars per passenger mile. It will be a challenge for self driving cars to close that gap, and this may not be fully achieved until all human drivers are banned from self driving roads.
Finally, let’s attempt to model the overall ‘experience’. We noted earlier that you have a lot more productive time in self driving cars; from the time of pickup to dropoff you don’t need to worry about interacting with any other services. A far cry from the other modes. Thus, you may be fine to take a longer ride in a self driving car than you would in a plane or hyperloop. Spending six hours productively may be better than a total trip time of four hours, three of which you spending fighting infrastructure (and other passengers).
The following is highly speculative, and different people will have different thresholds. Here we balance total time with productive time. At short to mid distances — up to six hours of travel — total time is less important than productive time. Above that, total time is the most important.
In this model, two modes will dominate the future: Self driving cars and planes. For trips under 800–1000 miles we will take self driving cars; it may take 6 hours, but we will work, sleep, play, exercise or socialize as there are no messy overheads to deal with — it will be productive time.
For longer trips we will take planes. Because of the speed advantage of planes over longer routes, both Hyperloop and HSR fall away (assuming Hyperloops max out at 300–350 mph as currently feasible, as opposed to the theoretical 700 mph originally proposed) They have the same hassles, but take longer. Assuming a region commits to self driving infrastructure, there is really no room for hyperloops. Tunnels, such as those proposed by The Boring Company, should be considered part of a self driving infrastructure plan. They may well make sense in highly congested areas.
Stop laying rail! Don’t build hyperloops.
So, what is the net of this analysis? We focused on medium to long distance trips above, but comparing self driving to LRT or buses for shorter distance trips is also compelling. Self driving will be cheaper, faster, and a much better experience than a bus or LRT.
- Self driving cars are going to be more disruptive than most people expect. The combination of the experience delivered (end-to-end service with no ‘central infrastructure’ hassle), reduced time, and exceptional cost/mile will have self driving cars replace more centralized modes in many use cases and for many people.
- Passenger rail — including high speed rail — dedicated bus infrastructure, and hyperloop will make no sense in five to ten years. Given how long it takes to deploy these systems, and how static they are, cities should be stopping development now.
Self driving bicycles, segways, and skateboards.
If we consider even shorter trips — say a mile or two, we already see another trend beginning. Motorized scooters, next generation segways, and electric bicycles are all filling niche areas. Imagine, for a moment, the self-driving versions of these. Instead of wasted bike racks across the city, bikes (maybe more like single person pods) arrive when required and disappear once you are delivered. They will take very little space and cause very little congestion. This could be the majority of inner city trips and utilize existing and extended bike lanes.
As today’s decentralized modes get smart they will displace more centralized modes.
The ‘smart’ versions of our existing modes of transport are going to eat into the sweet spots of our current centralized systems. There will still be a need for flights, but not for short routes and even some medium length flights will be displaced with self-driving bedrooms — a pod that picks me up at 9pm and delivers me to my destination at 7am after a nice long sleep. In fact, if the route is less than eight hours long I will instruct my sleep-pod to go slower — just fast enough to get me to my destination at the right time.
Of course buses, LRT, trains, and short distance airplanes are not going to disappear; however their ‘sweet spots’ are going to get smaller. We will still have many modes to choose from.
The technologies making this possible represent an inflection point in our development. We have been building more and more centralized systems for hundreds of years. Now the decentralized versions are going to become more efficient, more convenient, less expensive, and better user experiences. The implications of this will be more significant than most people realize.
More travel, more congestion?
If self driving infrastructure becomes so inexpensive, safe, and easy to use, we would expect to see more people use it more often. This may be counteracted by a change in how we work and live, with more telework and people living closer to their jobs…but we can’t count on that. So, we should consider that both time and experience could be compromised by congestion. This may well be a commitment and transitional issue: will cities and states commit to deploying dedicated self-driving right-of-ways? How long will that take to deploy.
If we take a long view where dedicated infrastructure is built or allocated, self-driving should allow for significantly less congestion, even with a dramatic increase in travel. A lot of optimization will come from packing vehicles tightly together, forming them into platoons or herds, and having them travel the same speed whenever possible. It is easy to imagine a modular system, much like containerization made shipping so much more efficient and effective.
- 1x1 pods that carry a single person or small packages
- 1x2 pods that carry two people, or one person and some luggage. Two people with luggage can also schedule a 1x1 to follow them if the 1x2 is too small.
- 2x2 pods for four people, fewer people with more luggage or mid size package delivery.
- 2x4 pods for food or furniture delivery
- 4x4 pods that can be stacked into containers for longer distance shipping
- Etc.
These pods will not require bumpers, side mirrors, trunks, or bonnets. Especially at inner city speeds, the system will be safe enough that you will not need three tons of metal to protect you.
With a modular design, and little need for inter-vehicle space, cities can pack many more lanes onto existing infrastructure, and do intelligent route planning based on peak inbound or outbound times. Traffic spacing, both front-back and side-side can be very tight; perhaps as little as a few inches.
A typical overhead look of traffic in 20 years may look more like the diagram below than a similar shot today where one or maybe two cars would fit. The herd below could be moving at 20–30 miles per hour in a downtown area, and reconfigure on the fly so that vehicles that need to pull over or merge in can do so. We should be able to fit 5–10 times as many people into our existing roadways as we do today.
Skeptics should remind themselves that self driving technology is probably following an exponential curve; in ten years it will not be a bit better, it will be hundreds of times better. So, if cities decide to keep up, the 20 years mentioned here is highly pessimistic. The cars will be ready for this challenge only a few years from now.
A modular, ‘pod’, approach may or may not work, and may take more force of will than cities and infrastructure planners have. The point is that once vehicles are self driving we can throw out a lot of our existing pre-conceived notions of what a car is. Most trips do not require trunk space — so instead of having storage all the time, you will just have a storage pod on demand; the storage pod will follow your personal transport pod. A 1x1 pod may be closer to the size of a segway than a small car. They will pack in tight; they have no need for a three car length safety rule.
Platoons or herds will make almost as efficient use of space as existing buses or LRT — perhaps even more. In fact, given loading levels on today’s infrastructure, a modular system may get much higher densities.
Intersections will not have timed lights, but rather demand based routing. A global view of the system can equalize traffic across multiple arteries, and schedule herds to minimize conflicts at intersections.
Smart engineers — or perhaps more likely smart AI’s — will figure out ideal packing, reconfiguring, and routing algorithms to handle many times our current traffic loads with little or no congestion.
The hard part of this vision is getting started. That is where infrastructure decisions and investments should be focused. Today we provision bike lanes and dedicated bus lanes. Tomorrow we should start provisioning self driving ‘spots’ (where self driving cars can pick up and drop off) and have them coexist with bus stops and centralized stations. Today we have LRT tracks, which are usually empty. Self driving pods could use those right of ways and ensure they do not interfere with existing trains. Instead of ordering more trains, cities could order self driving pods….or at least LRT with wheels instead of rails so that they can be optimized without laying new track. A little bit of imagination will go a long way.
Of course this is unlikely to happen smoothly, and vested interests will impede it. However, in cities that are developing brand new areas, or in newly designed cities, we would expect this type of configuration to be considered.
Ecosystem business model
Finally, the economic and business models for self-driving pods may be quite different than what we experience today. Instead of ‘purchasing’ minutes from a company — Uber, Lyft, Ford, Tesla, whoever — we may join a new ecosystem, say Pod, where all participants transact in Pod Coins, vote on updates and changes to the system, and manage the reputation of participants and users. Many participants will join, from car manufacturers to cities and governments, users, services…and some speculate Pods themselves may be motivated by service goals as part of their programming. The Pod Coin economy, built on top of a blockchain and smart contracts will better align all participants motivations, and generate a more meritocratic distribution of proceeds than today’s centralized firms. These models are not yet worked out, but there is lots of experimentation in building new economic models (many based on mechanism design), and on more distributed governance models. We will see new models of capitalism that will challenge today’s models.
It is important that an ecosystem model be worked out here. Self driving may lend itself to a winner take all model through network effects, which would not be ideal, for several reasons. First, it is a complex multi-stakeholder model, and all those stakeholders will want some input into the system. Second, a huge number of centralized jobs will be displaced as this transition occurs (rental cars disappear, no more bus drivers, oil and gas needs diminish, complex internal combustion car manufacturing dies away, etc.). Some of these jobs will be lost through automation — as AI and robotics displace repetitive specialized work — and some will be lost as modes of transport that use humans are reduced. In an ecosystem model there is the potential to generate new types of work — less repetitive, more decentralized. In-car entertainment and other services for while you are en-route, new types of meeting places made feasible by low cost self driving cars, etc. It is difficult to map these new jobs out now, but it is important that we build architectures that will allow them to evolve and grow within a healthy ecosystem.
Recap
Systems of Controls, which maps related systems by how decentralized or centralized they are, is an interesting way to look at the future of personal transportation. The emerging features of self driving cars, especially the user experience of seamless end-to-end travel with no hassles, points to them taking large bites out of existing centralized modes: from buses to trains to planes. Given this perspective, laying rails or building hyperloops does not seem like a good use of resources. Instead, accelerating the advantages of self-driving pods, and reconfiguring existing roads and right of ways to integrate them would be smart. If tunnels also need to be dug, that may also be feasible.
Personal transport is an example of a broader trend that we see. Today’s technologies — from GPS and AI’s to block chains and Ethereum are going to reverse the trend of ever more centralized systems. As decentralized designs become more cost effective and safer (including trust relationships and contracts) they will start to cannibalize or replace their centralized counterparts.
This fundamental shift in how systems are built will provide huge opportunities for startups, and huge disruptions for existing industries. This change will occur not only in personal transport, but in energy, food, financial services, and many more.
Thanks to Shawn Abbott for many thoughtful comments on this note.
Appendix: Assumptions
Here are the assumptions driving the graphs above. While the analysis above is sensitive to these numbers, it is not ‘highly sensitive’. Even if costs or times are off by a 25–50%, the basic argument still holds. The base assumption is that investment is put into self driving infrastructure instead of other options. The better ‘experience’ of the decentralized car experience drives this top level assumption.
- Self driving in city speed is assumed to be higher because of the investment in that infrastructure; with dedicated lanes / tunnels the average speed will be faster than today’s rates. 25 mph is optimistic given traffic in many cities.
- Recommended 1 hour prior to flight arrival times are assumed for planes, with similar times for hyperloop.
- Distance to and from centralized stations is from a small sample set of cities, including San Francisco, New York, and Los Angeles.
- Productivity percentages, which represent how much free time you have for different modes are high level estimates based on personal experience.
- Costs per mile can vary significantly for cars, buses and trains. These are averages based on looking at multiple sources.
- The fixed overheads and variable cost overheads were estimated based on checking ticket prices for common routes in economy class for airlines, trains, and buses. They can be viewed as the minimum expense. In the case of the self-driving car we use an approximation of the lowest fares on Uber and Lyft today.
For a given geo, a more detailed model certainly makes sense. The driving argument for this article — that the decentralized experience, where the user controls more of their time and spends less time dealing with centralized systems — may well have influenced these assumptions.