Once upon time there were horses and carriages. Carriages would travel for days between cities, reaching coach houses where to get fresh horses, some food and even some sleep. Carriages would travel on public paths, bridleways, in the dust, the sand, the stones … Paved paths followed historical routes, maintained here and there. As businesses and trade grew or vanished, as cities came up or down, new routes appeared organically, and old ones slowly faded away.
Originally written in 2016
Giving up our decision making for safer and smarter roads
And then came the automobile. Sharing the road with pedestrians, horses, and carriages, automobiles would blast through, in their pneumatic tires and combustion engines. Some routes were better than others, better paved, better maintained, harder soil. As more and more cars populated the paths over the years, the decades, maintaining the paths became a problem. Governments were faced with the decision of which paths to keep up, to upgrade and make more suitable for cars, which to transform into roads. Slowly a road network appeared, centrally managed, and cars were expected to travel on roads, not on public paths or bridleways. The freedom to travel wherever we wanted was being taken away in order to optimize public budgets, to build and maintain road infrastructure.
The number of cars continued to increase. The occasional car here and there, first a privilege of the very rich, became more and more adopted slowly trickling down the socioeconomic strata. Cars were not way more abundant, likely to meet each other in the road more frequently, at dangerous curves, at intersections or crossings. Road casualties escalated, and the need to regulate road traffic became obvious. Driving legislation restricting what could and could not be done on the road appeared, and with it the appearance of road signs, signaled intersections, and in general rules taking away the freedom to drive whatever way we had been doing in the past.
First, we gave up our freedom to drive at the macroscopic level, by limiting certain roads to car traffic. Then we gave our freedom to drive at the mesoscopic level, by limiting what we could legally on the road, to reduce collisions.
Global economies have continued to grow, and with them the number of cars on the road. Today there are over 1.2 billion vehicles roaming the roads globally. There are also 1.3 million casualties on the road. If we take the average societal cost per road casualty of 2.0 million USD, we are talking about a societal cost to the economy of 3.6 trillion dollars in road casualties, that all of us pay for, either directly or indirectly through our taxes. Driving costs our society 36 cents per mile –humans drive every year 10 trillion miles–. For the average commuter doing 10,000 miles per year, that’s 3,600 dollars, or 10% of a US household disposable income. In other words, we could be 10% wealthier instantly as a society if we were to eliminate road casualties altogether, in addition to eliminating the human drama and suffering.
All transport modalities have evolved over time to be more and more publicly regulated. We have reached a point where when we travel by plane, we expect regulation to ensure complete safety. Whenever there is air crash, long investigations across the whole industry are performed to ensure the root cause is eliminated and this type of crash does not happen again. We have stopped thinking about safety as an optional feature. We don’t decide to fly Delta or American, or Airbus vs Boeing, because of safety. Safety is no longer a discriminating factor, we derive no longer utility out of safety in our choice of airline. We have transferred liability to the airline, i.e. we are no longer accountable for safety and expect a 3rd party to take care of our safety. A similar set of measures and chances have happened in the railway industry. In both modes, air and rail, the possibility to transfer liability to a 3rd party that will have to ensure our safety has happened because of technological advances that make the monitoring of control of decisions of all agents possible. The number of vehicles in air transportation was smaller than in other modalities, and the appearance of radar and beaconing allowed for air traffic control to appear. The number of trains is still small, at least compared to the number of cars on the road, and the appearance of radio controlled braking systems allowed for automated signaling (although it’s not a completely solved problems). In any case, air and rail traffic management has been centralized, and the microscopic travel decisions have been removed from the individual vehicles.
Unfortunately, in road transportation we are not fully there yet, and the technology that makes it possible to manage road traffic to a precise enough degree so that we can transfer liability to a 3rd party ensuring our safety is not widely available, yet. The good news is that we are finally getting there. The appearance of advanced cellular wireless networks enabling vehicle-to-vehicle communications finally opens the possibility for cars broadcast their position, motion and predicted path, allowing other agents to plan, coordinate, and avoid crashes. But unlike in air and rail traffic management, road traffic management at the microscopic level cannot really be done in a centralized manner, but must be done a distributed, collaborative multi-agent approach. Whereas there are 10,000 planes at any point in time in the air, there are 250 million vehicles on the road. Also, cars travel at much higher densities and proximity, requiring much higher levels of accuracy and speed. The sheer complexity of managing 250 million paths, guaranteeing reliability, accuracy and latency, in millisecond by millisecond decisions, does not allow for a centralized system. Another factor that is slowing down, although not blocking, the appearance of microscopic road traffic management is autonomy. Human drivers are lousy at receiving and following instructions, we are either distracted or think we know better. Computers are much better at receiving all that information from the neighboring vehicles in order to reach consensus for the multi-agent collaborative driving. However, autonomy really is optional. Both trains and planes operate either with auto-pilot or human-pilot, but air traffic control and railway traffic control operate always, regardless of who’s at the steering of the vehicle. Additionally, we need to understand that for many years to come human-driven and automata-drive vehicles will have to coexist on the road.
We are in a path to put in place those microscopic traffic management systems. We will see the first versions of this with personalized Traffic Management Center (TMC) data being sent directly to vehicles, rather than homogeneous data on displays on top or the side of the roads accessible to all drivers. The information each vehicle will get will be different, leading to smart routing. Initially this will be limited to a one-way link from TMC to vehicle to optimize congestion, but eventually, as the density of equipped and vehicles connected to the TMC increases, it will evolve into a 2-way link with other applications, such as safety, platooning, capacity optimization, etc.
Just like we lost our macroscopic and mesoscopic decision making whilst driving, the appearance of technology making microscopic traffic management possible will lead to us giving up our last freedom whilst driving, for the sake of safer and less congested roads.
Obviously, we first need to get there. The key requirement to make this vision is a reality is the availability of such vehicle-to-vehicle communications network. This is both a technological issue, the existence of the technology meeting the safety requirements of road driving, but also a bootstrapping issue. Assuming the technology existed, it is clear that if this network was in place, it would add tremendous value. However, during the initial inception phases of the network build-out, the value that enabled nodes get is limited, or even none. The incentives that consumers and automobile manufacturers have initially to enable such network are limited. The lack of initial market incentives may seem like a reason for the regulator to force adoption. Unfortunately, lack of deployment is sometimes mixed up with lack of compatibility and interoperability across all vehicles, regardless of make or model. Whereas adoption is truly a problem of lack of market incentives, lack of interoperability is a problem of lack of engineering standards. Engineering standards are slow to be developed, but are best to be developed by taking existing implementations, ideally more than one, and achieving industry consensus to define such standard so that all contributing parties benefit from it.
This lack of market incentives has led regulators in US and EU to consider whether to legislate the creation of the network communication technology for V2V. In an unusual scenario, one where the regulator focuses on engineering standards rather than just on benefits, the regulator could force the adoption of such technology. The issue with a regulatory approach is two-fold: first of lack of societal returns, second of not being technically proven to be fit for purpose.
During the first years of operation, the network deployment would have a very significant cost, and at the same time add zero value. Over time, as density increases, the network would eventually become dense enough for V2V-based applications to start to appear. The lack of initial benefits to consumers is the highest risk of following a regulatory approach: it’s unproven whether consumers will adopt the technology. Rejection may not just be for V2V as a concept, but also the security and privacy concerns around the chosen implementation. For example, the proposed V2V communications network implementation over WIFI (802.11p), the Dedicated Short-Range Communications (DSRC) technology, has significant risks from a privacy, trust and security mode, which may deter consumer adoption. In addition to consumer rejection of the technology, the approaching of introducing a technology by ruling creates issues margin erosion for the manufacturer, and unfair subsidies, since only luxury vehicles equipped with computing instruments able to leverage the V2V network will offer V2V applications, hence subsided by cheaper vehicles, etc.
In addition to the societal issues around ruling, there is also the question around whether the technology is fit for purpose. The approach to rule first a network, to see V2V applications appear at a later point is flawed from an applied development stand-point: the fitness of a network communications technology should be first proven in the field, for actual real-world applications. Only then, once proven, and solved in more than one way, it makes sense for it to become a standard, driven by the major engineering standards bodies such as the IEEE. The regulator may at this point require homologation against this standard. But the order matters: first prove actual implementations in the field, then standardize, and only then regulate for homologation purposes. This approach is followed for other automotive technologies, such as Automatic Electronic Braking (AEB), where the automotive original equipment manufacturers took a step forward to adopt AEB, without the need for regulation. Once AEB is deployed, there can be standards defining the minimum performance requirements of AEB.
The intent to build a V2V network is not new. The first attempts where discussed in the late 1990s within ITS circles. Back then, the existence of such network was deemed as the only likely route to solve casualties on the road. Indeed, if all cars were equipped with the technology and communicating their position, motion and future path, we could completely remove collisions and congestion. However, the development of technology was slow, partially because of the lack of clarity around the immediate societal returns, and the significant cost of deployment. The engineering effort continued though, and the original IEEE 802.11a was modified and tested over the years, from 1999 to 2004. In 2004, field tests were performed on the modified specification, what was to become the IEEE 802.11p. Spectrum was allocated to the OEMs, and work continued to finalize the specification, which came out of draft eventually in 2011.
The issue is that in that decade the world of safety in the vehicle moved on. Traditional computer vision was introduced in the vehicle, along radar, to produce the first version of adaptive cruise control and lane departure warnings. Computer vision technology, now also based on deep learning and artificial intelligence techniques, continued to evolve to the point of being adopted in tens of millions of new vehicles sold. The original reasoning for the necessity of a V2V network weakened. Cars could now have Advanced Driver Assistance Systems (ADAS), or later as Autonomous Drive (AD), without V2V. V2V was conceived as the solution, but time has made it become part of the solution. The thing is though that V2V is still a necessary addition to the automotive stack because of reliability reasons. Line of sight for radar, camera and LiDAR might not be there, and vehicles should have more sensors to rely on, e.g. driving in night conditions. And viceversa, V2V cannot be the sole sensor for ADAS, the vehicle should be safe, even without connectivity to the nearby vehicles.
Because of these reasons, societal, technological and evolutional, the regulatory approach to build-out a network is unlikely to succeed.
We believe that we need that start the network now, deliver immediate value from the network for the consumer and focus on creating a virtuous cycle that feeds growth. The way we are approaching making the V2V network useful from the beginning, along the lines that V2V is no longer the solution but part of the solution, is to combine V2V with other sensor modalities, e.g. radar or computer vision. By doing this, the vehicle has, from day one, protection, in the form of ADAS/AD. Either way AD/ADAS will become multi-modality, with multiple sensor modalities operating at the same time for robustness and reliability, especially at night.
As we grow this network of AD/ADAS equipped vehicles, with combined V2V and vision sensor modalities, we are deploying nodes that can, from day one, be leveraged for microscopic road traffic management. We know that this microscopic road traffic management over V2V has significant benefits and that it creates social welfare: it removes collisions; reduces congestion and optimizes road infrastructure utilization; allows digitizing the public space to make our cities safer and smarter. We are then confronted with ethical questions around freedom and privacy. Are the benefits we gain as a society worth losing our freedom to drive where and how we like (and to kill others)? Are the benefits we gain by reducing congestion justified, if that means that some drivers will spend more time on the road as a result? Is our ability to digitize the public scape and any transient change in real-time in order to make cities smarter and safer worth the loss of privacy in the public space?
We think these trade-offs are justified as means to create social equity, and this is an essential part of why Nexar exists.