Driver-assist technology in cars has evolved substantially over recent years. Cars can operate in cruise control on highways, manoeuvre into parking spots with minimal driver intervention, and warn of obstacles that the human driver may not have seen. So it’s no great stretch of the imagination to suppose that car controls might before long replace the human driver completely, with fully autonomous operation.
Public transport has already seen the emergence of driverless trains, used on systems from Vancouver’s Skytrain to Beijing’s airport shuttle. Self-driving operation is of course a much simpler proposition for rail systems, where the path is already set by the rails and points, and signals give instructions to start and stop. But there are already prominent examples where self-driving ‘robot’ cars have successfully navigated public roads: the Google (now Waymo) self-driving car has famously done so for years—albeit on one specific set of streets in Mountain View, California)—and more recently a section of freeway in Adelaide was opened up as a self-driving car test site.
Like any new transport technology, however, self-driving cars have attracted some fairly lofty claims. According to some commentators, computers will be so much better than human drivers at piloting cars safely that they will not only eliminate all road trauma but also make traffic congestion a thing of the past (because cars can follow each other more closely and so achieve high-speed movement at higher concentrations). What’s more, by taking away the need to concentrate on driving and freeing up the occupants to do what they like while travelling from A to B, travel by robot car will become so attractive and convenient that no-one will ever want to walk or use public transport again.
The mind-boggling implications are just starting to seep into the public consciousness—the wholesale reduction of traffic, the freeing of urban land space once needed for parking, dramatically improved road safety, a mass boost to productivity, more accessible mobility for the disabled.
It suggests that perhaps our well-established transport debate—road-based private vehicles versus mass public transport—may be the wrong frame for the future.
Kate Burleigh (Managing Director, Intel Aus/NZ), The Age, 24 November 2016
The apotheosis of these claims is surely the bizarre suggestion, in The Atlantic in June 2018, that New York City rip up its subways and use the tunnels for robot cars. (As Jarrett Walker responded in CityLab, the likely effect would actually be to destroy New York economically.)
So far, such claims are little more than idle speculation given that no truly autonomous car is as yet even planned to be offered for sale to the public. A number of technical setbacks in 2018, including in March the first pedestrian fatality attributed to an autonomous car, and in November the admission by Waymo’s CEO that genuine so-called ‘Level 5’ autonomy will likely never exist, have punctured what increasingly appears to have been a bubble of inflated expectations between 2015 and 2017. Of course, even if at some hypothetical future time the technology does exist, it will take even more time before self-driving cars make up a substantial proportion of road traffic.
These genuine problems aside, quite a few researchers have attempted to pick apart the claims and conduct simulations to probe the likely effects of a self-driving car fleet on urban transport. Examples include an investigation of regulatory and traffic management issues by transport consultants Fehr & Peers in 2014, a 2014 MIT case study based on Singapore, and an OECD International Transport Forum (ITF) case study based on Lisbon in 2015. The latter is particularly relevant to understanding the likely effect of self-driving cars on day-to-day urban transport.
Quite a bit hinges too on the way ownership of and access to motor vehicles is organised in the future. On the one hand, it’s easy to imagine most people having their own self-driving car for their exclusive personal use, for the same reasons people like to own cars today. But it’s also plausible that many people will prefer to use self-driving cars on a shared model—what the ITF study calls ‘TaxiBots’ and ‘AutoVots’. This envisages that people would forgo owning personal cars and simply summon the nearest available self-driving car on demand, a model that combines aspects of taxi service, Uber-style ride sharing and car-share schemes.
On the whole, research and small-scale experience to date provides evidence for the following likely conclusions:
- The idea that robot cars might transform ‘road safety’ as we know it had a lot of early appeal. For a while, there was the encouraging fact that Google/Waymo’s self-driving cars had been circling Mountain View, California for years while being involved in very few collisions, all reportedly at low speed and attributable to the human drivers of other vehicles (though others point out the situation is far from clear-cut). But the later record is more ambiguous. Tesla’s ‘Autopilot’ partial automation system has been implicated in at least two fatal crashes. And as mentioned above, the first pedestrian fatality involving a robot car occurred in March 2018 when a person walking slowly across a multilane road in Tempe, Arizona, was hit by a speeding self-driving car operated by Uber. Supporters of the technology maintain these incidents involve ‘transitional’ automation systems that still depend on human intervention in emergencies, and occur against a status quo where thousands of people are hit by human-driven cars each year in the USA alone. Truly autonomous pilot systems for cars (should they ever exist) are not vulnerable to fatigue, anger, impatience, distraction or the myriad other human factors that figure in most real-world incidents. Accordingly, there are grounds for supposing a switch to self-driving cars—if the as-yet-hypothetical technology is sufficiently mature and developed so as to accommodate other modes of transport appropriately—will come with a dramatic reduction in road trauma. If so, public policy support for robot cars over human-driven cars could be justified for this reason alone. Admittedly, there are a lot of ‘ifs’ there. Technology writer Jason Torchinsky points out it’s a fallacy to assume robot cars can only eliminate causes of failure without introducing any new ones themselves, and highlights the lack of any evidence for safety benefits at mass scale.
- If a shared model for robot cars dominates, it appears likely the size of the car fleet could shrink dramatically. The ITF study in particular found that the equivalent of today’s transport task could be handled by between 80% and 90% fewer vehicles, using self-driving cars on a shared model. This is a consequence of the fact that cars typically spend well over 90% of the time parked.
- Closely related to the previous point, robot cars potentially have big implications for car parking. Instead of it being necessary to leave one’s vehicle at one’s destination, the vehicle can be instructed to return to a convenient storage location, or (if it’s a shared car) to pick up the next waiting passenger. This can potentially remove the need for car parking to be co-located with homes, workplaces and activity centres, providing the opportunity to reform land use in car-dependent areas.
- The big downside comes in robot cars’ likely effect on traffic volumes. Simulations and thought experiments alike tend to agree that a move from human-driven to robot cars will add traffic to the roads rather than reduce it. This follows in large part from the (otherwise positive) effect on parking behaviour: individually-owned cars will drive home empty and return empty when summoned later, while shared cars will travel empty between bookings or circle the streets in anticipation of bookings to be made. The most optimistic scenarios assume robot cars carry multiple passengers as a type of demand-responsive bus service, but still predict an increase in traffic. More pessimistic scenarios assume individualised robot car service replaces public transport use, and predict a doubling of traffic or worse. The Fehr & Peers study splits the difference, forecasting a 25% to 35% increase in traffic with a fully autonomous car fleet.
The so-called vehicle-mile problem with self-driving cars has been much discussed in transport planning circles, and is generally agreed to be a major reason why robot cars are no substitute for public transport in cities.
Although self-driving cars could one day help solve some real problems (foremost among them injury and damage due to poor driver behaviour), much of the enthusiasm for them appears to stem from the same misplaced source as other schemes to solve the problems of car travel by tinkering with the way we use cars rather than by developing alternatives to reduce the amount of car use. Whether it’s by carpooling instead of driving solo, or swapping petrol for electricity, or encouraging driving styles that reduce acceleration and braking, or using navigation systems to bypass traffic jams, or swapping personal car trips for Uber rides, none of these ‘motoring hacks’ directly deal with the biggest problem besetting urban transport: that the sheer volume of car travel and the number of cars on our roads means urban driving is neither convenient, nor efficient, nor environmentally sustainable.
We are hardly going to solve the problems of urban transport with a technological fix that’s likely to make this key problem worse; yet that‘s what the ‘self-driving car revolution’ appears to be offering us.
But what of the claims that self-driving car technology will allow more cars to fit on the road? Don’t these claims contradict what was suggested above?
The claims are of two distinct types. On the one hand, self-driving cars are claimed to ‘improve’ traffic flow through their ability to communicate with one another in real time and control their precise speed and heading, achieving flows at higher speeds and concentrations than human drivers can manage. In principle this distributed communication and control can also extend to the road environment itself, so that traffic signals too can respond to real-time traffic conditions.
While a nice idea in principle, this comes up against a few serious problems in practice:
- Actual simulations of these distributed controls generally only perform well on the assumption that the only other road users are other self-driving cars. As soon as one adds human-driven cars, pedestrians or cyclists to the environment, performance deteriorates (if indeed the model caters for their presence at all).
- The intuitive picture of robot cars following each other closely under cooperative control, in what’s often called a platoon formation, is specific to just one ‘use case’ – where everyone is travelling on the same long uninterrupted stretch of road to the same destination. It’s quite well suited to long-distance intercity travel—where congestion is rarely a problem anyway—but this is only a tiny percentage of the overall travel task, which is dominated by shorter trips in urban areas. And unfortunately, when there are large numbers of vehicles crossing an intersection in multiple directions, the most sophisticated networked controls in the world won’t override the basic fact that one line of traffic has to stop and wait for the other.
- Technical improvements in vehicle controls can only do so much to compress traffic flows. Even today, traffic in cities is often at a standstill simply due to the fact that two vehicles can’t occupy the same piece of road, and traffic density can only increase so far no matter who’s driving.
- Perhaps of most concern is that empirical tests of ‘platooning’ suggest the idea doesn’t actually translate well from theory to practice at all! In 2018, Daimler in Germany undertook comprehensive tests of platooning technology for trucks. The first indication that things were not turning out as expected came in September 2018, when CEO Martin Daum told an industry forum that “platooning might not be the holy grail we initially thought”. At the end of the year, Daimler axed the platooning trial entirely. For vehicle operators, the main attraction of platooning is the fuel economy benefit from smoother traffic flow, but while earlier trials had suggested a benefit for older vehicle models, it appears that for vehicles with current technology under real-world conditions the benefit is negligible.
The other way self-driving cars are claimed to reduce congestion has to do with route selection. The ITF study, in particular, found that the increase in traffic volumes with TaxiBots (shared multi-passenger robot vans) and AutoVots (shared robot cars with one user at a time) could largely be managed using intelligent navigation to send similar trips via diverse routes. This causes the occupancy of most streets to increase substantially, which is not without its drawbacks as the authors note:
The fact that both the TaxiBot and AutoVot scenarios lead to an increase in travel on local arterials and the local road network may imply changes in the performance and characteristics of those networks….
[In] the AutoVot scenario without high-capacity public transport…. road occupancy increases by 40 to 50% for all road classes, with the strongest growth occurring on local road networks. This implies poorer performance and possibly congestion. It may also mean a change in the ￼nature of local street traffic resulting from the presence of additional traffic on what were otherwise quieter, less-used roads. Such an increase in traffic could have a negative impact on the attractiveness and livability of local streets, as it reduces their availability for non-transport use.
—OECD/ITF, Urban Mobility System Upgrade: How shared self-driving cars could change city traffic, p.22
In other words, the increased traffic could be accommodated in this scenario, but only by turning every local street into a ‘rat run’ for robot cars trying to avoid congestion elsewhere. This is no more likely to be politically popular than it was in the days before traffic calming was introduced to deter it!
Notably, the increases in traffic volume predicted by these studies generally assume robot cars taking on the same transport task as is handled by regular car travel today. But this ignores the effect of induced demand that we know to occur in all other contexts where travel becomes more convenient. If robot cars really do make travelling by car more attractive—by freeing up motorists to do things other than focus on driving, or by offering the equivalent of a taxi service with zero driver costs—then people will rationally respond by travelling more, and by travelling further. Just as with the attempt to reduce congestion by building more roads, robot cars may simply encourage enough further growth in travel to wipe out any congestion relief that may occur.
Finally of course, it may not be universally true that people will find robot cars more attractive than human-driven cars per se. Many people are uncomfortable with the notion of handing over control to a machine, though this anxiety may subside with familiarity. People whose enthusiasm for car travel comes from being able to actually drive the car (as per the road lobby’s ceaseless ‘people love their cars’ narrative) aren’t likely to be enthusiastic about a vehicle they can’t drive themselves. And for the one in ten (at least) of us who are prone to a degree of motion sickness and are limited in what we can do in a moving car, a journey in a robot car has little to offer over one at the steering wheel of a regular car.
Research has shown a lack of demand for autonomous vehicles—nearly six in ten Americans do not want to ride in one. MIT found that drivers, even millennials, want clever technology to help the driver, not replace them (PDF). Ultimately, driverless cars are part of the tech utopia that nobody wants. But rather than technology, it is money and momentum behind them; they keep the share price up in the face of Google and Tesla.
And it’s a utopia that may never happen. Jeff Speck reminds us that the predictions of full autonomy are decades away. “I would challenge anyone in the automated driving field to give a rational basis for when level 5 will be available,” says Dr Gill Pratt, head of Toyota Research.
—Brian Sherwood-Jones, “Destroying the city to save the robocar”, The Register, 17 January 2018
In summary then, it’s almost certainly untrue that robot cars will have any positive effect on urban traffic congestion. Nor will they exert any magical effect on urban travel that obviates the need to develop public transport networks as a congestion-free and environmentally friendly alternative for moving large numbers of people.
[T]he new transport innovations will not increase the speed of travel. The car of the future will be electrically propelled, have extensive digital functionality and driverless options. But it’s unlikely to make much faster progress through traffic than the car of today.
These new transport innovations will not transform why and where people travel. Rather, they will offer incremental improvement to the quality of our journeys. As the auto industry switches to electric propulsion and develops driverless options, the lack of a transformational offering to car buyers could make it hard to recover the costs of development.
—David Metz, “Driverless cars won’t deliver a transport revolution—and the auto industry stands to lose out”, The Conversation, 18 October 2019
One vital fact often ignored, indeed, is that any advantage claimed for autonomous private cars translates directly to a similar advantage for autonomous public transport vehicles. As one Swedish transport authority has cheekily suggested, autonomous vehicle operation carries even greater promise for inexpensive personal autonomy delivered by a network of robot buses, trams and trains. And as routes are fixed and predictable, the automation task itself is likely to be much simpler for such applications.
Robot cars could certainly be welcomed for other reasons, but they will fit in our cities as an accompaniment to public and active transport—perhaps including networks of robot buses—not as a substitute for them.
Last modified: 17 February 2020