Category Archives: autonomous cars

On the Road and Off the Map: Mapping Roads for Self-Driving Cars in an Over-Paved World

Even as autonomous cars provide a more radical change in patterns of mobility than any change in transportation, the amassing of data to put places on the map comes at its cost.  Indeed, even the hopes to provide a high-density record to be able to navigate roadspace leaves an eery imprint for what it leaves out:  even as the existence of a crash-free world of the automated vehicle beckons, the high data density maps being developed to place space on a map presents a circumscribed landscape terrifying insofar as one rarely appreciates the costs for what is left off the map.  And despite the possible benefits for autonomous cars, the first maps made for ensuring safe driverless driving test the not only the huge amounts of data able to be included in maps, but the problems of data selectivity and of the prioritization of information, raising questions not only about the richness of the maps for self-driving cars, but the sorts of world we want to register for autonomous cars.  The eery ghostly roadways of the maps made for self-driving cars seem quite appropriate:  they track the road as able to be inhabited by the car, after all, rather than the pedestrian-centric spaces around them.  But is their ghostly character also not a cost?

All maps are made to meet demands, and the expanding market for maps for self-driving cars is no exception.  But if we have become able to map traffic and routes for some time, the ghostly sense of inhabitation in maps for self-driving cars seem worth reflection–for the image of the world they create; the ethics of mapping the road conditions, and how theses maps orient us to the world. Fort he intelligence of such maps, made to be machine-read rather than read by humans,  propose a different notion of the “inhabited world” that is in truth increasingly closer to the road-covered world that we increasingly inhabit.  While the safety of such maps effectively allow us to be passengers in such self-driving cars, they also render a new sense of the worlds in which we are inhabitants.

For the haunting ghostly worlds that maps for self-driving reproduce and create provide an odd record of our increasingly paved over and decreasingly roads-free landscape.

Driving is among the most familiar extension of an embodied experience, and the most familiar experience of navigation and way-finding that we have today.  But as maps are increasingly present behind the wheel, as it were, and built into many cars, today, both in the form of dashboard monitors, handheld devices, and disembodied voices, the relation of the map to the experience of driving has changed.  As maps have become data and datasets, we have no only constructed far more visually elegant renderings of roads and driving conditions.  As the maps for driving have departed from the over-folded pieces of paper, often ripped or worn at the crease, that used to be stuffed into the romantically named “glove compartment” and migrate underneath sun visors or into  the side-compartments on front doors, into interactive experiences that we read, they have in many ways transcended our abilities for attention.  And the increased demands for attention in our society and even for our drivers has led to a new market not only for for data rich maps, but for the maps that would help guarantee the safety of self-driving cars.

In an age where Google dominates mapping, creating the tools to develop maps for autonomous vehicles–“self-driving” cars that navigate by LiDAR software, real-time radar and laser sensors, streaming data libraries and programs–


perched processing directions for URBER.png


–which stands to prove the most important mapping innovation since the satellite, and perhaps the most valuable ever, as over thirty companies are applying to test-run their own self-driving cars in California, seat of the future, and the winner seems destined to be the one with the most complete and sophisticated mapping tools.  The tools planned to allow the cars to navigate real space don’t provide anything similar to a recognizable landscape, but Google’s driverless car division–Waymo–used the code-name ‘Chauffeur’ to refer to the armory of LiDAR tools as if to humanize the tools by which autonomous cars will be instilled with the ability to develop an effective cognitive relation to space.   Although autonomous cars may threaten to overturn the hegemony of Google has retained as a mapping engine,  the new remapping of the freeways also threatens a changed relation to most all extra-urban off-road space.   Is the growth of the market for self-driving cars not in itself emblematic of a new relation to space, where the car is less the instrument of exploration or navigation–the Keruoac’s image of being “on the road”–but a now bulky mode of transit and commuting, whose increasingly mechanical modalities of operation seem to be best performed by an artificial driver, built-in to the car.  Even as it is foretold AI is destined to replace increasing numbers of workers with world-changing effects that are only “50 or 100 years away,” we have kept fears of economic shocks and needs for massive retraining at bay, but face a profound fear of decreased human agency.

The diminished agency of the human is perhaps no more apparent than in the rapid race to design maps for self-driving cars–maps read by cars to familiarize themselves with traffic conditions and their routes, in ways that dispense with human judgment behind the wheel–one of the most privileged sorts of agency in existence–even if the maps for self-driving cars are now limited to the most mechanical forms of transportation on “smart highways” and shipping routes.

What sort of intelligence is lost, one might well ask, and what gained?


I.  The Intelligence of the Map

While not likely to ever be as dense with pictorial detail as a topographic map, but seeking to provide the crispness of an older road map shows a matrix of paved routes of shifting thickness that seem so eerily modern in their configuration in a Pennsylvania Road Map of 1926 that seems to invite roadsters to explore Western New York–





While the configuration of this disembodied network of pavement recalls the instructions for a map for self-driving cars, the range of maps that are currently being crafted for self-driving cars condense not only roadways, but road conditions, lane-changes, stop signs, speed limits, curving interchanges and current traffic conditions, all absent from the creased maps drivers once stowed in glove compartments to keep at hand, as well as the content needed for lane changing, intersections, speed limits, and navigating unforeseen obstacles that lie off the map.

But if the map is often imagined as an open book, rich with a variety of places and spatial reference points that can only be distinguished by diverse fonts and typography–




–the far more intangible nature of the algorithms and instructions comprising maps for self-driving cars are necessarily far more prescriptive than they invite exploration, as machine-readable texts.   Despite their high data density, such maps avoid the topical details encoded in early 1915 highway maps of local topography to suggest their continuity, that announced the cartographer’s art at encrypting information on two dimensions–





–maps for self-driving cars rather combine a somewhat skeletal sense of extreme data richness of local road conditions, to which they assimilate real-time information of the roads themselves, fading out all unessentials, without any expectation of addressing a human eye.  Indeed, they can’t be described, perhaps, as forms of authorship, in an individual sense, because they are “written” and “surveyed” by sensors of the very cars which accumulate data needed for their content.

These maps raise the question of how the set of instructions that they will give cars to navigate roads and relate to the highways on which they travel and the traffic they also have to navigate can be a text–able to be claimed as a form of intellectual property and as a product–rather than a set of instructions, and, indeed, what sort of liability will lie within the maps for any possible accidents that occur within autonomous vehicles.  Indeed, although the questions of culpability and liability are all too absent from the sleek maps that promote the range of data in such “high density” maps of the roadways, the questions of liability could not be far off the minds of their designers.  But they demand to be explored, anyway, for the powerful nature of the contents of their design.

The functionality of such maps reflect the sort of traffic-maps that have long been provided to human drivers, but are radically pared-down versions of the same.   Take, for example, not only Wayz maps–but the improved maps of traffic intensity that are produced in strikingly color-coded precision to foreground traffic flows, rather than human buildings or monuments, but offering an immediately striking means of showing the traffic conditions of a city and its routes of traffic in Washington, DC to a human eye.




Mapbox cartographers have worked hard with their style sheets to create a new iconography able to distinguish prominent the tunnels, show intersections of traffic and onramps, that blocked out the habited areas of lower Manhattan, unlike most any maps that were made of the area, but link driving routes to transit lines in ways that improved the legibility of routes, improving on the right-hand 1.0 to the 1.2 to the left.


tunnels, intersections mapbox.png.jpg


But if the problems of rendering the notion of progress or itinerary is already revealed in these color coded and sized streets, the density of data in the maps for self-driving cars, which dispense with the pictorial symbols in favor of parsing the density of data dots for ready access.

Similar sorts of rendering of interchanges have been developed by online mapping agencies that seek to render traffic flows and in real-time, as a way of providing the sort of up-to-date information for human drivers that are especially challenging in New York, where they seem to update radio traffic, but condense a wide range of news about relative traffic congestion in ways that can be readily grasped, so that drivers can maybe not use them to navigate city streets, but at least to survey the lay of the road.

The roadways indeed have been gaining an increasing entity of their own in some of the traffic maps provided by Mapzen’s Transitland, which offer a way to imagine driving in ways that almost verge on the ability of a self-driving car, although it is fair to say that maps as the below seem to use a human capacity from moving from the general overview to the specific interchange that autonomous driving cars may not possess.


Manhattan Dumbo.png


The considerable beauty of these renderings in real-time lighting offer their viewers suggest a particular chromatic appreciation of the salience of the roadways that offer a further pleasure of map-reading–here combined with a sense of building heights and shadows–to capture the time-sensitive notion of traffic conditions that can be readily appreciated and intuited, as if to make the map a “smart” surface, from early morning to night.


Lower Mtananhat AM.png

Lower Manhatan MapZen.png

Night Time Manhattan .png


Yet in capturing the essence of travel in the map–and distilling the voyage to the essence of road conditions–the maps for self-driving cars create an oddly isolated sense of the roadway, largely limited to the paved surfaces without much inclusion of the overall pan.


2.  The Road and the Path

The disembodied nature of the maps for self-driving cars provide a new avatar of the extended of artificial intelligence and its challenge to displace an embodied experience.  If maps remain among the most human of creations to orient viewers to place and space, allowing us to navigate and master a spatial continuity we can’t otherwise readily perceive, training cars to read this space depends on the challenge of conveying the tacit familiarity with roadscapes into machine-readable form.  While such maps are not often considered similar graphic embodiments of space, the embodiments these maps offer of roadscapes demand to be examined.  For the notion of driving as wandering–or wanderlust–seems to be skipped over entirely in the maps for self-driving cars, which one could only say that Jack “Nothing-behind-me,-everything- ahead-of-me,-as-is-always-so-on-the-road” Kerouac would find queasy-making if not downright repellent.  Rather than wondering insistently “what’s in store for me in the direction I don’t take?” the adventure of rolling on for the man who preached that “the road is life” would find the denatured aspect of the roadways and roadscapes all too much akin to disembodied instructions.

For the interest in making maps singly for self-driving cars is so removed from human skills of map-reading to make them profound alienating landscapes, and yet eerily inviting to inhabit, raising questions about the safety of their use but also awe at the level of details gathered and organized in their visual fields.  These are not landscapes occupied by humans, to be sure, but may suggest an increasingly common aspect of the landscapes which we are creating for cars, and the ways we are learning to navigate an increasingly paved world, where roadless areas have not only diminished but are increasingly rare and indeed unfamiliar.  The ways that the landscape that will be used by the self-driving car seems to register on-road experience for self-driving cars are ghostly landscapes of the “roaded” world, as much as they place us in a space where we are increasingly removed from reading maps.  The contrast to the human landscape through which we might walk to orient ourselves, as Jeremy Wood sought to render by showing personal travel as a sort of “geodetic pencil or cartographic crayon” that registered his familiarity during personal travels around London by combining GPS tracks over seventeen days of walking–



Jeremy Wood, “My Ghost” (2009)


–the maps for self-driving cars are less concerned with intensity of travel, or pulled by the interest of sites of pleasure, work, and curiosity, but they define a set of pathways abstracted from place, but designed to ensure safety.

Indeed, in contrast to the contingency of navigation or exploration in a world we know by traveling, and learn while walking, the maps for self-driving cars are a sort of synthesis of streets, stripped of cities or sites–rather than the sort of matrix of spatial relations to which we are accustomed.  We’ve long expected maps orient human readers to the continuity of inhabited space that affirm regional integrity and offer needed cognitive way-finding tools.  The continuity maps create makes them among the most humanistic of documents, extending cognitive skills and establishing needed frameworks for understanding space and place as less than abstractions but in concrete terms.  The cognitive magic of embodying spatial relationships to be grasped at a glance functions in large parts by the success of their selective conventions as tools to place ourselves in a form of vision accessible to the human eye, condensing markers that often lie in the land to a form we can readily engage.  The new maps that promise to orient self-driving cars to the roadways rather focus on charting data on flow, and the constraints of space, rather than they organize space as a recognizably continuous record of the roadways.  They offer what one might call the “First World” perspective on our changing inhabited space.

And so the dramatically curtailed nature of the mis en scène of the maps for made for self-driving cars offer seems to demand attention:  to be sure, such maps make less appeal to human cognition, offering a purely machine-readable sense of place and of space oddly removed from human habitation.  But they condense and transpose a human relation to the landscape into readable form that, in the end, serve to orient us to how we navigate and voyage through space.   Is this mapping from a machine point of view, or is evidence of the new dominance of road space and roadways in an increasingly over-inhabited world, where a sense of place has been increasingly eroded within a continuum of constant transit, far more sensitive to flow than place or space?

A rendition of the spatio-temporal database. Source: Zenrin Co.

A rendition of the spatio-temporal database. Source: Zenrin Co.

The uncanny absence of engagement with the natural–and the transformation of the map, as it were, to a network of man-made roads that is removed from a human landscape.  Once stripped of the real mosaic of place and removing human settlement from a web of forests, landscapes, or wilderness–and indeed entirely from the natural and non-man-made world:–the concentration on roads filter out any sense of motion through space from any experience outside a car.  In an overly detailed armature of isolated itineraries, organized from discrete data points, the maps designed for autonomous cars present a denuded landscape, stripped of context that is not immediate, and without any accreted human knowledge preserved in place-names.  If the map is often described as a text, Hi-Def maps replace text with rich data points and a feeds of real-time information from sensors.  Rather than locate oneself in relation to space by successive markers in a landscape along a road or pathway–



D.T. Valentine and George Hayward, Common Lands between three- and six-mile stones (1799) (Museum of the City of New York)

–the condensation of road signs and driving conventions seek to allow a car to orient itself to space in ways that strip space of its tactile surroundings, doing the work in many ways for orienting oneself to signs located along the roads.  Rather than chart location for a viewer, HD maps for self-driving cars are sensitive to capturing the constraints and flows by which a car can orient itself to the shifting traffic and crowded lanes that organize the changing terrain of highway space, an area to which so much increasing land cover has been dedicated across the inhabited world:  the HD maps take the space of the highways something like as a proxy for the inhabited world.  As records of the shifting space of the speedways, these maps promise to register the complex calculus of highways to promise a safe trip for future passengers.  If J. B. Jackson famously suggested the rise of “auto-vernacular” landscapes in the United States whose deceptive sense of ‘placelessness’ reflected the changing middle class lifestyles of suburban tract homes and suburban subdivisions and strip housing located along highways, the maps for self-driving cars are cultural landscapes stripped of signs of habitation save the roadways.


3.  The Risks of the Maps

The provision of these maps by sensors register real-time accounts of spatial relationships that suggest the range of tools by which we have come to orient ourselves to the world’s paved highways:  yet at a time when Google Maps register space universally, from a synthesis of local surveys and satellite imagery, the maps for self-driving cars provide a downloadable cognitive framework, preserved on the cloud, for how cars can orient themselves to space by surveying their real positions on roadways, and depend less on human drivers in an age when, the distracted driving is so widespread, according to analytics company Zendrive, that American drivers use phones on 88% of the journeys they make on the road–and in spend 3.5 minutes per hour looking at their phone, and not only for directions.  In an age of mass-distraction where drivers can’t be as trusted to watch the road, maps for self-driving cars are perhaps needed to reduce risk–even two-seconds of distraction on your phone increases individual drivers’ risk by 20 percent; the National Highway Traffic Safety Administration recently counted almost 3,500 people deaths in distraction-related car crashes in the United States in 2015, and almost 400,000 injured, and the problem stands to worsen as after a recent trend toward declining numbers of driving fatalities, fatalities due to distracted driving are substantially rising, and especially among teens.  The statistics are unclear, as accidents involving cell phones are under-reported by authorities.  At a time when paper highway maps are no longer consulted, and in fact rarely sufficient–and navigation has migrated to the hand-held phone, if not Siri, it’s no surprise the need and market for maps for self-driving cars has so markedly grown.


Screen Shot 2017-04-18 at 11.38.22 AM.png


The response to such distraction has been to provide a newly comprehensive map able to interface with self-driving cars, in ways that might allow our existing habits of distraction to grow.  In a sense, this is a local–as much as a universal–recuperation of the perspective of the single surveyor, providing a local automobilistic model of mapping with advantages one can’t gain from satellite/LiDAR alone, or a filling in of the gaps.   But the maps for self-driving cars are not made for human audiences, and as much as orienting individuals, offers perspectives of an inhabited world that is defined only by the extent of traffic on roadways–and which consumes increased cognitive attention of commuters–but which offers an extremely alienated perspective on the world.

Continue reading

1 Comment

Filed under 3-D maps, autonomous cars, HD Maps, machine-readable maps, self-driving cars