The self-driving car industry is growing up. Valuations of self-driving car companies and private investment in these companies are exploding. Bloomberg reports that private investment in self-driving and connected car companies in the second quarter of 2018 is more than the total private investment in this sector in the prior 4 years combined! Morgan Stanley has raised its valuation of Waymo from 70 billion in 2017 to 175 billion.
But this is only the tip of the iceberg. Below the surface, a major restructuring of the auto industry is underway where self-driving car companies are emerging as the pivotal element in the strategies for future mobility. Over the past years, different approaches to integrating self-driving car technology into auto- and mobility companies have been tried, ranging from various types of acquisitions (GM-Cruise, Ford-ArgoAI, Aptiv(prior: Delphi)-Nutonomy, Intel-MobilEye) to partnerships (Bosch/Daimler, Daimler/BMW, Baidu/Apollo) and go-it alone strategies (Waymo, Zoox, Uber and many others).
Leaving aside Waymo, GM may have found a winning formula, which is increasingly copied by its competitors: When it acquired Cruise Automation in 2016, it allowed the new subsidiary to continue to operate in a highly autonomous mode, its growth and speed largely unencumbered by the rest of GM. Successful collaboration with GM around the electric Chevy Bolt brightened the prospects of both companies and initially led to a significant increase in in GM’s stock price (which since then has fizzled out). In 2018 Cruise attracted a 2.25 billion USD investment from Softbank’s Vision fund. Being able to attract outside investment (as well as employees through stock options in Cruise) while having close connections to the resources of the parent company should be an ideal position for Cruise to quickly shift from start-up/development mode to commercialization. At the same time Cruise is insulated from all concerns related to building legacy cars and from the headwinds that classical car companies will have to face from the revolutionary changes in the auto industry. Other auto makers seem to be copying GM’s strategy. Ford has created a self driving division (which includes ArgoAI) and will also be open for outside investment. Volkswagen seems to have been in talks to buy Aurora, but was rebuffed. Daimler who was an early leader in self-driving technology is relying on a partnership with Bosch but is also splitting the company into three separate parts (cars, trucks, mobility (which includes self-driving technology). This has the effect of insulating the less vulnerable parts of the business (trucks and mobility) from potentially dramatic changes in the auto industry. Only Toyota, which has always been late to the self-driving race has chosen a different path by investing 500 billion USD in Uber, which minimizes its ability to leverage the opportunities associated self-driving car technology.
The last 6 months have shown that the auto (and mobility) industry is now finding ways to channel billions of dollars into the commercialization of self-driving car technology. Given the extent of changes, the capital associated with these changes and the increased ability of translating advances in the technology into actual products and services (which don’t have to be full fledged drive-autonomously-anyhwere solutions but can be very targeted) it won’t take several years until we see the first real impact on the streets…
Since 2014, many companies have applied in California for testing self-driving cars. The list of companies which have received a permit can be used as a measure for the innovation process associated with autonomous vehicle technology. The graph below shows how the number of companies active in California has only increased gradually from 2014 to the third quarter of 2016. A steep increase follows in 2017. The slope softens in the first half of 2018.
Of course it would be premature to conclude that we are already seeing the beginning of the end of the S-curve which is so typical for innovation processes. And the California AV permits can only be see as a proxy for the larger distributed self-driving vehicle innovation process currently unfolding across the world.
But neither the number of California AV permits nor the number of companies providing self-driving vehicle solutions will grow indefinitely. The time will come where the industry moves into the next phase, where the exploratory modes of development will be replaced by more systematic, managerial approaches and where commercialization will be the primary focus. A shakeout is inevitable. Time may be running out for those who still want to jump onto the self-driving car train…
– 58 permits have been issued by the end of June 2018; One permit has not been renewed (Uber), making it 57 active permits.
The two recent fatal accidents with self-driving cars by Uber and Tesla have not led to the major backlash which many people had predicted. While this does not come as a surprise (the predictions ignored the long history of technical innovations, where accidents have rarely slowed or even halted the advance of a technology), nevertheless, the two harrowing accidents increase the concern of the public and of regulators about the safety of self-driving cars.
Therefore this is the right time to perform a more careful analysis of the risk profile of this technology. As we will show in the following, the specific forms of risk, accident scenarios, and risk mitigation strategies for self-driving cars differ very significantly from other technologies that have been developed over the last centuries. To illustrate the differences, we will examine three key aspects of the risk profile of self-driving car technologies and contrast them with established technologies:
1) One- or two sided distribution of safety outcomes
Self-driving cars are an unusual product from the perspective of safety-related outcomes. Practically every product comes with the risk that it’s use may inflict harm under some circumstances. For most products the safety related outcomes are either harm (negative outcome) or no effect. A much smaller group of products can also lead to positive safety-related outcomes – their use increases safety. A self-driving car will prevent some accidents (positive outcome) or cause accidents (negative outcome); this two-sided distribution of safety outcomes contrasts with other product categories such as microwaves, coffee machines or electric drills which have only one-sided safety outcomes. From one perspective, products with two-sided safety distributions are preferable over products with one-sided distributions. But they present a challenge for risk analysis and for ethical considerations because uncertainty about the distribution of negative outcomes may need to be balanced against the certainty of positive outcomes. Delaying the use of self-driving cars for too long may cause harm (accidents that would not have happened).
In the health sector, this dilemma is a well-known problem for the approval of medical treatments. And the US Food and Drug Administration (FDA) has worked hard to balance both sides of the distribution (both by speeding up the approval process and by enabling critically ill patients to get access to experimental treatments in certain cases). But self-driving cars differ from medical treatments in a very positive way: Whereas the expected positive effects of a treatment often do not materialize (uncertainty on the positive part of the distribution), there is much more certainty about the positive safety outcomes of self-driving cars (accident prevention) and we already have statistical data for the safety benefits of some driver assistance systems.
Thus any legislative effort for regulating the approval of self-driving cars, needs to consider both sides of the distribution of safety outcomes.
2) Alignment of safety goals with development goals
For most products, safety is not an innate part or consequence of the development process. Over the last century we have learned the hard way that a large body of laws and regulations are needed (which then lead to well thought out internal processes) to ensure that safety is adequately addressed in all phases of the development process.
However, the situation is different for self-driving cars. For anyone developing an autonomous vehicle, the primary and overarching development goal of self-driving cars is to be able to operate the vehicle safely at all times. Driving as such is NOT the primary goal, it is a secondary concern because just navigating the car on the road and keeping control of speed and direction is only a very small part of the development problem.
The internal state of the car at any given moment is most important, because the car needs to constantly monitor its environment, identify road signs, traffic lights, predict actions of other traffic participants, etc. Therefore the main concern of development teams is to make sure that the car has a complete and accurate internal representation (of state and probable behavior) of what is going on around it. The key metrics in the development process are not just driving errors but their much earlier cause – shortcomings in sensing, interpretation, prediction. Thus the development of self-driving cars is a constant and intensive search for failures, potential errors, potential flaws. As a consequence, even in the absence of any safety regulations, it would not be possible to develop a self-driving car for the market without being constantly focused on safety. Of course, this is not a guarantee that no mistakes will be made. And this is not a guarantee that the development process will lead to absolutely flawless vehicles (that is not possible). But the technology of self-driving cars is one of only very few technologies where safety issues are inherently the primary focus of development.
3) Efficiency of recall process for defective products
Self-driving cars are almost unique in another, third dimension of risk: For most technologies it is difficult to prevent harm once a defective model is released to the public (and this has important implications for regulation). Once an Espresso machine, a drug or another product reaches the hands of thousands or millions of users it is very difficult to ensure that a defective product model will not lead repeatedly to harm somewhere. Recalls take time and rarely reach all owners. Again, the situation is very different for self-driving cars. They incorporate wireless communication and update mechanisms that allow the near-instant grounding of defective vehicles models. A worst-case scenario where a flaw is discovered after tens of thousands of vehicles have been released to public roads is not realistic: when accidents point to the flaw, the other cars on the road will quickly be grounded and thus further accidents will be prevented from happening. Of course this does not mean that standards for approving self-driving cars should be lax but rather that we should keep the likely risk scenarios in perspective, when we consider regulations for self-driving cars.
In summary, the risk profile of self-driving cars is quite unusual because it is positive on the following three dimensions:
— With self-driving cars, safety is the primary development objective and focus, it is an inherent part of the development process and can never be just an afterthought or constraint of the development process
— Self-driving cars have double-sided safety outcomes: Besides the risk of failure, they also increase the safety of passengers. Keeping self-driving cars off the road for to long because of worries about accidents may be harmful
— Self-driving cars allow instant grounding of defective models; defects can not harm large groups of customers
In the public and regulatory discourse we need to do justice to the unique risk characteristics of self-driving cars!
After the race for fully self-driving cars heated up in 2016, 2017 became a year with exciting developments – many billions of dollars changed hands for self-driving car related acquisitions(1) and many collaborations were started(2). But besides progress, 2017 also showed some limits: Tesla was plagued by defections from their SDC team and had to cancel their fully autonomous coast to coast test drive planned for the end of 2017 and shift the target date for their fully self-driving capability back by 2 years. Volvo effectively cancelled their planned Gothenburg self-driving car trials (by changing the scope to a test of driver assistance technologies). BMW did not change the planned release date of their first self-driving car models (2021) but toned down the expectations about the extent of self-driving capability available in the first models.
Nevertheless an enormously important milestone for the adoption of self-driving cars has been reached in 2017: Waymo is now operating self-driving cars without test driver on public roads in Phoenix, Arizona. Five years ago we had expected this milestone to be reached around 2018. This unequivocally demonstrates to the world that self-driving cars are viable and that they can no longer be considered a technology that is half a decade or more away.
This milestone (and the multitude of achievements of the many actors involved up to the end of 2017) also change the dynamics of the global distributed innovation process around autonomous vehicles. It is beginning to shift from the typical chaotic process involving many different actors with little formal organization trying out different paths and approaches to a more mature process. The acquisitions we have seen in 2016 and 2017 are an indicator that the global innovation process is consolidating and getting closer to move from the early stage of an innovation process (called ‘fluid phase’ in innovation theory) to the ‘transitional phase’. This is a major step typically associated with deep structural changes in the innovation process. We may reach a peak in the number of companies competing to develop self-driving car technology in 2018 or 2019 before seeing a market shakeout thereafter.
For the auto-industry, 2018 will be a crucial year because the time is running out for most OEMs to ensure that they can weather the changes caused by self-driving cars and – maybe even more importantly – that they can identify, understand and profit from new opportunities. There can be no doubt that car sales will come under pressure in the early 2020ies as autonomous mobility services (both for local and long-distance travel) grab a significant share of the mobility market, consumers fundamentally change their car-buying behavior and some emerging markets adjust their traffic infrastructure policies to take advantage of self-driving car technology.
OEMS that have not yet committed to a serious self-driving car strategy risk their medium-term competitive position. With every year that passes, it will become more difficult to adjust to the changes coming to the auto industry. It is unlikely that OEMs will be able to offset losses in demand for privately owned cars by building self-driving cars and selling or leasing them to mobility service providers (or operating them themselves). When the industry gradually comes to accept the reality of shrinking demand for automobiles, it will become more and more difficult to adjust because profitability will fall rapidly and with it the ability to change. Several automakers are likely to fall into the Kodak trap: Kodak was the first company to develop a digital camera. It always understood digital cameras but it failed to reinvent its business model in time and then was unable to turn around the already sinking ship which was bleeding from all sides. The European, Korean and Japanese auto makers need to strongly accelerate their self-driving car activities if they want to survive the coming turmoils of the next decade. General Motors seems to be the only OEM which currently is well positioned in this space. It is pity that Daimler, one of the earliest pioneers of self-driving cars, appears to be content to mostly watch from the sidelines.
In 2018, we can expect another change in the maturing innovation process: The focus will start to move away from the core technical issues towards the implications for the automobile as a whole (its interior, exterior and structural design, its supporting and sales infrastructure etc.) and towards the business models associated with self-driving cars. There are many more use cases for self-driving technology than just ferrying people around; many of these use cases have strong services components which OEMs (or their challengers) need to embrace. 2018 may also be the year where players beyond the auto industry start to seriously consider the implications, opportunities and risks. Retail will be deeply affected by dramatically falling local distribution costs. In the next decade nany supermarkets will have to close their doors as products can be delivered conveniently (and with very customer-flexible timing) to the doorstep. Hospitals, care and emergency services will need to adjust to fewer traffic related injuries. Most industries will need to consider the implications and opportunities associated with significantly lower transportation costs (affecting both inbound and outbound logistics and possibly providing new product or service opportunities). Cities, countries, architects, construction firms need to start planning for a future where mobility is provisioned differently and where space and capacity requirements for transportation are changing. Railways and transportation companies need to consider the challenges which will be raised by autonomous mobility services providers. Self-driving cars and machines will also have major impact on construction and agriculture industries and provide new opportunities there.
2018 may also be the year where the opposition to self-driving cars finds their voice. While self-driving cars have enormous benefits they will eliminate many jobs (not just professional drivers but also in the auto industry and many other industries). Society needs to find ways to cope with the fundamental changes that result from software-based devices with capabilities which some call ‘artificial’ intelligence and we all need to consider in depth how the fabric of society will be impacted and what changes on the different subsystems of society will be necessary. This process should not be underestimated and requires a major, multi-disciplinary effort.
In 2018, every business, organization, political actor, and any forward-thinking individual should take the time to look beyond the technicalities of self-driving cars and carefully consider their implications, opportunities and risks!
Self-driving mobility services are likely to be adopted quickly in high density urban areas. In these regions, car ownership is likely to fall significantly. Several studies have shown that one autonomous taxi might provide sufficient transport capacity to service the mobility needs which are currently fulfilled with 6 to 10 privately owned vehicles. These studies have considered local motorized mobility in large cities such as Ann Arbor, Lisbon, Austin and others.
But how will autonomous fleets impact mobility and car ownerhip in less densely populated areas? About 86% of the US population live in metropolitan statistical areas (i.e. areas that have a relatively high population density at its core). These are not limited to the great cities and agglomerations on the west and east coast but include much smaller areas such as the Grand Forks metropolitan area which comprises 2 adjacent counties in North Dakota and Minnesota with about 100,000 inhabitants (in 2014) and a population density of 11 people per km square. Of course, self-driving mobility services will be very viable in the urban core of this metro area where about 60,000 people live. The remaining 40,000 people living in rural parts of this area have significant, predictable mobility demands for trips towards and back from the urban core. Thus there is a potential for self-driving mobility services even in the outer, less densely populated parts of metropolitan statistical areas. A further 8.6% of the US population live in in micropolitan statistical areas (i.e. areas which are centered around an urban cluster with at least 10,000 but less than 50,000 people). The remaining 6% of the US population live neither in a metropolitan nor a micropolitan statistical area (see the white area in the map of metropolitan and micropolitan areas in US). It is instructive to consider their situation.
Let’s take Sidney, Montana as an example (Google maps): This is a small town with just about 5,000 inhabitants in eastern Montana. It is far away from more populated centers. The nearest larger city is Williston, ND with about 20.000 inhabitants at a distance of 70km. The next city with more than 100.000 inhabitants is Billings, MT at a distance of about 430km. There seems to be a significant mobility demand for trips to Billings: more than four flights leave for Billings every day (airfare about 40 USD). Uber is already active in this town and popular destinations/pick up spots include the airport, high school, health center and Holiday Inn Express.
The US currently has a stock of about 240 million light duty cars, which translates to about 750 cars for a thousand people. Because this ratio is higher in areas with lower population density, there should be significantly more than 5*750=3750 light duty cars in Sidney. Because a large share of the daily trips are local, and because their average speed is high compared to the speed in congested cities, autonomous fleets should be able to provide high-yield mobility services with a relatively small fleet. With a replacement rate of 1 to 7, about 535 self-driving vehicles could replace the town’s entire vehicle stock. The local mobility demands of 5000 people are also large enough that a mobility services provider can start with the smallest economically viable fleet size of probably somewhere between 10 and 20 cars and then grow the fleet as demand picks up. The low regional population density has an interesting consequence for non-local trips: The number of typical destinations is small; the number of routes people can travel from/to Sidney is quite limited. Therefore the potential for on-demand shuttles is high; Williston, with it’s Walmart (about a 1 hour drive) is an obvious target. Such shuttles have another side effect: they can provide the same mobility service to all location which they pass on their route. Such shuttles therefore effectively can bring access to self-driving mobility services to some very rural dwellings.
Today, households in low density areas of the US have much higher car ownership rates than the rest of the population: there simply are no viable alternatives. Self-driving cars fundamentally change this situation. Wherever there is a minimum of demand for personal mobility, self-driving mobility services become economically viable. The number of persons needed to sustain a self-driving taxi resource is rather small; towns with just a few thousand of inhabitants should always provide enough demand to allow a small fleet of self-driving taxis to operate. Initially it may only be the seniors who use these services but then households will start to think about the number of cars they really need and gradually demand for these services (and with it, supply) will increase.
In many lower density areas of the United States, car ownership is a prerequisite for finding work and – as a consequence – people without cars suffer and economic opportunities are lost. For seniors access to medical services and just getting can be extremely difficult. The young face similar problems. These examples show that we can expect sufficient demand for self-driving mobility services in most parts of the United States – including many small towns and even in many areas that have low population densities. The impact of fleets of self-driving cars will not at all be limited to big cities!
How will autonomous car technology generate profits? Among the many different business models – from self-driving mobility services to models centered on data, advertising or entertainment – platform-oriented business models are currently receiving much attention, not the least because Waymo seems to be leaning towards them.
The term “platform” can be understood in different ways: In the automotive context it is usually understood as a car platform where many different models share the same technology under the hood which reduces development costs and allows economies of scale. In a more general, wider interpretation platform business models aim to build a unique competitive position through a complex technology or service which is combined with an ecosystem of users and partners. Ideally the platform exhibits network effects: the larger the ecosystem, the more attractive it becomes to its users and partners and the harder it becomes for competitors to challenge the position.
Waymo’s integrated hard- and software platform
When Waymo’s CEO John Krafcik talks about Waymo’s strategy he emphasizes the integrated hard- and software platform which Waymo is building. Currently this platform is embodied in the ugly white box on top of Waymo’s self-driving Chrysler Pacificas which are occasionally driving around Phoenix. Most of the self-driving hard- and software in the box has been engineered by Waymo/Google: Not just the software, also a novel 360 degree spinning Lidar (with better performance than the Velodyne Lidar, costs reduced by almost an order of magnitude); radar sensors (with better short range detection of stationary objects); the computing platform (developed from scratch in collaboration with Intel); cameras, microphones. Ideally, this box, Waymo’s “better driver”, could be integrated easily into other car models. However, this will always require more work than just adding the box because some sensors will still need to be mounted on the car; more importantly, the car must be ready for self-driving (e.g. redundant safety components) and must be able to communicate with the box by reporting its physical conditions to the box and accepting driving instructions from it.
Can there be much doubt that such a universal driving module would be a highly profitable product? There are many application scenarios (vehicles for commercial use: taxis, buses, trucks, logistics) where self-driving modules would be economically viable for the customer even if priced at very high margins. Startups and established companies should see much opportunity for quickly bringing self-driving vehicles of many kinds onto the market. The technology provider could realize economies of scale while still keeping the total cost for the customer significantly below the alternatives (i.e. where self-driving technology is self-developed or sourced from a variety of vendors).
Platform economics in the consumer car space
Unfortunately, this calculation does not apply to the consumer car space: Consumers are not willing to pay a significant premium for self-driving car technology because they value their own time differently than commercial users of self-driving car technology. In addition, the equation changes for auto makers selling large volumes of vehicles: with a century of experience in managing and cutting costs auto makers will look for every way they can find to slash the price of the self-driving car technology and bring margins down. The larger the sales volume, the higher is the incentive to find other, more cost-effective solutions. Even if they initially agree to source the universal self-driving hard- and software modules, they will work hard to reduce their dependency on it. And they will find many ways to scale back the size of the external self-driving car module: they will want sensors to be integrated into the car – rather than to come with the self-driving platform – and they will want to source them independently. They will clamor to structure and compartmentalize the interface between the self-driving module and their vehicles and they will fight to standardize and take over some of those functions, so that they get control over them. There will be fights over access to the data, over controlling the interface with the user. And it will be hard for the universal self-driving module provider to beat all of those demands back because the OEMs have experience and market knowledge and their car models have special use cases in various segments that the self-driving module provider is not familiar with, does not own and therefore can not easily implement independently. If the provider of the SDC technology platform can not impose lasting, full control over the whole extent of the self-driving platform (prohibiting partial sourcing of components, keeping all modifications to the platform under their own control (even those developed in the context of a particular customer relationship) etc., avoiding any replacement of functionality by the OEM) his power position and margins are likely to deteriorate significantly over time. In the other extreme, the OEM risks losing their established central position in the market to a newcomer who now controls the ‘heart’ of the vehicles. The middle ground is a slippery slope characterized by an uneasy, highly unstable and competitive relationship between both partners where each continually tries to boost their power position to the detriment of the other.
Thus Waymo’s apparent lack of success at finding partners in the auto industry does not come as a big surprise. Why should companies that are used to investing billions for designing a new car model succumb to a company that has invested not much more than a billion dollars (approximately 1.1 bio $ between 2009 and 2015) into self-driving car technology? Shouldn’t they just follow the same path, jump-start their own efforts and ensure that they reduce the gap?
Self-driving software can’t establish a lasting competitive advantage
For anyone who examines the technology and its potential there can be little doubt that many actors will eventually master self-driving car technology. There are many commercial players who have every incentive and sufficient resources to solve the problem. This includes General Motors which has spent 581 million dollars to acquire Cruise Automation and is making a concerted effort to reach manufacturing readiness on the first self-driving car model. There are big European OEMs which are determined to solve the self-driving equation but there are also countries which regard the technology as vital to their economic and military interests. There are investors who understand the economic potential of the technology. Furthermore, although the self-driving car problem is exceptionally hard, it has a ceiling; it will not keep increasing and becoming more and more difficult. Over time, algorithms, simulation environments, tools test data collection and test case generation, hard- and software will become more refined and more easily available. Thus it is very unlikely that a provider of self-driving car technology will be able to establish a lasting advantage over the competition just on the basis of the technology. On the contrary: the time will come where the technology will be mastered by many and be commoditized. The time will come where self-driving car technology will be seen as a natural part of every vehicle, where cars will no longer be differentiated on the basis of their self-driving car technology and where customers will no longer care very much what kind of self-driving car technology is inside. Because safety requirements will be very stringent, vendors of self-driving car technology will have a hard time making the case that their technology is significantly better than the competing products.
Platform models with network effects?
But couldn’t there be a way for the first market entrant to establish a platform position in the wider sense where the technical self-driving car solution forms the base for a self-sustaining ecosystem of customers and partners which exerts a pull on the market and erects a powerful barrier against entry for competitors?
There are several strategies which could be applied toward this end: those who enter the market first and expand quickly can realize economies of scale, which keeps costs down and can discourage competitors by keeping prices low. But keeping prices down means foregoing much of the rents associated with significant productivity increases due to reduced costs of mobility. It is more than questionable whether this would discourage competitors or whether it would be interpreted as a play towards dominance in a lucrative market – an economic signal that might actually entice competitors to redouble their efforts.
Another approach would be to use current dominance in the technology to establish a hard-to-assail business position, a self-growing platform, around the technology. Self-driving car technology requires much more than the car’s hard- and software. There are many legal aspects which require substantial effort. Various service infrastructures need to be established – some to fulfill legal requirements, others out of practical necessity – and might become key parts of the platform ecosystem: California self-driving car regulations already mandate that operators of self-driving cars ensure that high-definition maps are kept up to date and are regularly distributed to the cars. The same regulations describe a remote operations service which assists fully self-driving cars in challenging situations (i.e. a 24/7 remote operations center). Infrastructures are needed for cleaning and maintenance, accident handling, secure over-the-air updates of self-driving car software. The scope of platform services could be extended further to include services for managing fleets of self-driving taxis, trucks and buses as well as associated customer facing services (reservation, payment processing etc.).
Companies which provide the full breadth of such services (or manage access to it) certainly have a favorable competitive position, but it is questionable to what degree this can protect the platform and establish a barrier against entry of competitors. Precursors to most of the platform services described above already exist today and companies exist already that would be willing to extend their services to the self-driving car market. Today many OEMs already operate remote assistance centers (GM OnStar, LexusLink, BWM Assist etc.) which could easily be extended to provide assistance to fully-self driving cars. Several companies are focused on building and maintaining high definition maps (among others Here which was purchased by the German OEMs). Rental car and mobility services companies already have experience with some of the additional services needed and would certainly aim extend their business models to the self-driving car space. Thus it is unlikely that such a Waymo self-driving platform could not be replicated with a determined effort by some of the OEMs or other players.
SDC platforms not similar to operating system or marketplace platforms
The market for self-driving car technology is not similar to other markets where we have seen platform models succeed. This is not like some of the operating system (Windows, Android) which have grown into a platform, where this platform is the base for millions of different applications and uses, where the platform grows because with more users the breadth of applications and uses increase. In contrast, self-driving mobility is a much more specific – and for safety and security reasons – limited application domain where scale effects matter but the diversity and number of applications will be comparatively low. A software platform for self-driving cars can never be as open as Windows or Android. A self-driving software platform will most likely evolve in a way that the platform has a very limited external application programming interface which partners may latch onto. But this also means that competitors which provide their own universal self-driving car modules or platforms should find ways to expose similar interfaces to their partners and these partners could more easily support multiple self-driving car platforms with their services and applications. Thus a self-driving hard- and software is not likely to achieve an operating-system like lock-in effect for its partners and customers.
The market also does not resemble an Airbnb, Ebay or Uber, domain-specific optimized marketplaces which link a large number of product or service providers to a large number of customers and which increase in value and attractiveness with an increasing number of participants, thus quickly erecting barriers to competition. Yes, self-driving car technology can be the basis for establishing mobility services which will tend to rapidly establish a dominant, hard-to-assail position in a region. This mobility-as-a-service business model does have a lock-in effect but this is a very different type of business model than the self-driving hard- and software platform model which we are currently examining.
Thus, the pioneers of self-driving hard- and software can base their business models on viable platform strategies centered around a universal self-driving hard- and software model complemented with associated services and business relationships. Given the economic value that can be realized in many markets and business scenarios with self-driving vehicle technology the business model will initially be very profitable. As in many other markets the pioneers have the potential of establishing a leading and hopefully lasting market position. But their competitive advantage will fall over time as the market becomes commoditized and it will be hard to keep competitors out – unlike the platform models in other markets which enjoy considerable network effects.
The problems with Waymo’s focus on a platform business model
Thus Waymo’s apparent focus on a universal self-driving platform-based business model seems to be questionable. When Waymo decided to shelve the activities related to their self-driving firefly electric two-seaters, they seem to have made a decision against squarely focusing on the mobility services model, the one business model in the self-driving car space that exhibits strong network effects and which would provide a permanent advantage for the first mover.
A side problem of Wamo’s universal self-driving platform is that it does not seem to be well executed. To make their platform truly universal, they would need to expose themselves to many different use cases and ensure that the platform works for cars, trucks, buses, even self-driving machines of different types. Many startups are currently working on products and services in the self-driving space and would be keen to cooperate with a provider of a self-driving car modules but there is no evidence, that Waymo is branching out to them. Companies such as EasyMile, Navya, LocalMotors, truck manufacturers, and many others would be more than willing to jump on the bandwagon and thus ensure that the platform really becomes universal. Waymo would profit from learning about differing requirements in different application scenarios which would necessarily lead to a more customizable structure of the self-driving “box” which Waymo envisions placing on top of a vehicle. That the top box may not be the best idea can easily be seen when we consider the context of trucks where a top box is much less compelling because it would not achieve full 360 degree unobstructed sensor vision. Another worry about Waymo’s approach to a universal driving platform is the reliance on their own sensors. With the current innovation in the automotive sensor market it is not very likely that their sensor suite can remain ahead of the competition for long. A universal self-driving car platform needs the ability to rapidly incorporate new sensors and even new sensor types. Impressive as Waymo’s self-developed sensors may be, there is also the risk of paying less attention to external innovations.
For the market as a whole, Waymo’s detour focusing on a business model based on some incarnation of a universal self-driving hard- and software platform (“the better driver”) may be a positive development. It reduces the risk that one player will dominate the field, has given auto makers time to understand the nature of the challenges better and increase their determination to close the gap. Most auto makers have now understood the dimension of the challenge (although some have difficulties balancing their priorities between autonomous driving and electric vehicles). General Motors is an excellent example of an auto-maker getting up to speed: their acquisition of Cruise Automation is a win-win for both companies and both companies together are not plagued by the competitive stalemate that a collaboration between a universal self-driving module provider and established auto makers would engender. Being the most advanced player, Waymo is likely to profit greatly from its self-driving car technology but a problematic platform-focused commercialization strategy may be giving its competitors some welcome breathing space for catching up.
Fleets of self-driving cars will reduce the cost of individual motorized mobility and increase its accessibility to people without driver’s license. Many city planners fear that this will induce additional demand and significantly increase miles traveled with the result of even more congestion in our already heavily congested cities.
Fortunately, there are many reasons why an increase in person-miles traveled with self-driving cars will not lead to an increase in congestion. The opposite may be true: we may find that self-driving cars, while certainly increasing person-miles traveled will actually reduce the congestion in our cities. Congestion is not a direct function of the number of vehicles on a road; it depends on driver actions, routes taken, road utilization per vehicle and systems for flow optimization (traffic management systems etc.). If we increase the number of miles driven and keep all other parameters constant, then congestion will certainly increase. But with fleets of self-driving cars, all of these parameters will change, some significantly.
In the following we will first look the reasons why self-driving cars are likely to reduce congestion compared to human-driven cars. Items 1 and 2 show that there is significant potential for congestion-reduction (which in turn means that the risk of induced mobility leading to more congestion is reduced).
1. Driving behavior: The driving behavior of a self-driving car differs from the driving behavior of human drivers. Autonomous cars don’t exhibit the lane-hopping and other congestion-creating behavior. Simulations have found that even a small percentage of self-driving cars among many human-driven cars on a lane reduces congestion because the self-driving vehicles help to smoothen the traffic flow. Self-driving vehicles also reduce the typical delay of the average human driver at a stop light turning green and thus ensure that more vehicles can pass that stop light in a given time frame. A self-driving vehicle will not sit idle for a second after the car in front has started moving. This number can be further increased if the self-driving car uses an optimized acceleration pattern at a stop light. Thus, with an increasing ratio of self-driving cars, the throughput will increase at the bottlenecks which will lead to significant reduction of congestion.
2) Road capacity utilization: 2a) Road space: Self-driving fleet cars used for urban driving will be smaller and thus use less road capacity. Self-driving cars will also systematically adhere to an optimal minimum distance to the car in front which significantly increases the number of vehicles that a given road segment can support during heavy traffic. 2b) Parking space: Fleets of self-driving cars will be in operation most of the time, especially when mobility demands (and with it traffic) is high. Thus cities will need much less parking space and can use parking space of other purposes. In some cases, parking spaces could be turned into additional lanes, further increasing throughput. This is an option but we expect most of the parking spaces that are freed up to be put to other use. Note that self-driving car fleets may need very little dedicated parking space because they could simply use existing lanes that are no longer needed during off-peak times or at night for parking. 2c) Convoy driving: As the ratio of self-driving cars in traffic increases, these cars will more frequently find another self-driving car in front or behind and can then coordinate their driving behavior. This can lead to further reduction of distances between the cars and can further improve reaction times at stop lights. 2d) Lane sharing: Self-driving cars can drive consistently with more lateral precision than human drivers. Thus they can operate on narrower lanes. This also makes it possible that more self-driving cars can drive next each other than the number of lanes available. For example, three self-driving cars may ride next to each other on a two-lane highway. This could be another variant of convoy driving and would need communication between the vehicles. 2e) Micro-cars: Very small self-driving pods could be built so that two of them fit next to each other on a single lane. An example has been proposed by Harald Buschbacher (although these two wheelers with auto-retractable stabilizer wheels are envisioned as personal rapid transit vehicles using their own very narrow lanes).
The previous 2 items (Driving behavior and road capacity utilization) ensure that the congestion-inducing effect of a self-driving car is much lower than the average human-driven car which in turn allows to significantly increase the number of person-miles traveled without increasing congestion. But the next item is the key reasons why we can be confident that self-driving car fleets will not increase congestion, even if they significantly increase the number of person-miles:
3) Internalizing the costs of congestion paves the way for combating congestion:
Today, congestion on our roads leads to enormous economic costs. Unfortunately, these costs are distributed among the many traffic participants which at the same time are cause and victims of congestion. It is difficult to unleash market forces to find ways for reducing congestion because it is difficult to set prices for congestion-free roads nor can we correctly attribute congestion-costs to those who cause it and make them pay. This changes once shared fleets of self-driving cars provide a significant share of local mobility because these fleets internalize a sufficiently large part of congestion costs.
Fleet managers will focus on the bottom line and they have every incentive to maximize their return on capital. They will try to minimize the size of their fleet and to maximize the throughput of their cars. To them, congestion translates directly to cost. When they send a car through a congested area, this increases the cost of the car, reduces revenue opportunities and it also reduces the throughput for other cars of the fleet that may need to take the same route a little later. After a few months of operations, fleet controllers will be able to quantify exactly how much their bottom line would improve if the throughput in a certain bottleneck could be improved by a few percent. They would find that many investments in infrastructure, signalling algorithms, routing methods etc. would have a positive return because their costs (of congestion-reducing activities) are lower than their benefits (increased fleet revenues, lower fleet size (capital stock)).
From an economic perspective, shared fleets of self-driving cars aggregate the mobility demands and the congestion-related effects of their large group of customers. This aggregation allows the fleet to find much better ways of handling congestion – taking into account both the preferences of their customers with respect to congestion-related costs, the congestion-inducing effects of different routes and mobility solutions and internal or external potentially costly mechanisms that reduce congestion. The fleet will very clearly understand (and be able to quantify) its effect and the effect of each of their customer’s trips on congestion. In contrast to the individual driver on the way to the office very morning, who is oblivious to his share in making congestion and who simply wants to take the fastest route, the fleet will not be concerned with the speed of the individual trip but will make sure that the trips are routed in such a way that the throughput of all their vehicles will be maximized. The goals of the fleet with respect to congestion are very much aligned with the goal of the city as a whole: that throughput is maximized.
This argument may sound academic. But the effects will be very real. Fleets that are small will not have a large impact on cities. But once fleets process a significant share of local mobility, they will have the best knowledge about traffic and congestion patterns in the city. Their cars will provide them with detailed up-to-the minute traffic information for all parts of the city. Economic rationale will lead them to build complex models of traffic flow and look for ways in which throughput can be improved and they will be able to very clearly indicate what approaches in which areas of the city could lead to which level of congestion reduction. They will work with city official to optimize their signaling infrastructure, they will even be willing to invest into that infrastructure (if the cost is lower than the benefits from congestion reduction). The fleets will also look for ways to shift mobility demand (so that some people defer their trips to non-peak times) and to reduce congestion cost per trip by combining trips (through ride-sharing or by inventing new variants of ride-sharing that actually appeal to their customers).
In summary, there is no reason for city managers to worry about congestion-inducing effects of shared fleets of self-driving cars. These fleets will have large benefits for the city. They will actively combat and reduce congestion because they are the first entity that internalizes the costs of congestion. They will reduce the ecological footprint of mobility because they will be mostly electric vehicles and the average vehicle will be smaller and lighter than the vehicles today. They will accelerate the transition to electric vehicles because the shared utilization of short-range vehicles is the optimal use case for electric vehicles. They will free up parking spaces and eliminate traffic looking for parking (which can be a very significant share in inner cities).
If you are still worried about the congestion-inducing effects of self-driving car fleets, here is a simple, political argument: Self-driving car fleets won’t increase congestion in our cities because we will not let that happen. Such fleets will not populate our cities over night. They will initially service a small fraction of the population and can not immediately cause significant increases in congestion. As these fleets become larger, politicians will certainly not sit idle if congestion increased and neither would the electorate accept more and clearly attributable congestion. This in turn would increase the economic pressure on such fleets to find ways for reducing congestion (the most straightforward would be to limit their size by adding congestion charges to their pricing structure).
Note: This is part of a larger series of misconceptions related to self-driving cars. The other misconceptions are discussed here. A PDF document with all misconceptions is also available for download.
In June, German Chancellor Angela Merkel provided the following forecast as part of her answer to a question about what the world would look like 20 years from with the the statement that “In 20 years, we will need a special permit if we want to drive a car manually.”…”We are the biggest risk.” (Source: Die Welt, Auto Motor Sport, 2017-06-09). Also in early June, Germany’s minister for transport and infrastructure, Alexander Dobrindt, said that he wants Germany to have the world’s most modern public transport by 2025, a vision which builds on self-driving electric buses (Source: Bayernkurier, 2017-06-09).
In a country, where many jobs directly or indirectly depend on the auto industry, these statements may be an indication that Germany is taking a future with autonomous vehicles more seriously and begins to give some thoughts to the near- and long term implications of self-driving car technology. There are few indications, however, that the transport ministry has begun to consider the effects of self-driving cars, trucks and buses on the national road infrastructure (the recently released national traffic infrastructure plan 2030 does not take self-driving vehicles into account). Plans to reduce the current and projected shortage of truck overnight parking spaces along German highways don’t take into account that demand for such spaces may peak in the early twenties. As with many other countries (including the EU, that is currently considering a costly infrastructure requirement for EV charging stations that completely disregards the likely changes for parking, EV charging and mobility caused by autonomous vehicles) , the enabling relationship between self-driving car technology and the adoption of electric vehicles is still not recognized. Neither are the impact on rail-based transport, which will likely see a decline on many routes and which will also loose much of its environmental advantage (so popular with many environmental thinkers and infrastructure planners) with the rise of self-driving scheduled and on-demand electric buses.
Today billions of Euros are mis-allocated in city-planning, construction, traffic infrastructure development because planners assume that mobility and transport patterns of the future will be similar today. It is time for German, European and the world’s leaders to seriously consider the changes that result from self-driving vehicles. Hopefully the recent statements from the Germany cabinet are an indication that politicians are beginning to slowly move into this direction…
The auto industry increasingly recognizes the threats and opportunities associated with self-driving cars. Unfortunately several impediments stand in the way of formulating and implementing a strategy for dealing with self-driving car technology and its impacts:
1) Time: lack of urgency
Although the competition in autonomous car technology has heated up considerably over the last 2 years, most industry experts continue to expect a slow adoption curve which could easily span two to three decades. Unfortunately, adoption of self-driving car technology (level 4 and up) will be much faster than traditional adoption rates of new technologies in the auto industry. A key accelerator is the enormous net benefit of the technology not just in terms of safety but also as increase of available personal time, competitive position (for companies and countries) and a significant decrease of costs (labour, fuel, insurance, capital). As a consequence there is much less time to formulate a sound strategy for self-driving cars.
2) Shared auto industry perspective clouds impact analysis
Shared convictions and experiences make it much more difficult for the industry (including their consultants) to think through fundamental, deep, disruptive changes in the architecture of mobility. Whether it is the joy of driving, the importance of brand for the consumer, the assessment of the legislative and regulatory environment, the consumer’s propensity to use shared self-driving mobility services or the likely business models, industry insiders tend to reinforce a perspective on the impact of self-driving cars that remains much too close to the current model, experiences and structure.
3) Lack of understanding for self-driving car business models
For many years, the auto industry has recognized a trend towards shared mobility services. Automakers understand that self-driving fleets will accelerate this trend. But they seem to spend very little effort to think through the dynamics of this market (which differs fundamentally from the traditional car-sharing and mobility-brokering markets), the way that shared mobility services will operate and compete, the regulatory environment that will emerge around fleet oligopolies, the differences between urban and long distance shared self-driving mobility services or the cost structure, maintenance strategy and model mix for such services.
In addition, there are many other business models besides shared fleets which may provide opportunities related to self-driving car technology which established players need to carefully consider, evaluate and prioritize.
4) Relationship between electric vehicles and self-driving cars not understood
In parallel to the self-driving car phenomenon the auto industry is involved in the switch towards alternative propulsion modes. But the relationship between self-driving car technology and alternative fuels is widely overlooked: Because self-driving cars will change mobility patterns (increase of urban mobility services, changes in long-distance travel patterns) and self-driving fleet vehicles will be able to refuel autonomously (or nearly-autonomously), the context for the adoption of alternative fuels changes dramatically. Battery range will become much less important; rather than optimizing cars for maximum range they will be optimized for an optimal range with respect to the mobility pattern which they are used for. When fleets carry a larger share of traffic the dimensioning of an adequate charging infrastructure becomes much easier and much more economically viable. Thus autonomous vehicle technology will serve as an accelerator for the introduction of electric and alternative fuel vehicles.
5) Fear of cannibalization / resistance to change
Any organization that faces major change and must consider the effects of a disruption of its primary business model will encounter tremendous internal resistance. Those who see the writing on the wall will hesitate to become advocates of (painful) change because internal opposition is fierce, uncertainty abounds and – as a result – career risks are high. It is useful to seriously study other industries and companies which had to face disruptive change. One of many examples is Kodak, a company that had developed the first digital camera already in the Seventies and brought the first digital camera to the market in 1995. There may be some parallels to the auto industry, which has a multi-decade history of developing technologies for self-driving cars. But Kodak hesitated far too long to adapt and rethink its business models, fearing cannibalization of their very profitable film camera business. When their profits began dwindling, it was too late. The auto industry cannot afford to make the same mistake.
For more on this topic please join us at the upcoming 1-day seminars on self-driving cars in Frankfurt (March 23) and Auburn Hills (May 16). The seminar will be run by Dr. Hars and will help to develop a better understanding and analysis of implications of self-driving cars. More info…
Autonomous cars will change the way we think about traffic. Today traffic is primarily regarded as the result of the independent actions of thousands of drivers. A view from above on any city would show large numbers of vehicles pursuing their own trajectories through the maze of roads. The cities’ traffic management systems try their best to observe, identify and somewhat channel the grand flows.
At first glance, autonomous vehicles do not seem to change this situation very much. From above, self-driving cars will not be distinguishable from human driven cars and they too, will seek their individual paths through the maze of roads. The picture changes, however, when we consider fleets of self-driving cars. Recent statements by Ford, Uber, BMW and others clearly show that fleets of self-driving cars will emerge early and have the potential to capture a significant share of individual motorized mobility.
This introduces a crucial difference: Fleet vehicles no longer pursue their local optimum; rather than completing the individual trip as quickly as possible, fleet management will seek to maximize throughput for all of its vehicles – for the fleet as a whole. The operational goals of fleet management are therefore very much aligned with the traffic flow goals of a city as a whole.
Initially, autonomous fleet vehicles will be instruments which fleet management systems can use to understand, model and predict the detailed traffic situation. The vehicles will be used as sensors and relay important information to the fleet management system.
As fleets grow, fleet managers will find that the vehicles can be used to influence the flow of traffic. Many different strategies are possible (and their effectiveness varies greatly with the ratio of fleet cars to total number cars): fleet vehicles can purposefully slow down the build-up of traffic ahead of arteries which are in danger of clogging. Fleet cars can reliably calculate and selectively or pre-emptively use alternative routes. As the percentage of fleet vehicles in relation to total traffic grows, fleet vehicles may travel part of the way in more densely packed convoys. They may even change their acceleration behavior at stop-lights (using a somewhat faster acceleration pattern than the standard acceleration pattern of human drivers) which may or may not be copied by human drivers.
Because both city traffic managers and fleet managers will recognize early on that their interests are very much aligned, we can expect many ways in which both parties will cooperate. Fleet managers will make real-time traffic information gained via their cars used as sensors available to the city traffic managers. Fleets are likely to ask city traffic managers to adjust stop light phases to improve traffic flow (and fleets will provide the data and models to prove that these changes will be beneficial). We can expect that this will lead to much more real-time traffic management for stop lights and fleet vehicles may come to very directly influence traffic signals. Eventually, as the differences in driving behavior between human-driven and autonomous vehicles become more apparent and fleet vehicles exceed 20 percent of traffic (initially mostly likely in urban centers), we may find that cities will reserve some lanes or roads for self-driving vehicles because they are more effective at providing local mobility than individual cars, or because the throughput on autonomous-vehicle-only lanes can be twice the throughput of human-driven lanes (mostly due to shorter distances between vehicles and better reaction times/acceleration behavior at stop lights, in some cases also because two conventional lanes might be re-fitted into three narrower lanes for autonomous fleet vehicles).
But this is only the tip of the iceberg. Fleet managers will understand local traffic very well and want to avoid their most valuable resources to be stuck in traffic. They will be able to predict the actual duration for a trip at any given point in time and will aim to minimize trips which incur heavy congestion. Instead of just driving a customer every day to work at a time of his choosing, they will look for ways to reduce the peak load on the fleet. Ridesharing is only one of many approaches: Fleets will provide rewards to those who stay out of the rush hour (or add congestion pricing, which in turn will drive down congestion). They may find ways to systematically phase traffic flows in certain areas, work with employers and schools to adjust working hours, provide an in-car environment that allows workers to begin their work while commuting (and ensure employers’ approval), provide a reliable forecast of trip times (and a clear indication how expected trip times can be reduced by leaving earlier or later).
Time will tell which of these many possible actions will yield the most benefit (and through which other approaches fleets of self-driving vehicles will improve the overall traffic flow in a city). But it is obvious that fleets of autonomous vehicles will lead to a very different thinking about traffic. Where today we have thousands of actors all pursuing their own little traffic goals, these fleets will start us thinking about how traffic can be optimized not just locally but as a whole. It is clear that this optimization does not necessarily start when a trip begins, but potentially already before – when a mobility demand for a trip from a certain location to another location in a certain time range is known. Fleets will pave the way by optimizing their trips against the whole fleet. And the lessons we learn from managing trips for autonomous vehicle fleets will deeply change our thinking about traffic and how traffic should be organized.
Thus, autonomous vehicles not only drive themselves; they change the cost structure of mobility, which in turn enables shared autonomous mobility services to grab a significant part of the market for motorized individual mobility. These shared services will necessarily implement a centralized perspective on mobility which requires finding (and negotiating) ways to optimize the mobility demands of large groups, even cities. In the end, we will likely think about all mobility – whether in a fleet vehicle, in privately owned autonomous or conventional car – from a perspective of global optimization. It won’t be long before our mayors, regulators and politicians will see the potential of self-driving vehicles for traffic management and begin to develop policies that lead traffic away from today’s heavily congested local optima towards structures that come much closer to the global optimum.