The death of the autonomous vehicle dream and the future of transport

Billions of dollars have been invested in autonomous vehicles to date. Eighty billion dollars spent, and for what?, asks Lloyd Alter, writing for Treehugger.com. It’s a fair question. The article makes a key point: We are not going to replace 95% of privately owned cars with shared autonomous vehicles anytime soon. Why does this matter? There are a couple of key consequences of this, I think pretty uncontentious, statement. 

1. Any cars with any autonomous capability will inevitably be limited in what they can actually do and where they can do it, and they’ll be expensive. They are niche, if not gimmick. 
2. We then come to the question of what is all this effort for? Why are we chasing the autonomous vehicle holy grail? Waggishly, we might suspect it is Silicon Valley’s typical approach to fixing the Bay Area’s transport infrastructure woes.

The Case for Autonomous Cars

There's a reasonable summary of the case for autonomous cars in this CIOReview article. We can imagine that the ever-constant vigilance of an automated system that doesn’t get tired or distracted will be safer (under certain conditions). There’ll be less traffic, lower cost, and less congestion. Let’s pause and unpack those points a little. Less traffic, as in fewer cars, is nothing to do with whether cars are autonomous, per se. It’s simply a consequence of other factors of the transport equation. Getting autonomous capability into every car and everyone continuing to own private cars, with the same journey patterns, will make no reduction in traffic volume. It’s true that driving at a steady pace, avoiding the stop-start of those frustrating compression wave jams, will improve fuel economy and reduce pollution. But self-driving cars are not a necessary condition for this.

The proposed cost reduction of fewer accidents may have less to do with cars being autonomous and more to do with having many fewer cars on the road. Improved traffic flow really just requires better networking between vehicles. This could be automated, but we already have live traffic on our satnavs that allow drivers to make real-time route decisions; all that’s fundamentally required is an additional notification in a driver’s satnav app that recommends a speed that will, for example, prevent such shock waves, and that drivers trust will improve their journey.

Then we have the issue of parking spaces. The fundamental crunch points of the day for congestion are the morning and evening rush hours. The rest of the time, most cars are static in garages and parking lots. Fancy autonomous driving capability isn’t doing much then. Ah, you say, but if the car was autonomous, it could drive off and serve others during these periods. Who would want their expensive, autonomous car going off on its own to carry Joe Public? We either believe that self-driving cars are being developed as a very expensive feature-add for the wealthy, and will do nothing to ease traffic problems, or we believe that it is all about providing a shared communal resource that slashes the number of cars in circulation. What auto manufacturer will be pursuing the latter expensive strategy to shrink its market? Immediately here, we arrive at a conclusion. The entire edifice of the case for autonomous vehicles seems to be predicated on the assumption that the car is the solution.

Yes, autonomous vehicles may well be safer than humans driving (conditions apply), but reduce the number of cars on the road by 90% and you’re going to see a huge drop in accidents anyway even if humans are still behind the wheel. Reduce the number of cars on the road, and how much smoother also would be the flow of traffic? Those multiple billions of dollars have undoubtedly advanced the technology of cameras and lidars, and advanced the field of computer vision. And they’ve most certainly funded many data scientists and machine learning engineers. But what have they done for the transport problem? What, fundamentally, is the transport problem that this work seeks to address?

If we accept that the self-driving car is not the answer to our transportation woes, then what is? Here we can take inspiration from a principle that underlies modern, robust, software systems that scale: decoupling. Making the car the universal solution and trying to couple it ever tighter into our lives is the antithesis of this philosophy; this forces you to think that every solution to the transportation problem must look, and fundamentally behave, like a car. Different modes of transportation already exist, and they have various properties of speed and flexibility etc.

Enter the Mobility Hub

What makes the most sense is not trying to pick one of those modes and shoehorn it into every use case, but instead to allow each to work in their optimum domain and provide a means to integrate them, to decouple the problem of transportation from the separate components of the system.

Such a means of decoupling is the mobility hub. The mobility hub acts as the physical transportation equivalent of the software API. This goes beyond the contemporary concept of the public transport park-and-ride, which is still intimately connected to the concept of private car ownership. Park-and-ride schemes serve mainly to provision large out-of-town parking lots where said vehicles can remain static until their owners return.

How can the park-and-ride concept be upgraded to become the mobility hub? Imagine cars rented by the hour, or even the minute, used just for that last mile component of the journey. But who wants to get into a car someone else has been in and messed up? Well, now we have the creation of a new job: car valet at the mobility hub! Electric cars are eminently suited to such last-mile journeys. Solar panels on roofs at the mobility hub can charge cars whilst they’re stationed there. Through advanced booking data and modeling, excess capacity could even be sold to the grid at peak times.

Instead of trying to navigate your way through the minefield of fixed timetables for various public services, wouldn’t it be great if there could be more flexibility? Trains may be harder to flex, but buses are the stalwarts of the custom mass transport function. Companies such as Zeelo provide such custom bus services and have compiled a really interesting ebook on reducing the reliance on car ownership for the daily commute.

Rather than treating a single service in isolation, however, it would be ideal for a traveller to enter their preferences for a journey and confirm their choice, or choices, which would then be used to ensure the required capacity and even adjust some service schedules. All of this requires infrastructure, which will be enabled by some tech, but nowhere are self-driving cars a necessity.

Takeaways for the Data Scientist

Where does all this leave data science? As a data scientist, I started thinking about that, and it led me to write this article. With billions of funding being ploughed into self-driving tech, and hype about the latest computer vision algorithms, many excited budding data scientists, machine learning researchers, and AI researchers have flocked to this cutting edge area.

Well, newsflash, it’s going to die. It will become the Klondike in 1899. If you rushed to learn some TensorFlow and CNN and YOLO algorithms, you’re going to find your niche experience, once very trendy and in demand, doesn’t get you very far. For the creative data scientist with a diverse set of skills, there will be huge opportunities in the new decoupled infrastructure.

Where should mobility hubs be sited? What services should they offer? What resources are required, and when, to service the expected demand? What jobs will they support (or create)? Can such hubs be developed in more of an “agile” rather than “waterfall” manner, adapting the offered facilities with demand over time and as the transport ecosystem evolves? What does all that even mean? These are questions on which a thoughtful data scientist can provide critical data-informed insights.

For data scientists, amongst others, there will be work on optimising the modes of transport accommodated and offered, as well as facilities to support them. There will be the need to determine bus routes and schedules. This could lead to more dynamic routing and scheduling, requiring algorithms.

Pricing strategies for electric charging need determining; is there an appropriate rate for charging private vehicles? Would rented, or communal, vehicles be charged less (or nothing)? Owners of electric cars, normally parked on the street, could use their local mobility hub as their charging point, increasing electric vehicle uptake by obviating the need to have power extension cables snaking out of kitchen windows and across the road!

The Bottom Line

To reiterate: the dream of the self-driving, privately owned vehicle in general use in the unmodified public environment is going to die. And it will happen in the near future, I think. The attempts are a pointless waste of investment money that could be better used now. Only a reduction in the number of privately owned cars, and a new mix of modes of transport, with suitable (de)coupling between them will solve our current pollution and congestion problems. All of this will provide exciting work for creative data scientists who can do more than lots of hyperparameter search for glorified computer vision tasks. Unlike the self-driving car, this approach truly has the potential to solve our transport problems.