Tesla Model S
Tesla Model S

For the past couple of years, we’ve seen a substantial amount of research committed by the tech and auto industry into self driving vehicles. Billions of dollars are being spent on a solved problem; depending on how you view the problem. If the domain of the problem is transportation and increasing populations within metropolitan areas, automated driverless trains have been a reality for quite some time. I’d argue that solving the domain space of individual automated cars, while contributing significantly to the fields of machine learning and computer vision, is a wasted effort when it comes to sustainable transportation solutions for the planet.

Recently Tesla suffered some major1 setbacks2 by what they have been mislabeling as self-driving vehicle3 accidents. Currently no consumer level fully autonomous vehicles are driving on the world’s roads today. Only test vehicles and pilot programs such as Uber’s driverless system in Pennsylvanian, which recently suffered the embarrassment of one of its vehicles driving the wrong way down a one way street4.

The idea of driverless car technology draws on our romanticism from science fiction. It brings us one step closer to a world where physical labor is phased out by robotic counterparts. Having fully autonomous vehicles on traditional roads would be a technological marvel, but it would only solve some transportation problems.

Automated Trains

Front of a Driverless Train on Singapore's Mass Rapid Transport
Front of a Driverless Train on Singapore's Mass Rapid Transport

In London, the busiest lines of the subway system, namely Victoria, Jubilee, Central and Northern, are all automated. The Victoria line has been automated since it opened in 1968, while the Central line converted to automated operations in the mid 1990s and the Northern Line in 2012. Although many of the automated underground trains still have engineers in the cabs who simply push buttons to close/open doors, the Docklands Light Railway (DLR) line has never had drivers since it became operational in 1987. By the mid 2020s, London plans to roll out a fleet of 250 trains on existing subway networks which can be fully autonomous, removing the need for a driver entirely5.

Singapore Mass Rapid Transport (MRT) system is also entirely automated and has a daily ridership of 6.9 million people6. Vancouver, Canada boasts having the longest automated train system in the world with a daily ridership of nearly 400,000 people. Other automated lines can be found in Barcelona, Paris, Nuremberg, São Paulo, Beijing and many other cities throughout the world.

The United States does have automated train systems, but the majority of them are inter-terminal trains at major airports. There are some commuter automated train systems such as the Morgantown Personal Rapid Transit system in West Virginia, the Las Vegas Monorail and the Miami Metromover.

Automated rails systems are a reality, and have been for several decades. Although they vary in the degree of automation, there are many safe, fully automated driverless trains with massive transport capacity that literally move millions of people throughout their cities every day; far more than could be moved by traditional buses or car networks.


Singapore’s North East Line is a completely underground and fully automated rail transport system. At a cost of S$4.6 billion, its sixteen stations have a daily ridership of over half a million people7. Chicago’s two busiest trains, the Red and Blue lines, have a daily ridership between 180,000 and 250,000 people8. Exact daily ridership figures for other systems are difficult to find as most cities publish rider per day/year metrics for their entire systems and not individual lines. Still, many large metropolitan areas move millions of people per day via commuter rail.

According to a study published by the Oregon Department of Transportation in 2013, the theoretical maximum throughput of a single lane of traffic, with passenger vehicles filled to capacity (4.2 persons) is 227,000 people per day. If you commute to work in a US city, you may notice that the high occupancy vehicle (OVH) lane with the diamond moniker is mostly empty. This suggests that most vehicles, like yourself if you’re stuck in traffic, only have one passenger. The theoretical maximum for single occupancy (1.4 persons) vehicles per lane is only 76,000 people per day.

Roadway People per day per lane (pplpd)
2008 AADT from I-405 in LA 38,000
2008 AADT from I-84 In OR 39,000
2008 Counts from 401 in Canada 48,000
Theoretical Veh Capacity (1.4 persons) 76,000
TCQSM Seated Capacity on Arterials 139,000
TCQSM Max Capacity on Arterials 209,000
Theoretical Full Veh Capacity (4.2 persons) 227,000
6 Walkers per lane 500,000
Theoretical Bus Capacity (TCQSM) 1,440,000
Theoretical Bus Capacity (OR Buses) 1,980,000

Source: Maximum Theoretical Person Capacity in a 24 Hour Period, Oregon Dept of Transportation9.

These theoretical maximums assume a lot of extremes such as the volume of traffic never decreasing, all vehicles moving at constant speed and all vehicles being of the same type and capacity. The observed maximums is considerably lower, in the range of 40,000 to 50,000 people per lane per day, depending on the freeway being observed.

Data Graph from Oregon Department of Transportation Study
Data Graph from Oregon Department of Transportation Study

Even the theoretical maximum bus capacity, making all the same assumptions, tops out between 1.5 and 2 million people per lane per day. Chicago’s Red and Blue train lines both transport the same amount of people per day on their respective two track systems than the theoretical maximum for a single lane of carpool traffic. Singapore’s North East Line, also utilizing two dedicated tracks, has a higher actual capacity than the maximum theoretical capacity of over two lanes of freeways consisting only of carpool lanes, and six and a half lanes consisting of single capacity vehicles.


Current estimates put the cost of a Google driverless vehicle at $300,00010. Obviously the research and development cost of a car isn’t comparable to the retail cost once an item is mass produced, but it does indicate that a lot of money is being put into the technology behind these vehicles. General Motors has invested over a billion into self driving vehicles with the creation of a tech center in Detroit specifically for autonomous car research11. Even the US government is investing over $4 billion of tax payer funds into autonomous car research over the course of the next ten years12. That’s a lot of tax payer money to spend on research that may or may not lead to viable self driving vehicles, as opposed to spending that money on trains which we currently do know how to build.

The cost of rail based public infrastructure is high, but it’s for solid technology that exists throughout the world. The amount of both public and private funding placed into self driving vehicles and infrastructure takes away from potential funding for fixing traffic congestion in the present. At the very least, public funds shouldn’t be given over to private closed research by technology and automotive industries, but rather be spent on existing transportation infrastructures that has a far greater reaching benefit for all people in a much more achievable time frame.

Software Licensing and Car Ownership

Tesla Model S (white)

Self driving cars are safest on roads with only other self driving cars. In such a system, you would no longer have a need for stoplights at intersections, except at pedestrian crossings. Cars could effortlessly flow by each other in an intersection, with complex navigation systems communicating velocity and direction instructions seamlessly between each vehicle.

But in such a system, autonomous car streets would need to be separated entirely from non-autonomous vehicle streets. Software improvements would need to be distributed to every vehicle for safety reasons. Security would need to be under continuous audit. Testing would need to be held to the same standards used for medical devices and airplanes. One rogue hacker or one bad software update would at best, cripple transportation and, at worst, kill an unfathomable number of people.

It may even make sense, in such a hypothetical system, to not allow car drivers to own their vehicles, but instead lease them from an organization that would assume the maintenance, as well as the liability, of operations. To allow car ownership would require a governing body that ensures vehicles have all their required updates and haven’t been tampered with. Modifications could potentially be very limited and the issue as to whether drivers truly own these vehicles comes into question. At this point we’ve closed the loop, and now you effectively have a public transportation system (although it could be privately owned, as many public transport services are).

If people were truly allowed to own their self driving vehicles on either autonomous-only or shared roads, there would be huge legal and liability concerns surrounding what one could and couldn’t do to their own property. It would tear into the core of the long standing hardware/software ownership debate in the open source communities.

Furthermore, if every manufacturer had their own self driving algorithms and implementations, would potentially life saving improvements be required to be implemented by other manufactures? Would there be universal test cases that each manufacture must pass for each software iteration (e.g. a miniature DARPA Grand Challenge which every car must pass on every potential software update)? Should all self driving code be required to undergo third party auditing? If When people are to be able to crack the security on the software controlling their vehicles, would they be criminally labial in cases where modified software leads to an accident?

Open Driving

It’s important to mention that there are open source automated driving tools in the wild, most which come with pretty big disclaimers. There’s Autoware, a BSD licensed suite of software which include computer vision, learning, acceleration/breaking/steering and simulation tools13. Openpilot is another research project that supports autonomous driving on two real world cars. Both projects have disclaimers about how the software is only for research and that individuals are responsible for complying with local laws14. Yet video can be found of Openpilot in use on active roadways15.

Commercial autonomous vehicles use real time operating systems with very low latency. Openpilot is a combination of visiond and python scripts running on an Android cell-phone. The fact that there is video of it being tested on a real roadway should be of great concern to people.

I don’t want to discourage research in the field of computer vision, and these open source projects do provide a lot of knowledge and research that’s accessible to the public. In fact, safety might be improved by regulating that companies do open parts of their software for inspection; allowing independent researchers to establish a baseline of safety and industrial standards for driving algorithms.

Still, students and researchers have been participating in competitions, such as the DARPA Grand Challenge, for years. Competitions allow researchers to experiment on closed tracks in controlled environments, where as these open source projects have researchers testing their devices on highways.

Legal Status

The legal issues surrounding autonomous vehicles are vague, with Uber disregarding the law and running unlicensed autonomous vehicles in production on the streets of San Fransisco16. Uber has a history of simply defying laws to establish their business in new regions. Where as in the past this has been framed an issue of workers’ rights or unfair monopolies, unregulated self driving cars go well beyond labor laws in their potential for damage.

In the US, laws regarding autonomous vehicles only exist in California, Florida, Nevada, and the District of Columbia. Since other states don’t explicitly address computer driven vehicles, self driving cars may or may not be legal in other jurisdictions. This vague legal gray zone is exactly what Google exploited for their early prototype vehicles17.

Before we’ll ever see production self driving vehicles in the hands of consumers, there are several legal questions that must be answered. Who is responsible for maintenance and security updates? Will owners be able to modify the software on their cars? And there’s the most important question, who is liable in the case of an accident, or a fatality?

Buses vs Trains

Buses may have more flexible routes, but they have limited capacity and still must share motorways with other vehicles. Some cities have express buses with dedicated lanes and stations. These systems remove the bottlenecks of shared roadways at the expense of flexibility, while maintaining the limited capacity compared to trains.

Even fully electric trolley buses on dedicated express ways still have the energy costs of tires, and pale in both capacity and maintenance costs of equivalent rail systems. Most express bus systems must drop from dedicated to shared lanes within metropolitan areas at some point, making it impossible for them to compete with high capacity rail systems. Some of the world’s more complex light rail system have trains arriving every two minutes or less on each platform during peak hours. This frequency allows them to approach the theoretical maximum carrying capacity of rail networks far more easily than with buses.

Autonomous self-driving buses would only solve the issues of labor. Although they have the potential to increase safety compared to human drivers just like with self-driving cars, their energy costs are still considerably higher than trains. At constant speed on level ground, pulling the same load, any steel wheeled rail engine in motion will only use 5% of the energy required by a tire based road vehicle in motion. Even with starting and initial acceleration, steel against steel vehicles only use 10% of the energy required by any large pneumatic tire road vehicles. What’s even more counter intuitive is that only in the case of railroads, train resistance (rolling resistance) is inversely proportional to the train weight. This mean that the heavier the train, the more energy efficient it becomes18.


Moving people from point a to point b in America is becoming increasingly difficult. Traffic and gridlock, as result of single passenger per vehicle highway systems, simply not scaling to meet population growth. Where large cities such as Chicago, New York City and Washington, D.C. have met this demand with by maintaining and expanding their networks of light and heavy rail, the majority of American cities have removed their light rail systems to make way for unsustainable car based transport.

Every city I have lived in had streetcars or trams at one point in time. From Chattanooga to Cincinnati, street cars carried large numbers of people for decades. Companies such as General Motors, Firestone Tiers, Ford and others, all pushed for bus based transportation and the removal of light rail19. The ability for buses to change routes seems like a positive, except when contrasted with their limited carrying capacity and non-dedicated right of way.

Trains can be easily automated, and there are several fully autonomous, safe rail networks around the world. A well constructed rail network does take a considerable about of construction time and funding, but the payoffs are large, scalable transportation systems that can move people at greater efficient, lower cost and lower pollution that cars or buses.

In places like Europe, autonomous vehicles could help bridge gaps in transport in areas that are already serviced by extensive and complex rail networks. However in places like the United States, they will not solve core transportation issues in a truly sustainable way. Even if we could mandate dedicated roadways just for fully autonomous vehicles; road networks without traffic lights and computer navigation based intersection exchanges, the problem will eventually hit a bottleneck due to the limited number of people per vehicle.

The capacity of such a system would still be below that of a rail network, and would essentially create a separate system for those who could afford self-driving cars. Alternatively, self driving vehicles would need to be owned and maintained by a central agency and individual users would rent time on them similar to Taxis, Zipcar, Uber and Lyft.

“Last thing I want to see before I die

Is the flash of twenty two inch chromes in my eyes

In America, in America

They’ll bury us with our cars”

-Bury Me with My Car, Ben Sollee (song)20

Self driving cars are cool. They are the romanticism of the science fiction world we were promised in our books and literature. However as cool as they may seem, they don’t solve fundamental transportation issues in ways that will scale for the future. Large scale human transportation is a solved problem in most of the world, and is continually being improved upon. Instead of continually dumping funding into autonomous vehicles, the United States needs to create solid rail transportation systems, both within cities and linking cities via high speed rail. With America so far behind in mass transit, now would be an excellent opportunity to invested in pure automated rail technology. It would be a substantial leap, moving the US from far behind our western counterparts to leading the world in autonomous train technology.

  1. Preliminary Report, ​Highway HWY16FH018. 26 July 2016. National Transportation Safety Board.

  2. Now a Third Tesla Crash Is Being Blamed on Autopilot. 11 July 2016. Silvestro. Road and Track.

  3. Germany Says Tesla Should Not Use ‘Autopilot’ in Advertising. 16 October 2016. Reuters.

  4. Uber’s self-driving cars are already getting into scrapes on the streets of Pittsburgh. 4 October 2016. Griswold.

  5. ‘Driverless’ Tube trains: See inside TfL’s new fleet for London Underground. 9 October 2014. Eleftheriou-Smith. Independent.

  6. LRT patronage up 10.9% and MRT up 4.2%; bus passenger trips rise 3.7%; cab trips down. 10 March 2016. Tan. Straits Times.

  7. Overview North East Line. Retrieved 18 December 2016. SBT Transit. Archived Version

  8. Annual Ridership Report Calendar Year 2012. 12 April 2013. Chicago Transit Authority.

  9. Maximum Theoretical Person Capacity in a 24 Hour Period. 27 November 2013. Bettinardi and Prusakiewicz. Oregon Dept of Transportation.

  10. Google’s Trillion-Dollar Driverless Car – Part 3: Sooner Than You Think. 30 January 2016. Mui. Forbes.

  11. GM to spend $1 billion on self-driving tech center in Detroit . 24 June 2016. Curry. ReadWrite.

  12. U.S. Proposes Spending $4 Billion on Self-Driving Cars. 14 January 2016. Vlasic. New York Times.

  13. cpfl/autoware. Github. Retrieved 19 December 2016.

  14. commaai/openpilot. Github. Retrieved 18 December 2016.

  15. Hack Autonomous Driving into Your Car with Open Source Hardware Comma Neo and Open Pilot Software. 1 December 2016. CNXSoft.

  16. Uber continues self-driving vehicle testing in SF in defiance of DMV. 16 December 2016. Conger. TechCrunch.

  17. Are Self-Driving Cars Legal?. Retrieved 18 December 2016. HG. ORG. Archived Version

  18. Why Rail Has 20X Energy Saving Advantage Over Rubber Tire Road Vehicles - The Science of Locomotion. Retrieved 19 December 2016. Brooklyn Historic Railway Association. Archived Version

  19. Kennedy, 60 Minutes, and Roger Rabbit:Understanding Conspiracy-Theory Explanations of The Decline of Urban Mass Transit. 17 November 1998. Bianco.

  20. Bury Me With My Car. Retrieved 19 December 2016. Sollee. (Song Lyrics)