In our home city of Melbourne, as in countless others, the COVID-19 pandemic was rapidly and starkly visible in the urban landscape. Emptied of human bodies, city centre public spaces usually teeming with people were suddenly still and quiet, apart from a handful of cleaners, construction workers, and homeless city dwellers. Architecture that usually formed a backdrop to bustling activity was suddenly, implacably visible. As photographer Pete Dillon put it, the city appeared to be “holding its breath” , waiting for normal life to resume. But of course, the city was not empty or still, or caught in a suspended intake of breath. In addition to the people who continued to move through or inhabit public spaces, there were a myriad of other animals, things, objects and technologies still shifting, flying, blowing and traversing the transformed urban setting. The city remained vital and lively, even if we now had to look beyond the familiar human forms to apprehend it. In this essay we will counter this vision of urban emptiness by focusing on one increasingly common technology: robots. During the 2020 pandemic, robots have worked to reveal human-centric understandings of emptiness, which mask the vital and dynamic presence of other types of organic and inorganic bodies.
This essay’s starting point is how COVID-19 has encouraged an increase in the use of robotic technologies, and how the growing presence of these technologies in our cities and towns during lockdown conditions demands our attention because of the distinctive ways they can potentially shape the future city. The examples of new robot uses driven by COVID stretch from hospitals to urban parks, and from food delivery to sanitation. While this is primarily driven by the need for infection control, from a robotic perspective, the pandemic has brought a welcome simplicity to the urban environment, providing conditions in which robots work best. Cities with more certainty and less contingency are amenable to the robotic logics we outline here. The startling emptiness of usually busy shared spaces therefore provides a productive backdrop to thinking about the capacities and limits of these technologies, but also opens a route to speculating about how they might better contribute to the complexity of urban space. As we will show, robotic technologies treat space in particular ways—as predicted, partitioned and datafied—and this has important implications for our cities as the use of robotic technologies grows.
In tracing these arguments, we build on Macrorie et al’s “whole city” approach to robots, one that explores their capacity as a “mode of urban restructuring”, where life is now “shaped by extended and expanded robotic and automation possibilities”.1Rachel Macrorie, Simon Marvin and Aidan While, “Robotics and automation in the city: a research agenda,” Urban Geography (2019): 2. DOI:10.1080/02723638.2019.1698868. We consider these possibilities in the context of the profound changes to our ways of moving through and inhabiting our shared urban environments imposed by COVID-19 restrictions, and use this unprecedented context to outline a framework for thinking about robotic logics of public space. Contrasting these with more ‘human’ treatments of space and place, we conclude by speculating on how this might continue to influence our urban public space as we emerge from the worst of the coronavirus pandemic.
Robotic logics of public space
One of the most valuable aspects of robots in the context of COVID-19 is that they are inorganic. They cannot themselves get sick, and can be disinfected to diminish the possibility of contagion as they encounter people. Despite this, they have a technological vitality that advances sociologist Deborah Lupton’s characterisation of data as ‘lively’. Lupton and others have shown how data is made meaningful because of the affective and bodily aspects of its creation and use, its potential effects on bodies, and its movement and fluidity as it is put to use.2 Deborah Lupton,The Quantified Self (Cambridge: Polity Press, 2016). An example is the data created by health-monitoring or self-tracking fitness apps where the meaning of data is constituted algorithmically, bodily, affectively, and spatially.3Shanti Sumartojo, Sarah Pink, Deborah Lupton and Christine Heyes LaBond, “The affective intensities of datafied space,” Emotion, Space and Society 21 (2016): 33-40. This work showed how research participants learned to understand their bodies and capabilities through the data that was generated with each workout. Machines, computers, people and their surroundings were all continuously brought together in a stream of data that was lively both in the conditions of its creation, and in its use to monitor, understand and potentially modify the human body.
Going beyond this, in addition to constituting their environments through the creation or receipt of lively data, robots also demonstrate their liveliness in the way they can physically act in and change their environments, for example by collecting rubbish or scrubbing floor surfaces. We often do not know the degree to which robots in public space are autonomous, or the extent of their capabilities, making their liveliness all the more intriguing. Moreover, they exercise this liveliness in a world that includes and must account for many elements and forces that extend beyond humans. For example, in work focused specifically on robots and more-than-human actants, geographers Christopher Bear and Lewis Holloway take a multivalent approach to technological liveliness.4Christopher Bear and Lewis Holloway, “Beyond resistance: Geographies of divergent more-than-human conduct in robotic milking,” Geoforum 104 (2019): 212-221. They detail the affective, technological and organic/inorganic combination of cows, people, milking stalls, and robots in automated milking systems. This research gives us an important account of robotic assemblage that shows what might come into view when we adopt more-than-human perspectives, but it is directed at the achievement of a particular goal—the efficient milking of cows.
In contrast, the robotic logics found in public space, outlined below, are not tasked with fulfilling one specific duty. Instead, they stretch across the many different things that robots might do—or might be able to do—which already include delivery, cleaning, food handling, health assistance, customer service, surveillance, health-service telepresence, and monitoring of spatial distancing. Our comments below are not meant to preclude any other activities, but rather to reflect on the general movements and activities of robots based on what we have observed over the first half of 2020, when a pandemic brought the risk of contagion between human bodies, and our cities shut down as a result. From the point of view of robots, and feeding into the logics we describe next, urban spaces were not so much empty as they were simplified. This favoured the robotic preference for certainty working under models of prediction, partitioning, and datafication, which we unpack in the following sections. The empty urban landscapes of COVID-19 made these logics evident and apprehendable in new ways.
Predicted
In the flow of everyday life, we not only move from one activity to the next, but we do several things at one time, start and stop, and may dream or fulminate about completely different things as we go. According to anthropologist Tim Ingold, we are enmeshed in our environments as we engage in the “tasks of human dwelling…within the process of becoming of the world as a whole.”5Tim Ingold, “Taking taskscape to task,” in Forms of Dwelling: 20 Years of Taskscapes in Archaeology, eds. Ulla Rajala and Phil Mills (Oxford: Oxbow Books, 2017): 23. That is, as time passes, our worlds are made through our activities, thoughts and feelings that cannot be severed or removed from our surroundings. We are in the world ongoingly, and the world is in us. This connotes a sense of flow—as time passes and as we and the world ‘become’ together—that we can discern in our everyday feelings.
Robotic logics, however, pull in a very different direction. Like many computational technologies, robots are programmed to work on identifiable tasks that must be anticipated before they can be attempted. Robots are designed to do specific things—attach car parts in a particular way, deliver parcels of a certain size, shine ultraviolet light long enough to kill dangerous pathogens—and these are often repetitive, highly precise tasks that can present considerable risks to humans. These tasks, however, are based on assumptions about what robots will encounter and how they will manage in their environments, which means programming behaviours that are based on prediction. Although some robotics research is addressing uncertainty, for example in the context of driverless vehicles, usually the world of robots is treated as knowable before it can actually emerge. Uncertainty is managed and diminished as far as possible rather than being treated as generative, although work on curiosity-driven robots is an exception.6Yoko Akama, Sarah Pink and Shanti Sumartojo, Uncertainty and Possibility: New Approaches to Future-Making in Design Anthropology (London: Bloomsbury, 2018); Pierre-Yves Oudeyer and Frederic Kaplan, “Intelligent Adaptive Curiosity: a source of Self-Development,” in Proceedings of the Fourth International Workshop on Epigenetic Robotics, eds. Luc Berthouze, Hideki Kozima, Christopher G. Prince, Giulio Sandini, Georgi Stojanov, Giorgio Metta and Christian Balkenius (Lund University Cognitive Studies, 2004): 117. This is fine for tasks that require efficient repetition in controlled circumstances, such as a manufacturing line or a warehouse shelf grid, but becomes problematic in contingent and complex public spaces. We get a taste of this in a 2019 report from Emily Ackerman, a wheelchair user in the UK who was not recognised by a Starship delivery robot when she was crossing the street, leaving her to manoeuvre abruptly over the curb before the traffic resumed. Here, the robot’s programming did not anticipate a person in a wheelchair, and so was not able to recognise her, to disruptive effect.
An outcome of this robotic logic of predictability is navigability—that the world is able to be understood through maps or other sources of data about the machine’s surroundings or the surroundings-to-come. This is important for mobile robots delivering things, cleaning or undertaking security patrols. However, as anyone who has used a GPS in an unfamiliar city knows, maps fall quickly out of date, and provide only limited forms of information, which can be catastrophic.7Allen Yilun Lin, Kate Kuehl, Johannes Schöning, Brent Hecht, “Understanding “Death by GPS”: A Systematic Study of Catastrophic Incidents Associated with Personal Navigation Technologies,” CHI ’17: Proceedings of the 2017 CHI Conference on Human Factors in Computing Systems (2017): 1154-1166. https://doi.org/10.1145/3025453.3025737. Even in recent advances in Simultaneous Localisation and Mapping (SLAM), when robots (or autonomous driving vehicles) continuously learn the map of their environment as they operate in it, the world is still understood as mappable, and therefore knowable, even if this understanding is arrived at on the move. Navigability makes salient the tension between the contingency of public space on the one hand, and the logic of predictability on the other. In always seeking to anticipate what the machine will encounter, robot programming treats the world’s becoming as somehow knowable or predictable. This is a way of understanding spatiality that is literally built into robotic technologies of many kinds, but that also drastically simplifies the environments they work in.
However, the city is not a simple environment. One response to this has been to make robots less autonomous than they may appear. Kiwi delivery robots in the United States, for example, are supervised by remote workers who monitor them every ten seconds and who can take over their controls if necessary, creating a demanding and heavy workload environment for low-paid labourers.8 Choon Yue Wong and Gerald Seet, “Workload, awareness and automation in multiple-robot supervision,” International Journal of Advanced Robotic Systems 14, no. 3 (2017): 1-16. https://doi.org/10.1177/1729881417710463. This limit in robotic capacities—and the demands on human workers that result—is becoming more and more important as delivery robots hit the streets in the context of the COVID pandemic. At George Mason University, for example, the return to campus in autumn 2020 will see robots used to deliver meals as part of the food service offerings, bringing them into shared spaces in greater numbers, but with as yet unknown effects on those spaces.
Partitioned
It follows that if robots treat the world as predictable, then their responses can be understood as partitioned. By this we mean that programming generates robot activities as composed of many smaller decisions and movements strung together. This implicitly treats the world as able to be partitioned into discrete units of apprehension, assessment and decision-making, just as it is able to be predicted. Put differently, robotic logics treat the flow of the ‘becoming of the world’ as not only divisible, but necessary to be partitioned if programmed goals are to be met.
Robots are necessarily task-focused—to deliver, clean or patrol, for example—and these tasks are broken down into smaller component parts. A simple example is when a robot moves to a particular place, assesses if the place needs cleaning, and then executes the cleaning activity. Along the way, it would need to monitor who and what was around it, stop if it encountered something in its path, and perhaps navigate around this impediment. Tasks might be tackled sequentially or in parallel, such as when a robot moves and monitors at the same time, or stops, assesses and then cleans in that order. That is, even when robots are able to do more than one thing at a time, in the main they still treat those things as separable.9There are, however, some exceptions, such as ‘end-to-end’ programming that seeks to train the robot system as a whole.
If a delivery or cleaning robot senses a person in its path, it may stop and wait for the person to go around it, seek more information from the person or from a monitor somewhere else, as in the example of the Kiwi robot above. However, it must first recognise this encounter as one that requires solving, and choose the right set of options for solving it. In Ackerman’s case, the robot did not appear to recognise her wheelchair as a person, and so was not able to respond appropriately. A robot must categorise its environment into aspects that it already knows about, and if it encounters something novel, its options are limited. Even if machine learning promises to enable robots to continually absorb and categorise new information, and perhaps respond to novel situations appropriately as a result, the simple tasks of delivery or cleaning robots do not yet stretch to this level of analytical sophistication. The robots that COVID-19 has made us more likely to encounter in our everyday lives remain limited by programming that treats the world as able to be partitioned, its flow able to be divided for analysis and action. In the tension between predictability and partitioning on the one hand, and the ongoing flow of spacetime on the other, the possibility of error is ever-present, as Ackerman reminds us.
The new urban emptiness of 2020, however, has greatly reduced the normal unpredictability of urban space. Combined with robotic resistance to viral infection, the pandemic has generated ideal conditions for robots in the city, such as those delivering medical supplies and food. Meanwhile, in smaller indoor spaces, robots are operating in physically distanced restaurants to limit contact between servers and the public. The less crowded conditions give them room to move and limit the unpredictability of busy indoor spaces. When physical distancing requirements soften and cities begin to fill with people again, however, the capacity of robots to manage the return to more complex environments remains unclear. Some reports claim that they do not make good waiters, because of their physical limits, or misgivings about how they interact with customers. Just as with the remotely-navigated Kiwibots, human helpers are usually close at hand.
Datafied
Like many other technologies that already enliven our smart cities, robots operate in a world of existing and potential data. In public space settings, they have the potential to connect to real-time information on gatherings of people, traffic flows or potential disruptions to their travel routes such as construction work. They are already being used to monitor infrastructure, such as water pipes, feeding back information about necessary repairs or upgrades.10Harutoshi Ogai and Bishakh Bhattacharya, Pipe Inspection Robots for Structural Health and Condition Monitoring (New Delhi: Springer, 2018). In the context of COVID-19, a robot was used to patrol a public park in Singapore, reminding visitors to keep their distance from each other and adhere to infection control protocols, while in China there are reports of robot drones being used to spot people not wearing face masks in public. As they connect up to other sources of information about the people, structures and other technologies in the city, the potential is clear for uses that are more integrated into the regular monitoring of urban spaces. The “social credit system” recently introduced in China is an example of the datafied spatialities that robots have the potential to participate in. Linking up data from a range of sources—financial transactions, facial recognition and even online dating sites—this system determines a score that can shape whether individuals are allowed certain activities, such as travel overseas or between cities.
Although this type of connectivity is not yet common, it nevertheless points to larger debates about the potential inequalities in code-based automated smart cities. Indeed, geographer Vincent Del Casino identifies the problematic discursive landscapes that can emerge in “a world of optimized ordering and regulation that relies fundamentally on the coding of social life into software.”11Vincent Del Casino, “Social geographies II: Robots,” Progress in Human Geography 40, no. 6 (2016): 850. Moreover, in a critique of “software-mediated techniques used to regulate and manage urban systems,” Klauser et al explore the power dynamics of everyday life “governed by code.”12Francisco Klauser, Till Paasche and Ola Söderström, “Michel Foucault and the Smart City: Power Dynamics Inherent in Contemporary Governing through Code,” Environment and Planning D 32, no. 5 (2014): 870. The dystopian potential of robots that work autonomously in response to ongoing data streams that reflect the ebbs and flows of city life has been explored at length in speculative film and fiction. ED-209, in the 1987 film RoboCop, “configured for urban pacification” and giving citizens “20 seconds to comply” with its orders, is an extreme example. Ethical concerns about data-based understandings of the world, however, are not only fictional. For example, in June 2020, Boston’s city council voted to ban police use of facial recognition technology because of the likelihood of misidentification related to skin colour. At the same time, major tech companies also withdrew or halted development of these applications.
Even if data and its algorithmic logics come to mediate our ways of interacting with each other more often, the inherent robotic bias towards orderly categories will always struggle with the contingent flow of everyday life. It is difficult to envision how robots can cope faultlessly with the inherent messiness of busy cities, and easy to see how this can result in undesirable and unjust outcomes. The COVID-19 shutdown has brought this into focus as quieter urban environments have highlighted the robotic logics we outline here.
Implications
More-than-human accounts of the city have explored how animals, ecological systems, technologies and other organisms and things play a role in shaping our shared environments.13Adrian Franklin, “The more-than-human city,” The Sociological Review 65, no. 2 (2017): 202-217. This has included algorithmic processes and technologies, as reflected in the growing literature on smart cities.14Jathan Sadowski, Too Smart: How Digital Capitalism is Extracting Data, Controlling Our Lives and Taking Over the World (Cambridge, MA: MIT Press, 2020). COVID-19 has accelerated the appearance of robotic technologies in our cities and towns, with new applications in hospitals and parks, for example, and wider uptake of delivery and cleaning applications, bringing them into the everyday lives of many more people around the world. In this short essay, we have begun to map out some of the robotic logics that govern these technologies, identifying their treatment of the spaces of our shared cities as predicted, partitioned and datafied. These logics necessarily seek to simplify urban complexity in aid of fulfilling quite desirable responses to the COVID-19 pandemic that have recently included effective disinfection, improved physical distancing and delivery of essential products to vulnerable people.
Given the desire to extend the use of robots to further uses in public space, there is the temptation to simplify their surroundings to make it easier for robots to be effective. Robots thrive in controlled conditions. In this sense, a COVID-19 world works to their advantage. Cities designed for robots would be more predictable, navigable and datafied, with knowable rhythms and patterns. We already see examples of robot-centric spaces with the creation of spaces on highways where it is easier for Automated Driving Vehicles to operate, and more recently, proposals for dedicated roads solely for these technologies.
However, such proposals come at a cost of the loss of the very things that make public space vibrant and valuable. Even worse, the increasing normalization of robotic logics in public spaces could reinforce the spatial injustices that already plague many cities. Urban public space can be wonderfully complicated, diverse and unpredictable; qualities that have long made cities irresistible for many people. COVID-19 gave us all a taste of what an opposing version of the city might look like—emptied of humans but populated by the logics of prediction, partitioning, and datafication. Looking ahead, the challenge for robotics researchers, urban planners, and others is to create robots that perform effectively in the face of urban complexity but do not diminish it. Indeed, the work of robotics should seek to actively contribute to this bustling environment in unexpected and generative ways. This means robotic logics that configure into and with public spaces as they currently exist, rather than impose new rules based on extant technological capacities. In other words, instead of designing cities to be simpler or more predictable, as we have all experienced during COVID-19, we should instead design technologies that take the unpredictability of cities as a starting point, and work to engage and even enhance them.
Shanti Sumartojo is Associate Professor of Design Research and a member of the Emerging Technologies Research Lab at Monash University. She researches how design and technology help to constitute how people experience shared spaces. Her Twitter handle is @bajak70.
Daniele Lugli is a PhD Candidate in Design at Monash University. As a design researcher and tattoo practitioner, her work investigates the sensory and social experience of individuals in the contemporary tattoo studio.
Dana Kulić is a professor in the Faculty of Engineering and head of Monash Robotics. Her research is focused on designing robots for people. She tweets at @ProfKulic
Leimin Tian is a research fellow in the Human-Robot Interaction group and the Human-Centered AI group at Monash University. Her research focuses on affective computing and social robotics. Her Twitter handle is @LeiminTian.
Pamela Carreno is a research fellow in the Human-Robot Interaction group at Monash University. Her research focuses on understanding and analysing human motion and behaviour for human-robot interaction purposes. Her Twitter handle is @PamelaCarreno11.
Michael Mintrom is a professor of public policy at Monash University, where he serves as the inaugural director of the university research focus area, Better Governance and Policy. He tweets at @mikemintrom.
Aimee Allen is a PhD candidate in the Australian Centre for Robotic Vision (ACRV) and Human-Robot Interaction groups at Monash University. Her research focuses on designing trusted and likeable robots.
Notes
↑1 | Rachel Macrorie, Simon Marvin and Aidan While, “Robotics and automation in the city: a research agenda,” Urban Geography (2019): 2. DOI:10.1080/02723638.2019.1698868. |
---|---|
↑2 | Deborah Lupton,The Quantified Self (Cambridge: Polity Press, 2016). |
↑3 | Shanti Sumartojo, Sarah Pink, Deborah Lupton and Christine Heyes LaBond, “The affective intensities of datafied space,” Emotion, Space and Society 21 (2016): 33-40. |
↑4 | Christopher Bear and Lewis Holloway, “Beyond resistance: Geographies of divergent more-than-human conduct in robotic milking,” Geoforum 104 (2019): 212-221. |
↑5 | Tim Ingold, “Taking taskscape to task,” in Forms of Dwelling: 20 Years of Taskscapes in Archaeology, eds. Ulla Rajala and Phil Mills (Oxford: Oxbow Books, 2017): 23. |
↑6 | Yoko Akama, Sarah Pink and Shanti Sumartojo, Uncertainty and Possibility: New Approaches to Future-Making in Design Anthropology (London: Bloomsbury, 2018); Pierre-Yves Oudeyer and Frederic Kaplan, “Intelligent Adaptive Curiosity: a source of Self-Development,” in Proceedings of the Fourth International Workshop on Epigenetic Robotics, eds. Luc Berthouze, Hideki Kozima, Christopher G. Prince, Giulio Sandini, Georgi Stojanov, Giorgio Metta and Christian Balkenius (Lund University Cognitive Studies, 2004): 117. |
↑7 | Allen Yilun Lin, Kate Kuehl, Johannes Schöning, Brent Hecht, “Understanding “Death by GPS”: A Systematic Study of Catastrophic Incidents Associated with Personal Navigation Technologies,” CHI ’17: Proceedings of the 2017 CHI Conference on Human Factors in Computing Systems (2017): 1154-1166. https://doi.org/10.1145/3025453.3025737. |
↑8 | Choon Yue Wong and Gerald Seet, “Workload, awareness and automation in multiple-robot supervision,” International Journal of Advanced Robotic Systems 14, no. 3 (2017): 1-16. https://doi.org/10.1177/1729881417710463. |
↑9 | There are, however, some exceptions, such as ‘end-to-end’ programming that seeks to train the robot system as a whole. |
↑10 | Harutoshi Ogai and Bishakh Bhattacharya, Pipe Inspection Robots for Structural Health and Condition Monitoring (New Delhi: Springer, 2018). |
↑11 | Vincent Del Casino, “Social geographies II: Robots,” Progress in Human Geography 40, no. 6 (2016): 850. |
↑12 | Francisco Klauser, Till Paasche and Ola Söderström, “Michel Foucault and the Smart City: Power Dynamics Inherent in Contemporary Governing through Code,” Environment and Planning D 32, no. 5 (2014): 870. |
↑13 | Adrian Franklin, “The more-than-human city,” The Sociological Review 65, no. 2 (2017): 202-217. |
↑14 | Jathan Sadowski, Too Smart: How Digital Capitalism is Extracting Data, Controlling Our Lives and Taking Over the World (Cambridge, MA: MIT Press, 2020). |