Not Just A Dot on The Map: Food Delivery Workers as Infrastructure

Food delivery platforms are location-based services that rely on minimal, quantifiable data points, such as GPS location, to represent and manage labor. Drawing upon an ethnographic study of food delivery work in India during the COVID-19 pandemic, we illustrate the challenges gig workers face when working with a platform that uses their (phone’s) GPS location to monitor and control their movement. Further, we describe how these, along with the platform’s opaque, location-based logics, shape the delivery workflow. We also document how the platform selectively represented workers’ bodies during the pandemic to portray them as safe and sterile, describing workers’ tactics in responding to issues arising from asymmetric platform policies. In discussion, we consider what we can learn from understanding gig workers as ‘infrastructure’, commonly overlooked but visible upon breakdown. We conclude by reflecting on how we might center gig workers’ well-being and bodily needs in design.


INTRODUCTION
Food delivery platforms allow customers to have a meal delivered to their doorstep at the push of a button.To successfully deliver an order, food delivery workers (also referred to as riders) need to navigate an urban setup "that is a complex and interactive network linking together disparate social activities, processes, relations with a number of architectural, geographical, civic, and public relations" [20].Food delivery platforms are location-based services in that they rely on GPS data to allocate, direct, and monitor work.While platforms observe workers primarily through their phones' GPS location, this study attends to workers as more than just dots on a map: we present an ethnographic study on what it is like to work for a platform that relies so heavily on location and location-based rules.
Drawing upon our study of platform-mediated food delivery work in Pune, India, during the COVID-19 pandemic, this paper examines the challenges food delivery workers faced during the pandemic and their tactics in interacting with changing platform representation.Our analysis encompasses feld notes, telephonic interviews, audio-video recordings of delivery rides (where the frst author rode with delivery workers), and detailed conversations with riders after a delivery.It illustrates workers' complex relationships with their surroundings and how their situated eforts lay the ground for the food delivery platform to function.
Our fndings, frst, show the tactics riders developed to work with the platform's location-based, opaque logic of order allocation.The platform's strict location-based rules created trouble during pick-up and delivery, and, in order to complete deliveries, workers needed to 'repair' the situation when something went wrong.Second, during the pandemic the platform chose to selectively represent workers in a way that would make their service appear safe and sterile.This resulted in three asymmetric changes that caused further challenges to the riders: measuring riders' body temperature, a masking mandate for riders, and the option for customers to choose a contactless delivery.
Our primary contribution is empirical: We present an ethnographic study, with supporting qualitative methods for added context, regarding the tactics gig workers used during the COVID-19 pandemic, and the challenges they faced due to asymmetric platform policies.Our study highlights food delivery workers as a critical yet often hidden layer of the platform economy: They not only contribute to platform business' functioning but also work in critical situations like the COVID-19 pandemic as essential workers, often putting their health (and economic gains) at risk.Further, our study illustrates the power asymmetry among stakeholders, including the platform, customers, workers, and even the restaurants.We conclude by considering our fndings through the lens of infrastructure, drawing on the foundational work of Star et al. [6,35,53], and by emphasizing the importance of centering gig workers' well-being and bodily needs in the design of platform infrastructure in the feld of Human-Computer Interaction (HCI).

BACKGROUND
Our review of prior literature is structured into three parts.First, we provide a short overview of scholarship that considers infrastructure as relational and ecological.Second, we consider gig work as a part of location-based HCI.Third, we discuss prior research on workers' well-being and corporeal issues in gig work.

Infrastructure
Within HCI and Computer-Supported Cooperative Work (CSCW), the concept of infrastructure has been important in attracting attention on the critically overlooked role of background structures and systems in enabling complex collaborative technical practice.Star (e.g., [53,54]) introduced the notion of infrastructure as going beyond physical hardware and software to encompass the sociotechnical and organizational elements that support collaborative activities.In CSCW, this is known as the work that makes the network work, often operating in the background, unnoticed until the arrangement breaks down.This "hidden", "invisible", or even "boring" part of technology that people and organisations rely on daily includes network protocols, standards, and databases, alongside the social norms and practices relevant to their upkeep [5].
This approach considers infrastructure as relational and ecological -not as a single, fxed system but a multiplicity of interrelated components.These components are fexible and adaptable, allowing for diferent confgurations to suit the needs of various collaborative activities.Understood in this way, infrastructure is also something that is subject to ongoing maintenance and repair.When disruptions occur, it becomes particularly evident how critical infrastructures are to the functioning of collaborative work.Building on this, researchers have documented at length how different users and infrastructures are constructed, spanning diferent physical, social, and digital infrastructures [35].In ethnographic work [14,27,55], authors have used the notion of 'infrastructuring' to highlight the process through which infrastructures are made in interactions between people, technology, and networks.This work has proven particularly valuable in understanding infrastructures that need to be actively reconstructed, particularly in non-western contexts [27,28].
In this paper, we connect these notions of infrastructure to gig work and, in particular, the work of food delivery riders during the pandemic.Recent, critical research in HCI has highlighted the often invisible or ignored work that supports the smooth and efcient functioning of technology platforms [18,41].As Irani points out, the invisibility of workers' labor is what "sustains the joyful optimism and celebrations of creativity" associated with technology mediated work [25, p.736], while simultaneously producing conditions of exploitation and precarity [44,60,61].Our research responds to the call this line of work has made to pay attention to the crucial role of workers -people who are locally embedded and bring along deep social knowledge -in aiding platform functioning [40].

Gig work as location-based HCI
Our focus is on platform-mediated food delivery, a type of gig work that is characterized by little control over when and where the work is done [62].Many gig platforms are geographically situated, relying on digitally mapped territories to plot their services.Car-hailing and delivery platforms, for example, need to know at least three 'live' locations marked on the digital map -the worker's current location as well as the pick-up and delivery locations -to match workers with orders and to plot and manage the work.Thebault-Spieker, Terveen, and Hecht [57] examine this dependency in UberX and TaskRabbit to highlight how four fundamental principles of human geography -distance decay, structured variation in population density, mental maps, and spatial homophily -manifest in these platforms.Our research connects with these principles and especially the key insight regarding the infuence of socioeconomic status since the Indian city where our study is located experiences continuous gentrifcation that impacts delivery workers' decisions and work performance.
Location-based interactions, enabled by smartphones and the burgeoning of internet connectivity, have been studied extensively within HCI and CSCW (see, for example, [3,4,16,63]).Literature on geographic HCI (GeoHCI) considers these interactions primarily from end-users' point of view [23].As one example, Colley et al. [12] studied the geographic efects of location-based gaming with Pokémon GO.As another example, Guha and Wicker's [21] study of homophily in friendship and surveillance networks among foursquare users discusses positive surveillance, with participatory, social, and collaborative interactions.This has implications for on-demand platform design where, unlike in the location-based applications these authors considered, workers have no option but to share their live locations.As Shklovski et al. [50] point out in their study of how location-based technologies transform the relationships between parole ofcers and parolees, the settings where these technologies are deployed are already sufused with complex social dynamics, and the implications of location becoming a tradable, technological object are complex.

Well-being and bodily needs in gig work
On-demand service platforms ofer a certain kind of fexibility, freedom, and independence to the workers -a reprieve from the demands of traditional work and, in some ways, the realization of the promise of development [51].However, despite the low entry barriers and benefts associated with fexibility, research has consistently highlighted the challenges this work entails for those doing it.Kinder et al. [32] note that digital platform's algorithm-based management contributes to information and power asymmetries that are pervasive in the gig economy.Drawing on a qualitative study with online freelancers and clients of the Upwork platform, they also highlight the multiplicity of actors already in place for creating alliances to work with, through, around, and against this form of management.In a project focused on the problem of unfair reviews that consider factors outside of a worker's control and can lead to lost job opportunities or even termination from the marketplace, Toxtli et al. [59] created a tool, Reputation Agent, to promote fairer reviews from employers or customers on gig markets.The tool was designed to automatically detect when an individual has included unfair factors into a review and, subsequently, prompt the individual to reconsider the review.The authors discuss how tools that bring more transparency to employers about the policies of a gig market could be a means to build empathy and foster reasoned discussions around potential injustices in gig work.
In response to the challenges gig workers face, further work has called for worker-centered design (see, for example, [19,36,38,64,65]).Ma et al. [38] encourage further exploration and advocacy for work support, inclusion, justice, and a visionary outlook for gig work, while emphasizing the need for a cautious approach to ensure that any changes do not inadvertently lead to further worker exploitation or damage the economic feasibility of gig work.Zhang et al. [65] used a participatory design approach to co-design interventions that center workers' lived experiences, preferences, and well-being in algorithmic management.Their research highlights how issues like information asymmetries and unfair, manipulative incentives, hurt worker well-being.The authors propose worker-generated alternative designs to address these issues while considering the competing interests of platforms, customers, and the workers themselves.Working with a policy focus, Schwartz and Weber [48] argue that information asymmetries are, in fact, a key feature of platform capitalism: platforms rely on them in their pursuit to turn a proft.Describing how these asymmetries impede workers' awareness of new rights and their access to policy benefts, the authors advocate for distributing responsibility in the enforcement of platform regulations, involving gig workers in policy design and implementation, and addressing information asymmetries in regulating platform companies (including but not limited to the gig economy).
Informed by research on platform responses to the COVID-19 pandemic [52], we advance this worker-centered line of research with a particular interest in corporeal issues.As gig workers gained the stature of an 'essential worker' during the pandemic, platforms had to acknowledge their embodied presence on the ground.Katta et al. [30] have argued that COVID pushed Uber to take a temporary turn towards 'embeddedness' and 'decommodifcation' of its drivers' labor.The authors highlight that the pandemic "made clear that platforms like Uber are inextricable from the local by reminding us of the materiality of the bodies they govern."Gig workers' 'essential' performance during the pandemic also emphasized their economic precarity [2], and triggered discussion about the need for a better safety net within the gig economy [10].However, limited attention has been paid to the bodily risk [8] experienced by these workers during the pandemic or to platforms' approaches to mitigating it [24,52].
In addition to centering workers' health and well-being in general, we might also ask whose bodily needs are discussed in appeals for more worker-centrered approaches.Many gig work platforms -ride-hailing and food delivery services as the most prominent examples -have failed to attract many female workers and, thus, rely on a male-dominated workforce (for more details on the gender distribution of gig workers in India specifcally, see [29]).Male-default values, as discussed by Ma et al. [37], permeate the platform's design as well as the manner gig work gets completed on the ground.While these gender biases have been refected also in researchers' focus in this domain, there are notable exceptions: Ticona and Mateescu [58] investigated care work platforms to show the limitations of the informalisation of work.Raval and Joyojeet [42] ofer a detailed account of the constraints female gig workers face in working with beauty and wellness platforms, and Anwar et al. [1] further explain how the platformization of traditionally gendered beauty work continues to align and retain the socio-cultural logic restricting women's lives in India.Next to gender issues, it is worth noting how location-based gig labor has been designed primarily with able-bodied workers in mind, in an 'ability-unaware' manner [46], although people with disabilities are over-represented in contingent and part-time work [47].

MATERIAL AND METHODS
This paper builds on data collected between May 2020 and April 2021 in Pune, Maharashtra, India, for a broader research project on how online food delivery workers cope with the drastically altered working conditions during the COVID-19 pandemic when they were allowed to work as 'essential workers' [9].Adhering to rules and regulations related to the COVID-19 pandemic, research methods evolved from remote telephonic interviews to feld visits, and, subsequently, to riding and recording with delivery workers to document their work performance in situ.In this paper, we draw upon four types of qualitative research material: (1) six semistructured telephonic interviews, (2) 11 audio-video recordings with six diferent food delivery workers performing deliveries in situ, (3) 11 in-depth interviews conducted during the ride-along sessions post-delivery, and (4) ethnographic feld notes capturing observations and interactions with 20+ food delivery workers from four diferent locations in Pune.
All participants considered in the analysis worked with Zomato (directly or via a third-party last-mile delivery service named Shadowfax), the largest online food delivery and restaurant aggregator platform in India [43].All participants who contributed to the study were male, despite the frst author's eforts to locate participants of diferent genders.During the feldwork, the frst author asked the male participants if they knew of any female food delivery workers in the city, before or during the pandemic.He also inquired with friends and colleagues in the town whether they, as customers, had encountered any female delivery workers.Finally, the frst author performed desk research to locate any online references to female food delivery workers in the city.Despite these eforts, he did not locate any women who would have worked in food delivery in Pune during the pandemic or before it.Our study, then, echoes the masculine shape taken by gig work platforms [29].Despite the low entry barrier work ofered by food delivery platforms, our observations from India show no female participation in this type of gig work.

Data collection
Respecting the health and economic hazards imposed on the workers during the pandemic, we evolved our methods in a situated manner, to understand both the work and the workers.A rule we established for our project, and that the frst author followed closely during recruitment and throughout subsequent engagements, was to not disturb, interrupt, or delay (potential) participants' work.While recruiting participants as well as during any informal, ad hoc conversations during feld visits, the frst author introduced himself as an academic researcher, mentioning his university afliation and explaining the purpose of the proposed interactions: to collect food delivery workers' frst-hand experiences for an academic study of how their work was performed and how the work practice was afected during the pandemic.The frst author explicitly mentioned that he has no association with any food delivery platform companies, and that any information collected would only be used for academic research.He also highlighted that the worker could opt out of the conversation at any point in time, that they need not answer every question, and that they should only answer to the extent they were comfortable with.If and when the frst author needed to record the conversation, he asked the participants for permission to do so, explaining that the recording would not be shared publicly but rather only used to document the conversation for analysis.He also explained that the data would be shared with other researchers involved in the study and its use would strive to preserve participants' anonymity, that is, the participant's name, voice, or any visuals disclosing their identity would not be shared publicly.This approach was followed throughout the feldwork to ensure informed consent from the participants verbally, even as the frst author often interacted with delivery workers ad hoc, without prior arrangements, in line with when they happened to be available to participate.
The frst data collection phase started during a strict lockdown.Unable to meet or observe workers in person, the frst author conducted six semi-structured interviews over the phone to understand how riders were performing with the constrained work conditions.The frst participant was recruited via an acquaintance, followed by snowball sampling.Some workers were asked whether they would like to participate when they came to deliver food to the frst author.After receiving participants' verbal consent to participate in the study, the frst author called them at their convenience, to their mobile number or via an audio call over WhatsApp.The interviews, conducted in Hindi and Marathi (the regional language of Maharashtra), were loosely structured and open-ended, so as to allow participants to share their understanding of the situation on the ground.At the beginning of the call, verbal consent was ensured for audio recording the conversion and participants were told they could stop the conversation at any point, without needing to give an explanation, and without any repercussions.The interviews, ranging from 35 minutes to an hour, were transcribed and translated into English by the frst author.
As the Indian government initiated an "unlock" phase [31], the frst author started feld visits at four locations that served as 'waiting spots' where many food delivery workers waited alone or in groups for new orders or for restaurants' order preparation.These were public spaces, mainly located near popular restaurants.The feld visits, conducted between January to April 2021, entailed observing workers around popular delivery hours: breakfast, lunch, and dinner.The frst author maintained a feld diary documenting observations, thick descriptions, pictures, and conversations.
Finally, the frst author recruited workers to record their delivery rides with a chin-mounted action camera on their helmets.The camera recordings aimed to capture the details of food delivery work in situ.When the frst author proposed the idea of a ride-along session to a group of workers, they were keen to record their work and even urged the frst author to share it online to voice their issues.However, in line with our data management protocol, the frst author explained that we would not share the video recordings on social media platforms and would use them solely for research purposes.Altogether eleven rides were recorded with six riders, six during the lunch slot and fve during the dinner slot.The frst author accompanied the rider on the pillion seat during the recordings on ten of the eleven rides.This allowed for a detailed conversation between the frst author and the riders, explaining their performance.One participant chose to ride alone.In this instance, the frst author followed the rider on his own bike and spoke with the rider about his experience once the order was completed.The conversations were in Hindi and Marathi, selectively transcribed and translated into English during the analysis.The frst author tried to keep his distance at the pick-up and drop locations, intending not to interrupt or infuence the workers and their interactions with various actors or the platform application on their phones.The frst author interacted with the worker at the pick-up spots only if the worker had to wait longer while the food was prepared or packed.The frst author explained the purpose of recording a delivery ride before asking whether the participant would be willing to wear an action camera on the helmet: the intention was only to document food delivery work, as it happens in practice, from the worker's perspective.He also emphasized that the aim of the recordings was not to capture every moment of the worker's interactions with other actors or devices.The frst author also explained that participants were encouraged to retain their focus on their work and that they need not worry about the orientation of the camera in order to record any particular details.Participants were informed that they could take of the camera or stop the recording at any time, and that they did not need to record the entire ride.The frst author demonstrated how to turn the camera on and of, and answered any other queries participants had regarding the camera before the ride.This protocol was followed at the beginning of every ride-along session to ensure participant's informed consent verbally, reiterating that the collected video would be used solely for academic research purposes, maintaining participants' anonymity.While recording the rides with a chin-mounted action camera, the incidental appearance of bystanders or customers was inevitable as the rider was in control of the camera's orientation and the recording.While the frst author was not in a position to solicit the consent of those who were captured incidentally [7], we anonymize any personal information of bystanders that was captured in the footage to protect their integrity.
Participants were ofered cash compensation for their contribution: INR 100 for each telephonic interview and INR 200 for every ride-along session.The amounts paid were to outmatch their average hourly income during the pandemic.While we made signifcant eforts to compensate the drivers, some refused or asked for the money to be donated to charity.Compensation for participants is not a straightforward matter, as it can transform the relationship, particularly in long-term ethnographic research, from one based around help and assistance into an, at least partially, monetary relation.Our challenge, then, was to compensate participants fairly while also respecting their right to refuse compensation.For any participants who we were unable to compensate we, instead, donated the money to a local charity.

Analysis
Aligning with the broader research project, during the initial rounds of thematic analysis [11], the frst author considered the varied ethnographic research materials together to produce a range of descriptive codes.For the analysis presented in this paper, the frst author revisited this preliminary coding to identify material that highlights the COVID-specifc changes introduced by the platform.Through discussions among the authors, this early round of analysis led to "quantifcation of the body" and "selective representation of body" as two themes for deeper exploration.The frst author, then, revisited the materials beyond the COVID-specifc data points, guided by these two themes, uncovering further data about the platform's reliance on the workers' GPS location and how this was refected in the workers' practices.The resulting materials were analyzed collaboratively with a focus on how the platform represents the workers and how this relates to their situated eforts to complete deliveries.As the representations of a worker as 'a dot on the map' and, alternatively, 'a superhero with a cape' were central to our analysis process, we also use them to organize the presentation of our fndings in the following section.The frst representation captures issues related to working with a location-based service, while the second speaks to the platform's eforts to portray delivery workers as safe and sterile, along with the asymmetric changes this led to.A summarised thematic analysis map in Appendix 1 provides further details on our coding process both specifcally for this paper and within our broader project.In presenting the fndings, we use pseudonyms to refer to the participants.We have obscured participants' faces and other identifying features from the visuals included, opting for representative illustrations where necessary to protect participants' and bystanders' integrity.

Refection on positionality and research ethics
The frst author, who was in charge of the feldwork, is currently based in Europe, yet born and raised in Maharashtra, India, with Marathi as his mother tongue.He spent signifcant parts of the COVID-19 pandemic back in Pune, so he was experiencing the situation there frst hand and directed the unfolding of the study in line with the situational awareness this allowed.To the participants, then, he appeared as a local, middle-class male, studying in a European higher education institute.Prior experience conducting contextual inquiries in rural India as well as acquaintance with the local language and social arrangements helped the frst author to build rapport with the participants and gain their trust.To reciprocate for participants' contribution beyond the monetary incentives that were ofered, the frst author chose to remain available for them also outside of the feldwork, assisting them where possible when they reached out with queries, for instance regarding employment and education.The frst author did not provide or propose any such assistance to the participants while conducting the study, so the provided assistance was strictly in response to participants' own initiative.The other three authors, who are from Indian and European backgrounds and currently based in Europe and the US, collaborated with the frst author on the analysis, drawing upon their expertise on the study of gig work and work practices, yet with sensitivity to the frst author's position as the expert on the feld site.
The feldwork was devised and conducted with consideration and empathy toward the devastation the COVID-19 pandemic had brought to society.The pandemic impacted severely the informal labor sector to which our participants belong.We worked to recognize the risk these workers were taking, respect their eforts, and learn about their work without causing further disruption.The frst author closely observed and followed all pandemic-specifc rules imposed by the state and the central government.Our study does not clearly fall into any of the areas of application specifed in the Ethical Review Act in the country where the frst, third, and fourth authors are afliated, and their national ethics review board only considers research conducted within the country.The second author, afliated with an academic institute in the US, joined the team during the argument and paper writing process and does not retain any data.As per university mandates, since her contribution was primarily intellectual and had no involvement during data collection, she could not obtain an IRB approval from her institute.The data is stored securely, in line with the frst author's university's guidelines, and we plan to dispose of it as per the guidelines.While we did not have access to a local ethics board that could review and approve our study, we have striven to conduct the research ethically.To guide ongoing conversation among the author team, we have attended to the ethical standards for ICTD/ICT4D research [15] which outline ethical research practices for research projects situated in diverse and complex contexts like ours.Further, we have sought to frame contextually situated ethical responses, instead of going into the feld with a static and formalized set of rules, resonating with the 'In-Action Ethics' framework by Frauenberger et al. [17].As the frst author had previous experience of performing feldwork in India and of working in Indian academia, he also drew on his experience and social network within the local academic community to discuss the methods and research process.This was done to ensure an ethnographic sensitivity where the local perspective and cultural reality was taken into account.
We wish to highlight ethical limitations in the feldwork and the subsequent analysis process.Even after attending to ethical research conduct as outlined above, the gap between the researcher and the researched remains, where the research is not directly accessible or benefcial to the researched.The experiences and interactions with the riders during the pandemic pushed us to refect on the limits in our procedures in ensuring consent and managing data: Notably, the riders were keen to make their issues visible to larger audiences on social media platforms but our commitment to preserving their anonymity, in line with established ethical research practice, did not allow us to facilitate it.Further, we intended to show and discuss the ride recordings with the riders with whom they were recorded.However, it proved challenging to even locate the same riders again, and we were also hesitant to ask them to invest further time in reviewing data with us.Similarly, inviting participants to evaluate and comment on our analysis would have further strengthened our understanding of the studied gig work and the platform.While we have not managed to do so, we acknowledge this type of engagement as best practice whenever it is possible without imposing strain on the participants.

FINDINGS
We organize our fndings into two sub-sections: First, tracing how the platform relied on a 'quantifable minimum' representation of delivery workers, we outline how the platform uses workers' (phone's) GPS location to monitor and control their movement.This shapes the delivery workfow at three crucial stages: order allocation, pick-up, and delivery.Second, we illustrate how the platform began to acknowledge workers' bodies selectively during the COVID-19 pandemic, moving from the solely location-based representation to highlighting specifc physiological aspects of delivery workers' bodies.We interpret this as an efort to portray workers' bodies as sterile and safe, in order to retain and regain customer trust and to position the platform as an essential service.

A dot on the map: Working with a location-based service
The worker's GPS location works as a crucial point of reference in how the platform sets up and controls the delivery workfow.Throughout the delivery workfow, from receiving an order to picking it up, and fnally completing the delivery, the location information the platform provides and the location-based rules it imposes upon the workers add to the challenges of getting the work done.
4.1.1Order allocation:Spatial tactics and collaborative sense-making.
The platform uses the rider's GPS location during order allocation to estimate and plot the next steps in the delivery workfow.When the platform ofers an order to a rider, it shows information like the travel distance and estimated order completion time.Once the rider accepts the order, the platform informs the customer of an estimated delivery time.
The logic of order allocation remains a point of speculation among workers since they are not party to exactly how the orderworker matching is performed.Among the workers, though, their distance from the prospective pick-up spot was broadly considered the key factor in who gets the next order.Workers believed that "(they) will receive an order only if they are near the pick-up location." With this belief, riders tried to mark their location as closest to the nearby popular restaurant, shown as a pin on the map.Ratan explained his understanding of how the pin location worked (See also Figure 1): "... one who used to match his location with the pin location (the restaurant location) used to get the order.You would not get the order if you were even a little away from the pin.(So) guys used to stack their mobiles at a spot matching the (restaurant's) location!"At the time of the feldwork, the platform switched to a heat-map that indicated an area from where, according to the platform, riders were most likely to get an order.The heat-map ofered riders a wider territory to cover, supposedly making them more dispersed rather than encouraging them to gather at one specifc spot.However, we found that riders largely ignored the heat-map and, instead, continued to rely on their own heuristics of the area and the possibility of getting orders.At a waiting spot, a group of workers explained how the heatmap works and expressed their doubts: "According to them (Zomato), the heatmap highlights the place with more orders.We do not know if it really does have (more orders) or not." So, instead, the workers chose to rely on their experience and local knowledge: "But we know where the orders are... at what time, which restaurants (names a few restaurants) get orders... " During one of the rides, Rakesh ignored the heatmap suggestion as we were passing by the suggested area, opting to return to his regular waiting spot.Aiming to appear available near popular restaurants, riders gathered around spots from where they believed they would get an order.As documented by Shaikh et al. [49], these 'waiting spots' became more prominent during the pandemic when there were fewer orders, and riders spent more time waiting.
While riders at a waiting spot might have appeared to the platform as a cluster of active GPS locations, our feld visits revealed that riders shared and collected information about each other.Riders were aware of who else was at the spot.They often socialized to inquire and discuss each other's earnings, along with the platform's rules and guidelines.The accumulated knowledge was later used strategically as riders competed for order allocation.Many riders kept moving between waiting spots in line with what they believed would increase the likelihood of receiving an order.For example, they often decided to leave a spot if they suspected that most co-located riders were paid less per delivery and were, thus, more likely to get the next order [49,51].Others, like Vishu, wanted to ensure orders from nearby locations and believed that staying in the same location helped: "I stay in the same area to get orders within the radius of 3-4 KMs from where I wait...only from the nearby hotels." Moving or static, as long as riders are logged in for work, they remain available to the platform as GPS data points which can be used for algorithmic calculations.Workers, however, are motivated by their income goals and strive to enhance their order allocation.This leads them to make choices about where to wait and when to move, in line with their best guess of how the platform's order allocation factors in their location.

Pick-up:
Trouble with location-based rules and erroneous addresses.After accepting an order, the food delivery worker has to drive to the pick-up location and mark 'reached' on their mobile app, informing the restaurant about their arrival and readiness to pick up the order.To ensure that the worker can make this announcement only after physically reaching the pick-up location, the platform allows riders to announce their arrival only within a radius of 500 meters from the pick-up location.The platform relies on the rider's mobile GPS location to confrm that the rider is within this defned boundary.Rakesh, a part-time delivery worker, demonstrated how the platform monitors his location by trying to mark as 'reached' when still a few kilometers away from the pick-up location.The platform refused his submission with a pop-up notifcation: "You are not near the restaurant".A green button directed him to "Go to restaurant location." In another instance, Vinod, a full-time delivery worker for the last four years, hit the 'reached' button instantly after accepting the order at the waiting spot.He explained: "(I) Can mark 'reached' if within the range of 500 meters (from the pick-up location).See, I can mark it as reached from here as the hotel is nearby." While the platform relies on strict and simple rules to manage the work, on the ground, delivery work is often more complicated.During the early phase of the lockdowns, restaurants were not allowed to serve in-house customers.In response, some small restaurants shifted their kitchens to owners' homes to save on operating costs, yet without updating their address information on the platform.The platform kept imposing the pick-up radius on riders based on their and the restaurant's location, learning about the ground reality only after a delivery worker reached the pick-up location and found out about the situation.Vishu explained how he had to work extra because of this: "After reaching the given address, I had to call customer care, informing them that I had reached this hotel, but it was shut.Then they (the platform) used to call the hotel to let them know that I had reached the mentioned location; then after, the hotel would share the new address!And I had to go there." As Vishu's live location was also shared with the customer, they would see him moving away from the purported pick-up location.By then, if the platform had not informed the customer about the problem, they might report Vishu's seemingly abrupt movements.Vishu mentioned that customers sometimes cancelled the order after learning about the issue, and he ended up earning nothing despite putting in extra efort.Imposing location-based rules and showing workers' location to customers caused trouble when restaurants' location information was unreliable.This also made it harder for workers to repair the situation.

Delivery:
The cost of repair.In the last part of the delivery workfow, the rider has to reach the given delivery location and mark the delivery as completed, after handing the parcel to the customer.Crucially, the delivery needs to be marked as done while still at the customer location.Here again, the strict reliance on location data to manage the worker becomes problematic.
During a ride-along session, we witnessed Rakesh struggling to locate the customer after he had reached the address provided by the platform.He, then, found out that the customer lived a short distance away from the given address, in a newly constructed building.This new construction was an extension of the residential complex whose address had been given to Rakesh.The building was not yet marked on the digital map, so the platform was unaware of its existence.Rakesh kept waiting at the given address for a few minutes and then called the customer several times to fgure out that he was, in fact, not at the intended delivery location.Even then, he was initially reluctant to move away from the address provided by the platform: "We can not go further away from the given address.If I move a few meters away, they (the platform) will know and call me.It is not allowed." Eventually Rakesh was able to complete the delivery by locating and reaching the customer's actual location with the help of bystanders who knew the area.Security guards at the address where the app had pointed him provided initial instructions towards the correct address.When Rakesh reached a spot in a dark alley that appeared like a dead-end, a shopkeeper pointed him towards the desired address (which was not represented on the map).At the end of the trip, we saw the platform rating the trip as 'not great' since it had taken more than the estimated time (due to the time Rakesh spent fguring out the correct delivery location after reaching, ahead of time, the location he was sent to).
In another incidence, Mahi discussed with a colleague who had returned to the waiting spot after delivering an order.The colleague had forgot to mark the order as delivered at the delivery location, which was around 3-5 Km away from the waiting spot where he now wanted to complete the task.The platform was still showing the delivery location actively pinned on the map.Since the worker's current location did not match the pinned location, the platform was not allowing him to mark the order as completed.He asked Mahi to help him reset the pinned location to his current location.Mahi explained that he had to go back to the delivery location to mark the order as delivered: "They will block your id if you try to re-set the pinned location.It means you are adding fake kilometers to the ride.Go back and complete the trip." His colleague kept tinkering with the app for a while, then gave up and rode back.
The platform's strict reliance on riders' GPS location to manage and assess their performance made it hard for workers to recover from errors, be it erroneous data about restaurant or customer locations, or a rider's own mistake, such as Mahi's colleague's failure to mark the delivery as complete in the right place.Workers, to the degree possible to them, made eforts to repair and recover from such errors, bearing the cost even when the mistake was not theirs.

A superhero with a cape: Portraying a sterile, safe delivery worker
We now turn to describe how the platform, moving beyond workers' GPS location, started to selectively acknowledge certain physiological aspects of their bodies during the pandemic.We argue that, at the onset of the pandemic, the platform recognized the rider's body as a 'threatening body' [34], a potential conduit of the virus.
In response to the new circumstances, the platform made three major changes in the delivery worker's app and related tweaks in the customer app.These changes were measuring riders' body temperature, a masking mandate for riders, and the option for customers to choose a contactless delivery.All three were promoted as precautionary measures taken by the platform, and they seemed aimed at retaining customers' trust in online food orders [52], by portraying a sterile, safe delivery worker.Perhaps the most striking visual of these optics of safety appeared on the customer app, where the rider was illustrated as a superhero with a cape (see the worker icon on Figure 5), framing them as an essential worker who is risking their own health to deliver food during the pandemic.

Body
Temperature: Threatening bodies and bodies at risk.Since the beginning of the pandemic, a rise in body temperature, from what is otherwise considered to be normal, was widely accepted as a key symptom of a COVID-19 infection.While a rise in body temperature alone is not enough to conclude an infection, during the early days of the pandemic, trust in such readings was instilled in the public discourse to motivate diverse, self-disciplined, precautionary measures.For example, we observed some restaurants proactively recording and showcasing staf's body temperature readings on the food parcels (see Figure 3).Cashing on and contributing to this discourse, the food delivery platform devised a way to measure delivery workers' body temperature and share it with customers.The platform asked partner restaurants to record the workers' body temperature when they arrived to pick up an order.This reading was, then, submitted to the platform and displayed on the customer app to inform customers that the person carrying their food had a 'normal' body temperature.Riders' bodies, thus, not only became visible as possible carriers of infection, but information about them was made available to illustrate a concern for health.
As a part of the change, the workfow on the worker's app was modifed to ask the worker to confrm whether the restaurant had recorded their body temperature (see Figure 4).Many restaurants invested in contactless thermal readers to monitor the body temperature of delivery workers entering their premises.Vishu remembered the early days of these arrangements: "They refused to hand over the order if the temperature is above 38°C.Before that, they used to ask us to sanitize our hands." Figure 4: Worker is asked to confrm whether the restaurant had recorded their body temperature and helped with hand sanitation.
Our video data, captured in an 'unlock' phase during the pandemic, does not show any restaurant recording body temperatures.However, we can see the delivery worker's app asking them to confrm whether the restaurant measured their body temperature and provided hand sanitizer (see Figure 4).In response, most workers clicked 'yes' even when no such check was performed.Anam explained the reason for this false confrmation: "They [the restaurants] do not check [the temperature].But they might inform the platform that they did!In that case, the platform might block my ID." Once again, we see workers navigating opacity: The platform's exact arrangement with the restaurant to acquire workers' body temperature remained invisible and unclear to the workers.They did not know whether a restaurant shared their body temperature with the platform or if the platform confrms workers' answers with the restaurant.Without revealing the arrangements regarding rider's temperature data to the riders themselves, the platform made temperature readings visible to the customers in their app on multiple occasions, in order-status notifcations, on the map along with the rider's live location, and as a part of the order summary.Similarly, the customer app communicated that the rider had sanitized their hands (see Figure 5).It is worth noting that while the platform came up with means to track each rider's body temperature -or at least to appear to do so -and share it prominently with customers, it made no efort to measure the temperatures of any other actors, such as restaurant staf or customers, and share that data with the riders.The rider's body gets acknowledged solely as a threatening body [34], rather than also as a body at risk [52].

Masked
Selfie: Asymmetric measures and missing communication channels.During the pandemic, the platform started to monitor that workers wore a mask while delivering food, and to promote this to customers as another precautionary measure.Many mentioned that a masked selfe was mandatory to log in for work (see Figure 6).As Rakesh explained: "It is mandatory to take a selfe with the mask to log in.(This is) to make sure we are taking safety measures, else have to pay a 100 rupees fne." Workers either had to pay a fne for not submitting a masked selfe, or risk getting their account blocked temporarily or for a longer duration.This demand for a masked selfe was an expansion of the requirement to submit a picture with the company's delivery bag, wearing the company T-shirt, that pre-dated the pandemic.Rakesh recalled: "I had to submit one (picture) with their bag as and when they asked.If I could not, then they put a fne of 500 rupees.I once had to pay.That is why I carry this bag all the time." During a ride-along session, we witnessed Shiva adhering to the platform's injunction for a masked selfe, making eforts to submit the selfe showing both the mask and the company T-shirt.Eventually, though, he received a warning for not wearing a mask while delivering food, although he wore one.Shiva had no channel to explain the situation or challenge the platform's assessment.With a small addition to the customer feedback form, the platform asked customers to confrm whether the delivery worker was wearing a mask or not (see Figure 7).As the workers seldom used the customer application, many participants were not aware of this verifcation, and the ones who were, still did not know how the platform used this customer feedback to assess their performance.When asked if any such feature was made available on the delivery worker's app to report if the customer had worn a mask or not, Vishu explained: "No, there was no such a condition (of wearing a mask) for the customer...They said we could also ask the customer to wear a mask, but we never did.Because not every customer is considerate enough." The platform's assessment of a masked selfe was fnal and assertive, with no channel for the worker to challenge it.While the riders were monitored with the help of image recognition technology and customer input, the platform made no efort to impose a mask mandate on customers -refecting the same asymmetric approach to health risk mitigation as seen in the previous section.

Contactless Delivery:
Exposure to risk of infection and economic loss.Aligning with social distancing measures imposed by the government during the pandemic [39], with a minor tweak in the customer app, the platform introduced a new feature allowing customers to opt for a 'contactless delivery'.If the customer chose this option, the delivery worker was requested to leave the order at the customer's doorstep, instead of handing it over in person.During the early days of the pandemic, almost every gated community and residential apartment limited visitor entry, directing delivery workers to leave the parcel at the entry gate with the security guards.The customer had to collect it from there.Such arrangements, along with the 'Contactless Delivery' feature, may seem to reduce the risk to delivery workers and make their work more convenient.Yet, while more often able to keep a safe distance from the customers, delivery workers now had to interact with the security guards to confrm the delivery location.What is more, they often had to call and wait for the customer to come to the entry gate to collect the parcel.The asymmetry of the arrangement was quite striking: It was optional for the customer to choose a contactless delivery, whereas the workers had no option but to comply with customer preferences.Ramesh believed that concerned and considerate customers would opt for contactless delivery: "If he (customer) is truly concerned and did not want to infect anyone else, he would select the 'Contactless Delivery' option, even if it is allowed in his gated community to enter." The platform placed no restrictions on delivery locations.Ramesh had once needed to make a contactless delivery in a COVID quarantine center, established by the state government to treat COVID patients in isolation.Ramesh explained: "Patients were not allowed to carry mobile phones in there, so someone from the outside used to order on their behalf giving quarantine centre as the delivery address...They had a check-post, followed by lab and waiting area.They kept some tables in the waiting area to keep packed food.That was the delivery spot.I had to wait until the customer or someone comes to pick the order.I cannot move!"Workers could not suggest or demand a contactless delivery.Raj described how a customer had insisted on doorstep delivery when no visitors were allowed to enter the premise.The local governing authority had marked the premise as a quarantined zone after confrming COVID cases.With a 'no visitor' notice on the main gate and security guards not allowing Raj to pass through, he had called the customer: "I told the customer that I am not allowed to enter.But she was insisting me to come upstairs for the delivery...the customer told me that she had spoken with the security guards and they should let me enter.Raj's explanation illustrates how customers could exercise their agency to reach out to the platform and insist on receiving the service they wanted: "...She threatened me that she will fle a complaint against me with Zomato for not delivering the order (at doorstep)...when I made her talk with the security guards, they told her that they are not allowing anyone to enter but she didn't agree and continued insisting for doorstep delivery." Raj eventually left the order with the security guards.Later, he came to know that the customer had fled a complaint for torn-of packaging and a leaking container, for which the platform charged him for the amount of the order.The delivery workers' situated eforts and their social coordination with other involved actors, like security guards at the entry gate, made the contactless delivery feature work, amidst the asymmetric arrangement that prioritised customer preferences rather than functioning to maintain social distancing across the board.At times, this pushed riders into situations that increased their exposure to either the risk of infection or economic loss.

DISCUSSION
Our riders strived to make the food delivery arrangement work, despite location-based rules that restrict their action, the upheaval of the pandemic that made address information less reliable, and asymmetric health measures that, while operating on the riders, were not devised to care for them.We now discuss what could be learned from attending to workers as not only 'essential' but 'infrastructural', and what it might look like to center workers' well-being and bodily needs in design.

Food delivery workers as infrastructure
Customers saw only the most schematic of visualisations of delivery riders' work.This reduction of the work in the customer's interface meant that, for customers, ordering food was a simple operationjust clicking on the phone's interface and paying the delivery fee.One way of understanding this reduction is to draw on the idea of infrastructure [53].This concept has been productive for both HCI and CSCW, capturing some of the complex relationships around technology and its use.With an infrastructure, the complexity of a service or activity is 'black boxed', that is, reduced so, as Leigh Star put it, "the cook considers the water system as working infrastructure integral to making dinner" [54, p.133], or, in our case, customers expect to have their lunch delivered at the push of a button.
Much like the complexities involved in ensuring the availability of water and electricity -utilities that get used in many parts of the world without much thought to where they come from and howthe continuous efort, skill, and repair needed to complete deliveries remains out of sight as long as deliveries arrive on time.For the platform to keep providing food delivery as a service to its customers, it does not need to visualise what the workers are experiencing on the ground.As Star and Ruhleder [54, p.133] write, however, for us as analysts we need to see it as "a relational property, not as a thing stripped of use".This paper has documented something of how this 'infrastructure' is made possible -how work is represented in a simplifed way, yet how it exists as a series of relationships between restaurants and workers, workers and customers, workers and the platform, and so on (See Figure 1).Each of the constitutive relationships builds up to create a complex service which is, then, inverted into the simple form shown to the customer.
While most of the time the delivery services successfully produces an infrastructure for customers' convenience, infrastructure also breaks down on occasion -and this is what happened with the restrictions to prevent the spread of COVID.The complexity of food delivery surfaced as the relational arrangement itself broke down in a way that workers could not repair on their own.Considering workers as infrastructure lets us see how their contribution can remain invisible until, upon breakdown, "'the normally invisible quality of working infrastructure becomes visible" (ibid).Clearly, the COVID-19 pandemic is a drastic example of this.While gig work platforms deliberately ignored many aspects of workers' bodies with the onset of the pandemic, their infrastructure needed to be repaired with some recognition, so that the platform could appear to provide a safe and reliable -even essential -service.
The measures taken by the platform to portray safe delivery workers can be seen as eforts to repair a broken infrastructure.After the outbreak of COVID, the platform had a new focus on the bodies of the workers, not simply as dots on a map, but through a quantitative measurement of body temperature, selfes to enforce a mask mandate, and the option given to customers to maintain distance to the workers.This focus had a number of implications for the workers.They had to work in hot conditions, which at times overheated their bodies, yet needed to regularly produce a 'normal' body temperature, which would be measured, alongside a masked face, which would be photographed and verifed with image recognition software.The breakdown of the infrastructure revealed how the arrangement had worked with its representation of workers as dots on a map.It was, then, put back together in another way, with workers represented on the customer app as safe, temperature controlled cartoon superheroes.
Whatever these eforts at the platform level -and the attempt at valorisation of the workers in the interface -workers' labor to repair the arrangement went unvisualised.Road blocks, erroneous address information, or security guards could prevent the timely delivery of food.Despite workers' strenuous eforts to deliver food in a timely manner, any failure to do so resulted in penalisation of the worker.So, while in one way the infrastructure was repaired, and even began to acknowledge those delivering food as heroes, the worker behind the representation was a 'hero on a reverse timer', penalised if they took too long, in a city disrupted by a pandemic.
In their book on 'ghost work', Gray and Suri [18, p.122] highlight how, in the absence of platform-provided tools, crowdworkers need to create the collaborative infrastructure they need for themselves: "The need for social ties associated with traditional nine-to-fve employment persists, even in the absence of a physical workplace.Because there is no infrastructure for workers, they create it for themselves.Workers collaborate to overcome social and technical problems stemming from the platforms they work on.More specifcally, they collaborate to reduce the transaction costs imposed on them by the platforms to get the work done and provide social support to one another." Prior research and the analysis we have presented illustrate the same for our food delivery workers: they, too, resort to one another and the 'kindness of strangers' to get the work done.Our argument here goes one step further: Not only do workers create the infrastructure they need to do their work, but what they do on the ground is a crucial part of what holds up the whole food delivery arrangement.The infrastructure breaks apart without the work that is rendered invisible to the customers.

Centering workers' well-being and bodily needs in technology design
Our fndings speak to how the platform kept working in a time of crisis, but at a cost to the workers who were exposed to both the risk of infection and economic risk, as they delivered food in a city with restaurants that had, unannounced, started operating from home, and customers who were not necessarily concerned with the workers' well-being.While there is a humane management and labour aspect of this that is important, there is also a set of questions around how platforms could have supported the workers themselves better, through both policy and technology.Our discussion, here, joins prior work that has highlighted power asymmetries in the gig economy [32,48,59], and called for workercentered design [19,38].Distinctively, our particular focus is on the corporeal aspects of gig work.
As location-based services, gig platforms rely on plotting their workers on digitally mapped geography to structure and manage the work.What gets cleaned out of the picture with simple representations of workers as dots and heroes, is their sociability with co-present others, their tactics in working with location-based platform logics, and their situated, bodily eforts that are needed to get the work done.We may contrast this overlooking and downplaying of workers' contributions with contemporary eforts to bring 'humanness' to AI technologies by focusing on three essential elements: embodiment, emotion, and sociability [56].Our fndings illustrate how workers' situated eforts enable the functioning of food delivery platforms.These eforts scale beyond the physical labor of bringing parcels from one location to another: Workers endure bodily pain through long motorbike rides.Their emotional involvement is needed in negotiating with restaurant staf, security guards, and customers.They have to compromise and reschedule meals to deliver food on time.Overall, they need to respond to unexpected circumstances, often by resorting to the kindness of strangers to repair 'broken' situations.Platform technology is also devised to ignore -even erase -the variation of bodies at work.It disregards gender and age, and workers' bodily abilities and disabilities are overlooked.Whoever logs in for work gets rendered as the same kind of icon on the map.Only aspects of the worker that can be tracked and timestamped are operated upon in allocating work and monitoring its completion.Workers' corporeal actuality gets reduced to the bits of information essential to steer the labor, and, in the process, information loses its body [22].
Interpreting gig workers as infrastructure highlights their crucial contribution and guides us to explore what it might mean to center workers' bodies in designing platforms and the ecologies of technologies that surround them.While the pandemic prompted the food delivery platform to acknowledge workers' bodies, our fndings highlight how the rider's body got attended to solely as a threatening body [34], rather than also as a body at risk [52].Riders' bodies and practices involving their bodily presence were operated upon -with the monitoring of body temperature, the mandate to wear a mask while working, and with the contactless delivery feature -but this was not done to prioritize their well-being.While the emphasis on customers' preferences and comfort is perhaps not surprising, the asymmetries of these health measures are stark and make them, from a public health standpoint, a sub-optimal approach to prevent the spread of infection.Next to increased surveillance of the workers, this meant exposure to the risk of infection not only for workers but also for those they lived with.An obvious design opportunity, then, would be to imagine what it would look like to redesign food delivery platforms in a way that prioritizes workers' health and well-being.In this vein, we need to attend to workers' corporeal variation: Who does the work?Why them and not others?What kinds of bodies are centered, and in what capacity?How might we support these capacities during the labour of food delivery and other geographically tethered gig work?For example, we might ask what it would look like to design for fatigued food delivery workers, in between orders, wanting to take a break without impacting their income negatively?Or, what would it take to support gig workers' eforts to attend to their bodily needs, from eating to peeing?We might even ask ourselves why it can be so hard to imagine a set-up where a female food delivery worker could fnd the time and space during logged-in hours to breastfeed a child or pump milk, without missing out on order allocation?After all, being able to take care of such needs is not that unimaginable when it comes to many other types of workers and work settings.
Considering the constraints on this kind of design, we need to acknowledge that existing platforms are, at present, not incentivized to redesign their workfows and business models with particular care for workers and their bodily needs.Here, the limit is not so much in the capacity to re-imagine technologies as it is in the capacity to make policy change -and we can draw on the work of scholars like Schwartz and Weber [48] to consider what engagement in public policy making could look like.When it comes to design work, the more promising opportunities may lie in two diferent directions: First, we may focus on alternative platforms [13] that work with diferent starting points and, likely, diferent organizational models (for example as cooperatives [33].Second, we can make eforts to build up successor systems [26,45] that co-exist with current platforms yet carve out room to critique them, and to generate knowledge and activity that is motivated from the perspective of the workers and attentive to their needs.

CONCLUSION
With their situated presence on the ground and their capacity to weave together the complex relationships between the platform, restaurants, customers, and co-located others encountered in the urban landscape, food delivery riders are not only essential but infrastructural in making the food delivery arrangement function.Like working infrastructure, their eforts remain hidden and become visible only upon breakdown, when the need for repair and rearrangement arises.Our key contribution is empirical: Drawing upon an ethnographic study of a food delivery platform in Pune, India, during the COVID-19 pandemic, our fndings illustrate (1) the challenges gig workers face when working with a location-based platform that uses their (phone's) GPS location to monitor and control their movement, and how these shape the delivery workfow, as well as (2) workers' tactics in responding to issues arising from the asymmetric platform policies introduced during the pandemic.Our study highlights food delivery workers as a critical yet often hidden layer of the platform economy: They not only contribute to the functioning of the business but also show up in critical situations, like the COVID-19 pandemic, as essential workers, often putting themselves at risk.In resonance with prior work, our study speaks to the power asymmetry among stakeholders, including the platform, workers, customers, and even the restaurants.Refecting upon the asymmetry of the health measures the platform implemented during the pandemic, we call for discussion of what it might look like to center workers' well-being and bodily needs in technology design, and what avenues such design work might take given that platforms are, at present, not incentivized to redesign their workfows and business models with particular care for workers and their bodily needs.

Figure 1 :
Figure 1: Food delivery workers' socially situated presence in waiting spots reduced to GPS-locations on the platform's map.

Figure 2 :
Figure 2: The platform asking worker to go near the restaurant to mark as 'reached'.

Figure 3 :
Figure 3: Restaurant recording and showcasing staf's body temperature on the food packaging.

Figure 5 :
Figure 5: The platform showing riders' body temperature on map and in a notifcation, portraying the rider as an illustrated superhero.

Figure 6 :
Figure 6: Worker is asked to take a selfe with a mask to go online.

Figure 7 :
Figure 7: Customer is asked to report on whether the valet (the rider) wore a mask.