Recently we have to come to hear the phrase “Connected and Autonomous Vehicles.” To the educated listener, this may sound oxymoronic, because in many ways, it is. Autonomy is about independence, connectedness is about interdependence. History reminds that all of today’s networked technologies began as isolated, autonomous instances. Personal computers (minicomputers, mainframes) predated the Internet (Arpanet, etc.). Books, magazines, and newspapers predated the World Wide Web. Motors and generators predated the electric grid. Short, independent, often differently-gauged railroads predated national and international rail networks. Cisterns and septic systems predated water and wastewater pipelines. Crystals and single-celled organisms predated the web of life. It is the natural order of things.
“Connected and Autonomous Vehicles” is not just a US expression. The UK uses it too. They write:
About connected and autonomous vehicles
Connected and autonomous vehicles incorporate a range of different technologies, facilitating the safe, efficient movement of people and goods.
Increased connectivity allows vehicles to communicate with their surrounding environment. This provides valuable information to the driver about road, traffic and weather conditions.
Vehicles with increasing levels of automation will use information from on-board sensors and systems to understand their global position and local environment. This enables them to operate with little or no human input (be driverless) for some, or all, of the journey.
To be clear, an autonomous vehicle needs to be able to make real-time decisions independent of any explicit communication with other vehicles.
A connected vehicle gets and provides information to infrastructure (Vehicle to Infrastructure or V2I) and other vehicles (V2V).
The example (due to Volvo) that is trotted out for the value of connected vehicles is the ice patch. A vehicle driving a road at too high a speed for conditions (because if it weren’t too fast for conditions, the ice patch wouldn’t matter) detects an ice patch, hopefully doesn’t crash, and relays that information using some V2I protocol back to a traffic management center, which then relays it to all other vehicles traveling that section subsequently. This seems fine enough. It would of course be better for the first car to have detected the ice patch before driving on it at too high a speed, and decelerate to an appropriate speed. But what harm can come from this? I suppose very little. More information, presuming it is accurate, is usually better than less. Yet, I am not very moved by this example because it is rare. Usually roads are either covered in snow and ice or clear of frozen precipitation. In Minnesota, prior to global warming, the first condition was standard. This is a “nice to have” and may reduce some number of crashes per year.
A second example is also innocuous if largely unimportant. The infrastructure advises the vehicle about controls. Instead of relying on the vehicle’s cameras (or the driver’s eyes) to read signs and signals, or digital maps to know of their existence, the intersection would broadcast its state to the vehicle. This could be an improvement, but unless all intersections are so instrumented (a tall order in a big country with many levels of government), it cannot be relied on. An uninstrumented intersection does not broadcast that it is uninstrumented. Instead, the vehicle or driver must know that.
A third example that is potentially valuable is the simple provision of traffic information like speeds back to the system. Waze is an early version of this, but anytime you use Google Maps or Apple Maps, your GPS location is fed back, and if you are in motion, a speed is derived, along with a timestamp. This is turned into traffic information. More complex reports, like text based data from Waze or even Twitter, are additional less systematic examples. If there is enough data, that can be weighted appropriately, these kinds of data might be sufficiently accurate to provide useful travel speeds where there are no inductive loop detectors. It might even generate traffic or OD flow data, if the share of vehicles acting as probes at a given time is somehow estimated.
A fourth more dangerous example that is trotted out is the 4-way intersection where no one has to stop. In 2002!, I wrote this with a student, Xi Zou, in a (sadly unfunded) proposal (and this was hardly an original idea, it was in the air):
Concept (uniqueness) By integrating intelligent agents and Mobile Ad-hoc Networks (MANETs), the management of low-volume intersections can be devolved to vehicles, without requiring conventional traffic signals, or forcing extended stops at red lights. Advanced techniques such as digital maps, GPS, in-vehicle computers, and mobile wideband communications provide cornerstones of this new framework. Intelligent agents implanted in the vehicle represent the aims of drivers and management, as shown in Figure 1. The intelligent agents embedded in vehicles know the vehicle’s destination, and like adaptive cruise control, adjust the vehicle’s speed up or down. These agents continuously announce their id, position, speed, and acceleration to inform other equipped vehicles over a Mobile Ad-Hoc Network. (While radar and GPS can be used by one vehicle to determine another vehicle’s position and speed when it is in line-of-sight, it is insufficiently accurate to determine acceleration, or other attributes when the vehicle is obscured). Based on the position/speed/acceleration of other vehicles, a vehicle proceeds through the intersection at its current speed, slows down, or speeds up to avoid a collision. A consistent protocol used by all vehicles (based on each vehicle’s position, speed, and acceleration) determines which vehicle passes through the intersection conflict point first, both avoiding collisions and ensuring safety, as shown in Figure 2. Thus intersection management becomes a decentralized operation of a community of agents that might be part of a future mobile society.
Why is your concept better than those of others? (competitiveness)
Traffic signals were first installed in Cleveland in 1914, and there are more than 300,000 traffic signals now operating in North America, which control two-thirds of roadway travel each year (FHWA 1995). While great progress has been made in terms of passenger safety and road efficiency by this technique, there are still some limitations. Improperly operated traffic signals cause excessive delays that sacrifice productivity, waste fuel, and pollute the air. While side collisions are reduced, rear-end collisions are increased at signalized intersections. Dissatisfaction in intersection operation has become a serious problem faced by traffic operators.
In intersection management, the number of accidents and total travel time of traffic can be reduced if drivers/vehicles are aware of the states of other vehicles near the intersection. Current technologies enable vehicles to be aware of their own real-time states such as position, speed, and acceleration, and to communicate with other entities (vehicles or management center). Furthermore, vehicles near an intersection can acquaint themselves with the overall and detailed information about other vehicles. They might evade each other efficiently, avoiding potential collisions. Unlike conventional signal control, we propose microscopic control in which the behavior of each vehicle is adjusted individually. Instead of using stop-or-go control, more flexible passing maneuvers can be used to increase the capacity in every approach and reduce average travel delay, as shown in Figure 3. On the other hand, without traffic signals that suffer from improper settings, delayed maintenance, and malfunctions, the proposed framework takes advantage of the distributed sensing and computing resources in vehicles, which makes the system more robust, flexible and economical. Furthermore, a reality we will face in the future is that every vehicle will be capable of communicating with other nearby vehicles, forming a mobile ad-hoc network. This advanced traffic management will become a software function that is implanted into the vehicle information system.
So in short, Vehicle A approaches from the west, vehicle B from the north. They will both hit the intersection conflict point at approximately the same time if neither changes speed or trajectory. The traditional rule for this is a stop sign (or signal), either 2-way or 4-way, which delays at least one vehicle needlessly. With V2V (or V2I) communication (what we called MANET), some protocol is established (e.g. yield to the right) whereby one vehicle (say the southbound vehicle) decelerates relative to the other, and they miss each other by a fraction of a second, traveling almost full speed. There is no delay, no needless braking, no collision, and best yet, for purposes of greenwashing, the environmental impact is lowered because we somehow still all drive gasoline powered vehicles in this imagined future.
The video below from the UT Austin CS department (which is highly confused about what autonomy means and other things in many many ways) illustrates the concept.
Even more ridiculously there is this:
Could this last case be done with autonomous vehicles? We argued at the time and I would continue to argue, it can only be done safely with autonomous vehicles. You cannot safely rely on the other vehicle saying it will decelerate, you must ascertain that yourself. What if it was you who was to decelerate, how do you know the other vehicle doesn’t decelerate as well, just because it said it would maintain speed? The answer is, you measure it with your own sensors. Perhaps a deer (or a child!) jumps out into the road just past the intersection. The other car slams on its brakes near the intersection. You cannot rely on the pre-established protocol. Perhaps they communicate this change of plans to you. But there is a lag (or worse, buffering). Even a fraction of a second is too much. Your vehicle must detect its change via sensors (which are always faster than communications, just try to beat the speed of light) and react accordingly (decelerate/accelerate/swerve/brace for impact).
The algorithms, even if coordinated, would need to be able to resolve the potential conflict without communicating. To illustrate conceptually. Each vehicle, if it believes it would reach the conflict point first (to, say, 8 decimal points of precision) accelerates, if it believes it would reach second, it would decelerate. If one vehicle accelerates, and the other vehicle decelerates, we have harmony, and can avoid the collision as each vehicle continues to update velocities and speeds in reaction to the other vehicle’s behavior until the possibility of conflict is avoided entirely.
Certainly there are possibilities that both cars accelerate or both decelerate. This could be due to GPS location measurement error (each thinks it reaches first (or second)). In this case a second tie-breaker might be required. For instance, the easternmost car always wins. Both cars should be able to detect with sufficient accuracy which car is easternmost (in this case, the southbound car). So then the eastbound car decelerates, and the southbound car maintains speed or accelerates. Hopefully there is harmony, both cars adjust speeds accordingly as conflict is avoided. No explicit communication was required.
One can imagine other, better tiebreakers, or the need for third tiebreakers (the intersection is X-shaped rather than +-shaped, so they both think they are easternmost), but this can be mapped out. In fact it only requires one vehicle to be autonomous. If after, say, three rounds of chicken (both vehicles increasingly accelerating to the intersection) which is ascertained in less than a second, then the AV decelerates hard to avoid a conflict.
You don’t have to like the conceptual algorithm I made up here to believe that such algorithms are possible.
Now, if all vehicles were centrally controlled, what would happen? While this might appear to improve safety (if the central algorithm is indeed safe), it does so at huge cost. The prospects and magnitude of systemic failure just increased enormously. Any bug, any hack could shut down all transport or cause multi-vehicle collisions. This seems terribly unwise.
So why this recent interest in connected vehicles from the government? It comes from the same people who were pushing Automated Highway Systems (AHS) in the 1990s. Full AHS was a great technology with no deployment strategy, it was a classic chicken-and-egg situation with no middle ground which could be ratcheted to. It only worked on dedicated infrastructure with dedicated vehicles. No one would build dedicated infrastructure if no one owned dedicated vehicles. No one would buy more expensive dedicated vehicles in the absence of a large network of dedicated infrastructure. This was obvious to some of us in the 1990s, but nevertheless AHS was funded until after the technically successful NAHSC Demo, at which point it was cancelled. Certainly there were some autonomous technologies mixed in with the more connected (platooning) technologies, and some of those have reappeared, so I don’t want to imply the research program was a total loss. In contrast, most progress in automated vehicles comes from outside the traditional transportation community, namely the roboticists. Their demos were funded not by the US Department of Transportation, but by the more outward looking Department of Defense, which supported a series of grand challenges.
Attaching the governmental word “Connected” to the more Silicon Valley “Autonomous” is a type of techwashing.
In short, connected vehicles, if used for information, can be a benefit. If CVs are to be used for any kind of control, if designed with any algorithms, they are at best a luxury, and possibly a threat, but definitely cannot be relied on. They are neither necessary nor sufficient for safety. Autonomous vehicles, designed with the appropriate algorithms, in contrast are both necessary and sufficient for safety (as human-driven vehicles have already proven themselves unsafe).
In the end, all vehicles may be both autonomous and connected, but it is far more important they be autonomous than they be connected. Fortunately, that is the sequence in which the technology will be meaningfully deployed.