Co-design Accessible Public Robots: Insights from People with Mobility Disability, Robotic Practitioners and Their Collaborations

2.1 Inaccessible Public Space and Mobility Technology

In the United States, regulations require that sidewalks be accessible to all individuals, including those using mobility devices such as wheelchairs [3]. Unfortunately, for people with disabilities (PWD), using sidewalks can present challenges that impact their quality of life [27, 43], independence [53], physical activity levels [53], and ability to participate fully in society [27]. These challenges are often due to barriers created by poor sidewalk design or construction, insufficient maintenance, or natural terrain features [27]. Studies on sidewalk accessibility have identified the challenges experienced by people with disabilities on a daily basis. Gharebaghi and Amin categorized these barriers into spatial (e.g., surface quality, slopes) and social (e.g., transport policy) factors [31]. Mehmet et al. identified various physical barriers, including uneven surfaces, obstructions, and insufficient width that impair sidewalk accessibility for PwMD [51].

Work by Froehlich and colleagues uses design and technology to address sidewalk accessibility challenges [29]. For instance, their team’s research has explored the potential of crowdsourcing platforms and deep learning techniques to map sidewalk conditions, allowing for the assessment of sidewalk navigability [61]. Teams such as Kasemsuppakorn and Karimi have used similar data [40] to develop wheelchair routing systems that can avoid obstacles and impassable sidewalks [41, 42]. While such technology is helping PwMD better navigate sidewalks with fixed issues, the introduction of robots to public spaces raises questions about how PwMD navigation strategies may be affected. Our research focuses on how PwMD will interact with robots while also managing other fixed sidewalk issues. Work exploring prior deployments of public robots may suggest some ways in which they will alter public spaces and people’s interaction with them on the sidewalk.

2.2 Robots in Public

Currently, there are various deployments of robots in public spaces, both in research contexts and as part of new services. Real-world deployments of robots in public include security robots that monitor parking lots and pedestrian malls¹ and dog-like robots to patrol public spaces and encourage social distancing during the COVID-19 pandemic [74] or support police operations for dirty, dangerous, and dull tasks [16]. The most widely deployed robots in public are delivery robots deployed by companies such as Starship, Kiwibot, Amazon, and FedEx. These robots often travel slowly along public sidewalks to deliver goods to people. In general, these robots do not often aim to interact with people, except when passing. However, prior work by HRI research teams suggests that even when robots have incidental interaction with non-users in public, they will nonetheless cause changes in how people behave, thus making it important for robot development teams to understand how the public might interact with their robots [60].

Prior work in public HRI studies suggests some ways that people will interact with different robots in public. Yang et al. [84] used Wizard-of-Oz field studies to develop socially acceptable movement and interactions for a robotic trashcan, finding that many people would actively ignore the robot if they did not want it to approach, would speak to the robot to summon it or request it leave, and would try to entice the robot with trash. A more recent deployment in New York City has replicated these results and shown that people are generally receptive to trash barrel robots when they are in public squares [15]. Work by Schneider et al. [64] testing a humanoid robot in a Japanese mall also found that roughly half of the people walking by ignored the robot unless the robot spoke to people and asked to not be ignored. Thunberg and Ziemke [72] found that many people in a European train station were not accustomed to interacting with a humanoid robot (Pepper by Aldebaran). [22] found that some people in Estonia helped delivery robots stuck in the snow, though they raise ethical issues about people providing free labor for delivery companies by helping the robots. [80] also observed people helping or ignoring stuck robots in a US city. These results suggest a split among people’s desire to interact with robots in public. Moreover, the works specifically observing delivery robots suggest that the robots have similar challenges on the sidewalk as PwMDs and that they too may become obstacles impeding walkways.

2.3 Public robot’s impacts on sidewalk accessibility

We now consider prior work that further explores the potential impacts robots may have on the accessibility of pedestrian walkways. Weinberg et al. [80] found that robots could cause distractions and obstructions with different sidewalk users, such as blocking part of a sidewalk when stuck or causing dogs to become perturbed, by barking or lunging at the robots, and causing a minor scene on the sidewalk. Gehrke and colleagues [30] found that delivery robots deployed on a college campus provided convenient last-mile delivery but also caused conflicts in walking paths and potentially unsafe travel conditions, especially for cyclists.

Speaking directly to people with disabilities, Bennett et al. [12] highlight the safety concerns and real-world incidents of people with motor-related disabilities when encountering delivery robots on sidewalks, such as blocking access to a curb cut and preventing a wheelchair user to get from the street to back to the sidewalk. Bhat and Zhao [13] explored the experiences of individuals with visual impairments who directly interact with different mobile service robots, describing these robots as “dangerous” and “unfamiliar moving obstacles” due to their lack of predictable behavior or clear intent. Due to such incidents and the potential harm that they may cause pedestrians and PwMDs, researchers such as Salvini et al. [62] have recommended government regulation of robot deployments in public areas to address the psychological and physical safety risks that arise when pedestrians interact with sidewalk robots in highly crowded street environments. Thomasen [71], however, argues that regulations must be carefully considered as different strategies for managing robots in public may help some people while encroaching on the rights of other’s access to public space.

Our research builds upon these prior works exploring the accessibility issues of public robots to understand the potential challenges that PwMDs believe could exist. Further, we also explore the perspectives of roboticists in considering accessibility during robot design. While regulation is one proposed way for maintaining accessibility, it will also need to be paired with accessible robot design strategies that roboticists can build upon, as described in the next section.

2.4 Accessible design strategies in Human-robot interaction

Both HRI and HCI communities have proposed design frameworks for robot designers to follow [11, 19]. However, there is less existing work on designing accessible service robots. We found only one proposal, evaluated by Qbilat [58], that specifically addressed accessibility in social robots by incorporating six other accessibility guidelines based on computer interface design (e.g. WCAG 2.0 [36], IBM accessibility [35]). However, accessibility recommendations based on screen-based interfaces may not provide useful guidance for a robot’s physical form and movement behavior. Prior HRI studies have shown various factors of robot behavior design that may affect people’s perception of safety and trust in robots. Previous experiments have shown that robot size, speed, and approaching behavior can alter people’s trust in robots in indoor environments [55]. Further work has explored proxemics and people’s comfort with robots at different distances and when approaching from different directions [63, 70, 78]. Overall, there are many factors robot designers must consider in making robot movements safe and accessible.

Due to the complexity of creating safe and accessible robot interaction and the current lack of clear guidelines and understanding of people’s responses to different robot designs, many teams approach designing robot behaviors through co-design workshops. Prior works on assistive robotics have found success in co-developing robots tailored to the needs of people with disabilities  [26, 45, 75]. Based on these prior successes, researchers exploring public robots are also exploring participatory design approaches. Early work by [8] engaged people with vision disabilities (both designers and non-experts) through interviews, group co-design sessions, and Wizard-of-oz testing sessions to design robots that provide navigation guidance. Tian et al. [73] used participatory prototyping methods that involved programming robot interactions with the support of an expert and viewing them in a simulation to imagine how a humanoid robot could interact with people in public spaces. Sumartojo et al. [68] used images of public spaces overlaid with different types of robots to elicit reactions from people about where such robots would or would not be appropriate.

Our research seeks to build upon the aforementioned participatory design research on public robots to better understand how the design choices involved in developing and deploying sidewalk robots might influence existing accessibility challenges. While Participatory Design methods have been shown to be fruitful in generating new robot concepts and surfacing important concerns from users, there is little work engaging PwMDs in discussing robots on public sidewalks. Our work builds on the methods of prior works by leveraging interviews and co-design sessions involving imagery and simulations [8, 73] to consider how robots on sidewalks should interact with PwMDs, described in Section3. Our study seeks to bring together insights from PwMD and roboticists on the design and presence of sidewalk robots. By learning from both groups and facilitating a collaborative design process between PwMDs and roboticists, we aim to inform accessible robot development in public space—whose intricacies haven’t been fully examined.

3.1 Interviews with People with Mobility Disabilities

In the interview with PwMD, we first asked our participants to share demographic information about themselves, such as age, gender, and their mobility impairment (should they wish to disclose). Then, we delved into the following topics:

3.2 Interviews with Robotic Practitioners

In the second set of interviews with practitioners, we centered our conversations around two primary questions: (1) How do robotic practitioners consider accessibility within their current practice, and what opportunities do they see for improvement? (2) Given the challenges highlighted by PwMD, how can public robots be made more accessible?

Initially, we delved into their perspectives on the state of accessible robots in their respective contexts, whether in companies or academic institutions. We sought to understand the obstacles they might face in ensuring accessibility for the robots they work on. We then explored potential avenues for change, posing questions like, "What values, methods, and research-based strategies are essential to incorporate accessibility features into public robots?" These questions were tailored to their specific roles (e.g., designer, researcher, or developer). Furthermore, we addressed the practical challenges outlined by PwMD in our initial interviews. For instance, we discussed issues like competition for sidewalk resources and invited suggestions on potential solutions.

3.3 Co-Design Workshop

Building on our findings from interviews, in the final stage of our research, we organized four co-design workshops pairing PwMD and robotic practitioners to explore how their early collaboration could inform accessible robot design. The procedure of the workshop was:

3.3.3 Activity 2 – Robot Communication & Interaction Under Different Scenarios:

After participants converged around a particular robot idea, they moved to the second activity, which involved picking a card with a scenario and developing the interaction and communication of the robot (refer to 3b). The scenarios contained situations from the first activity, as well as predetermined encounters, such as robots stuck in the road. To support the activity, we provided the interaction modalities and communication channels we used in the first interview for their reference.

Figure 3:

View Figure

3.4 Data analysis

We analyzed our data using a mixed inductive and deductive qualitative analysis [25]. To begin, three research team members developed a set of deductive codes by reviewing the interview and co-design study protocols. The deductive codes were mainly based on the pre-set topics in our interview questions. For example, the interview of PwMD contains pre-set topics on interaction preferences, design factors, and attitudes toward sharing the sidewalk. Then, three team members collectively coded two interview transcripts and two workshop transcripts, combining the deductive codes and bottom-up coding strategies to build an initial codebook. The first author independently coded all remaining data, with all authors convening to discuss and resolve conflicts (e.g., missing codes, disagreement on codes). Each code, along with its corresponding interview transcript section, was sorted into an affinity diagram. This process continued until broader patterns and relationships appeared. The semantic analysis steps stopped when all team members reached a consensus around key themes, including PwMD’s perceptions of sidewalk robots, practitioners’ challenges for ensuring accessibility, and robot ideas in co-design.

4.1 Concerns on Robots’ Presence and Design

4.1.2 Sense-Making and Communicating Sidewalk Robot Behavior.

Interviewees described how explicit communication could prevent and mitigate potential conflicts. All participants emphasized the significance of communication, and nine (P1-5, P7-9, P13) suggested that robots should announce their presence when turning, changing speed or trajectory, or approaching PwMD. Voice announcements were viewed as an effective and intuitive method for robots to communicate their presence and intentions on public sidewalks. P4 and P5 additionally suggested a polite, soft beep as an acknowledgment of PwMD’s presence or intent to change directions.

In addition to auditory cues, visual cues such as text and icons were deemed lightweight yet effective communication channels, especially in “crowded situations where auditory cues might be lost” (P13,15). However, participants also highlighted some limitations of visual cues. For instance, the robot’s low height might make legibility difficult, and the cognitive load associated with comprehending text might be a barrier to quick transmission of the message (P5,7). Therefore, combining text with icons could improve understanding (P7).

Nevertheless, P9 pointed out the potential for misinterpretation with oversimplified signals, such as confusion over who should move when a left turn icon is displayed. Auditory cues could also lead to confusion on crowded sidewalks, where the source of sounds might not be immediately identifiable.

Figure 4:

View Figure

4.2 Designing Interaction with Sidewalk Robots

4.3 Expectations and Aspirations for Accessible Public Robots

5.1 Organizational Practices for Infusing Accessibility into Robotic Development

5.2 Deployment and Operation of Public Robots

Designing robots for public venues is complicated, and practitioners in our interviews discussed the multiple challenges inherent in this domain. R2 emphasized the diverse reactions of bystanders, pondering over the dilemma, "How can a robot engage without disrupting regular activities in public spaces?" Echoing concerns expressed by PwMD (refer to Section4.1.2), R2 noted that a robot’s unfamiliar presence might cause individuals to halt their activities, suggesting a need for clearer robot communication. This would require modeling human perception to the robot behavior (R1), and versatile digital, and physical engagements (R8). R1, an HRI researcher, articulated a longstanding challenge:

Appropriately modeling human behavior[to the robots]...is a huge challenge that we still need to do a lot of work [...] to have robots actually behave the way we want them to and expect them to.

Thus, prior to any large-scale robot deployment, practitioners argued that it is pivotal to discern how PwMD interacts with robots in real-world settings. R2 and R6 pointed out potential disparities between what users say and how they act. Achieving this understanding would require the development and use of research methodologies that uncover how people interact with robots, such as ethnographic observation (R4), and establishing robotic simulation platforms to test and tweak robot behavior (R2, R8).

Designing robots for complex public spaces also requires an understanding of the environment and its possible impacts on human-robot interactions. As highlighted by R6, even seemingly defined environments, such as a museum with flat floors, might surprise roboticists. Such a place can have “variations in flooring—shiny, matte, black, white, all things that make robots upset... and exponentially increase the complexity of running safety analysis." In even more complex and unpredictable public spaces, R4 remarked that robots will inevitably fail, which makes designing for failure cases and immediate damage control crucial to mitigating negative or dangerous interactions.

Offering a provisional solution to such failures, R2 suggested robots be equipped to guide the public in fixing emergent malfunctions, and apologize for mishaps, an idea corroborated by prior research [21] and echoed by R3. However, as R6 emphasized, failure control might not suffice in extreme situations, such as a robot running into a toddler. Further, R6 asked, "How effective would human intervention be in these scenarios?" Drawing from their experience in safety assessments, R6 elaborated, "Merely making marginal safety enhancements to a system capable of inflicting potential harm doesn’t fundamentally alter the risk equation." R6, therefore, advocated for stringent regulations overseeing public robots, up to and including potential product recalls when necessary.

5.3 Regulatory Frameworks and Guidelines for Robot Accessibility

Given the intricacies of the robotics industry and its interactions within public spaces, five participants(R2, R4-5, R6, R8) also shared the need for more regulations and guidelines for all robotic development stages. R6 observed that while certain robotic firms have established safety protocols, emerging sectors within robotics lack the institutional "muscle memory" or corporate culture for structuring design processes with safety and accessibility at the forefront. The absence of regulations and guidelines has implications; for example, roboticists may be inclined to prioritize novel features over core concerns of safety and accessibility. This leads to a potential introduction of features that may compromise safety or inclusivity. R2, R5, and R7 also called out the need for well-defined design standards, particularly given a lack of concrete, user-friendly accessibility guidelines for robotic design. Drawing a parallel to web accessibility standards such as Web Content Accessibility Guidelines (WCAG)³, R7 underscored how integrating accessibility checking into workflows to evaluate and address accessibility issues can empower practitioners.

Several participants requested action for overarching regulations, possibly spearheaded by governmental bodies. Current regulations mainly focus on parameters like weight limits and speed [77], but R8 argued that there’s a broader spectrum to examine. Interviewees proposed rules like a robot being required to maintain a certain distance from people (R2), automatically replanning when it detects people approaching curb cuts (R3), following the social norm of staying on one side of the road in most cases (R3), and restrictions on the concurrent number of robots on a single road (R3). Other ideas included changes to the built environment that would impact how robots and people interact. For example, R2 and R8 shared ideas of dedicated pathways for robots on sidewalks, especially if their prevalence surges. R8 visualized a synergy with smart infrastructure:

When the robot is coming, we will know that it’s coming, and it’s going to use this space. But at the other time, we can definitely use the whole space when there is no robot... I do not even have to look at the robot. I know, because of the lighting that is displayed on the road or maybe projected on the road, the LEDs (R8).

However, our discussions primarily revolved around regulating robotic behavior rather than changing the environment, likely due to participants’ HRI backgrounds.

6.1 Overview of the Generated Robot Ideas

Below, we present each of the four robot ideas from the collaborations. To contextualize each idea, we describe the team composition and how each team interacted with one another⁴. These ideas focus on robots that could support PwMD and include a cargo-carrying robot, a robot to grab and hold groceries in a store, a crosswalk guide robot, and a snow plow robot for clearing sidewalks.

View Figure

Figure 6:

View Figure

6.2 Uncovered Accessibility Needs and Concerns

As part of our protocol, we asked workshop teams to center the needs of PwMD in creating their robot concept. This led to designs that focused on specific needs for PwMD such as carrying or reaching items independently (W1, W2) and providing increased safety on public walkways (W3, W4). Throughout the co-design engagements, there were instances where teams needed to work through conflicts around robot features desired by PwMD and the technical feasibility of implementing such features (frequently emphasized by roboticists). These disagreements would often lead to more accessibility needs and concerns being uncovered. To illustrate such conflicts and how they realized more detailed accessibility issues, we present two vignettes from the workshops.

Vignettes 1: Why Opt For a Robot if it Can’t Outperform a Human Assistant?

In Workshop 2 (W2), Adrianne, a power chair user with limited hand dexterity, expressed a desire for assistance in fetching groceries. When Michael, a roboticist, sought to understand how long Adrianne’s grocery fetching task would take, he learned that with human assistance, Adrianne could fetch five items in approximately 20 minutes. "But would a robot match a human’s efficiency?" Adrianne asked. "Not a chance," Michael admitted frankly, acknowledging the technical reality that robots would likely lag behind humans in terms of efficiently navigating a grocery store and locating items. "My life is about taking things slowly. It’s fine if it takes three times longer," Adrianne responded. She shared feelings of being hesitant to inconvenience others when she needed assistance in the store. Adrianne explained that a robot might circumvent such social discomforts and support her autonomy. Through this exchange, the core value of the robot Adrianne proposed became clear to Michael: not to offer mere efficiency, but the prospect of furthering her autonomy.

During the co-design sessions, the PwMD and roboticists often came to a compromise on technical features, but there were also cases where PwMD remained unwavering in what they believed were their most important requirements.

Vignette 2: I Value Your Concerns, But This is What I Need As a Cane User

In W1, Lily, a cane user, and Bella, an HRI researcher, decided to conceptualize a cargo-carrying robot. Lily immediately proposed that the device be built to “adjust to different heights for people to easily get access to it” because of her inability to bend over. She elaborated: “If I need to press a button or tell it to do something, it’s important to me that I don’t have to really stretch up or really bend down to do that because that’s something that would catch me off balance somewhat.” Bella scrutinized this functionality, worrying that “it might be terrifying to other groups like elders” when changing shape. Although Lily admitted there was a possibility that a robot adjusting its height could be disturbing, she insisted it would “be useful for broader groups of people with mobility disability like wheelchair users” This feature later became one of the main attributes of the robot because Lily suggested various heights the robot should cater to, taking into account situations where she might be using a wheelchair or standing with the support of canes. She felt that she was shorter than the average person.

Across the Workshops, teams discussed a variety of accessibility needs that their robot concepts aimed to address. While the true feasibility of each concept would require extensive testing, we believe that each brings up more needs that could be generalized to other robots for public use. The following summarizes the core accessibility needs of each workshop, grouped by functionalities and interactions.

View Figure

6.3 Summary of the Implications to Future HRI Practice

Abstracting beyond the specific needs of each robot, the collective needs uncovered in the workshops appear to address three areas. First, there is a need for different robots to carry and transport items for PwMD as they are moving about the world. Second, various communication modalities were desired, including voice interaction and touchscreen-based interfaces. Across workshops, it was clear that these should be carefully considered to meet people at their abilities, with examples of smartwatches on the user’s wrist or touchscreens that can move to meet the person, rather than requiring someone to reach for a screen. Thirdly, the workshops revealed a need for robots to clear the sidewalk and make it more accessible for PwMD such as by communicating with other technologies like cars to have them avoid PwMD or by clearing the sidewalk of obstacles.

Overall, these examples suggest opportunities for robots that work on behalf of PwMD rather than simply getting out of the way of people. While each design concept might suggest specific technical requirements for future public robots, we acknowledge that our results represent only four groups who all ended up having fairly different design ideas. In lieu of providing specific recommendations, we believe that further co-design sessions among people with disabilities and roboticists can reveal even more opportunities for supportive public robots and that designs addressing the specific use cases and contexts are needed. Based on our sessions, two areas for future exploration include robots that operate on sidewalks that can communicate with other road agents (as suggested in the crosswalk robot) and robots that can clear debris from a walkway, making it more accessible to others (as suggested by the snowplow robot). Further, robots aimed at carrying goods might consider how their service could be used for people carrying their own items rather than receiving a delivery. Finally, all designs suggested clear communication between the robot and people, however, communication modalities should be tested with a diverse set of users and should likely be multimodal to accommodate multiple communication abilities and preferences. Overall, we believe HRI practitioners leveraging similar co-design methods can uncover new technical requirements for improving accessibility using public robots.

Figure 8:

View Figure

Figure 9:

View Figure

7.1 Learning from Interviews and Co-Design for More Holistic Public Robots Research Methods

Inspired by prior qualitative-based HRI studies [12, 13, 75], we leveraged semi-structured interviews and co-design workshops to study two sets of stakeholder perspectives concerning emerging public robots, as well as collaborations between them. In our co-design workshop, we intentionally invited robotic practitioners from diverse backgrounds, from design and engineering to industry and academia. We found that their backgrounds and expertise influence their views and, in turn, the co-design outcomes. For instance, a roboticist with extensive experience in proxemics frequently raised questions regarding how wheelchair users navigate robots in public spaces. We interpreted this in two ways: firstly, it shows the opportunity space of incorporating accessibility in different facets of HRI; secondly, future studies should be more mindful of balancing the professional participants to avoid the outcome being over-directed by the participant’s personal experience. We recommend future studies adopting iterative and long-term co-design sessions [56] to facilitate reflective design processes. This can be effective in providing a more comprehensive lens on the accessible feature and also improve the achievability.

Through interviews with PwMD, we learned of potential conflict scenarios between PwMD and the robots (refer to Section 4.1.1) influenced by a range of HRI design factors (refer to Section 4.1.4). Building on these scenarios, future work could concentrate on refining the interaction design of sidewalk robots [37], accommodating the multifaceted requirements of PwMD [18, 48], and taking into account diverse robotic design dimensions such as anthropomorphism [9] and non-verbal signals [10, 28] [47, 49, 65], [18, 24, 48, 69] Nonetheless, there remains a pressing need to test these interactions and qualitative findings in real-world evaluations. For instance, as discussed in Section 4.2.2, our interviews suggested PwMD have a preference for voice interaction, valuing both its inherent naturalness and avoiding direct, tactile engagements while navigating the street. In contrast to our findings, some literature suggests a general expectation that robots remain non-vocal [72], and that the willingness of individuals to instigate conversations with robots might be minimal [38, 81]. Yet, in scenarios laden with potential conflicts, such as robots obstructing vital sidewalk resources, lacking different forms of interaction, including speech, could lead to issues. Consequently, to truly cultivate an interaction paradigm that is both usable and societally acceptable, it is imperative to delve deeper into the disparities between PwMD’s anticipated and actual responses to robots in public via comprehensive contextual evaluations and observations [32, 80].

7.2 Inclusive Stakeholder Engagement: A Prerequisite for the Ethical Design of Public Service Robots

Our interview findings explored opportunities to make public robots more accessible, such as adapting to their mobility aids (Section4.1.3), enhancing regulations to move beyond physical size and speed to consider robot behaviors (Section5.3), and providing mechanisms to address robot malfunctions (Section 4.3). However, we need to be more cautious about techno-capitalism: there is a risk that robot companies might leverage these enhancements to assert the accessibility of their robots, which contributes to rationalizing the deployments. Indeed, certain segments of the disabled community have already voiced reservations concerning private-owned delivery robots taking up space on the public sidewalk  [12].

So, what are the ways that accessible robots can enhance the accessibility of public space? PwMD in the workshops suggested that the robots could enhance their daily living, including improving their carrying capabilities and alleviating transportation anxieties (Section 6.1). They also collaborated with roboticists, making design decisions that led to accessible features such as adjustable robots and examining the value of the robots within their technical viability (Section 6.2). The accessible features they raised might also challenge current robot capacities and roboticist’s conception of universal design (Section6.1). Such co-design engagements may offer a chance for roboticists to learn more about accessibility and reflect on their practices. Thus, to ensure accessibility, it is imperative to engage PwMD both in the evaluation of robot designs and early ideation of robot features and interaction.

We also recognize that solely focusing on people with mobility impairments is insufficient,and may pose a challenge to the generalizability of our findings. For example, in the conflict scenarios discussed in Section 4.1, interviewees highlighted design factors that led to a sense of competition for the curb cut, a resource indispensable to PwMD. However, people with visual impairments may have different perceptions of robot behavior than those seen as obstructive by PwMD. Therefore, design suggestions such as voice interaction may be invalid or even unfavorable to them. Even though the findings presented here are specific to the populations with whom we engaged, our interview design (ref to Section 3.2) is generalizable as it effectively introduced the concept of public robots, different design factors, scenarios of encountering the robots, and a comprehensive set of interaction modalities. All of our PwMD participants were able to comment on the robot even without prior exposure. Thus, future research can reuse and adapt our interview and co-design methods to engage other groups to compare and extend our findings on accessible public robot design.

Moreover, findings from our engagements with robot practitioners suggest collaborations with other parties, such as policymakers and urban planners, are also necessary to realizing the vision of accessible robot design (refer to Section 5.1). The idea of broader engagement beyond practitioners and everyday people echos recent HRI studies on ethical congruent operations [57, 82], which prompts vital inquiries such as "Who gets to be included in the robot design process?" [57]. We argue that future robot design should include multi-stakeholder engagements through co-design workshops with appropriately designed participation frameworks [85, 86] that enable such collaboration.

7.3 Enhancing Standards, Policies, and Regulations for the Development of Accessible Public Robots

Accessibility in robot design and development remains insufficient, even as sidewalk robots are increasingly deployed in public spaces. Prior approaches to ensuring accessibility found in desktop computing accessibility may offer promising opportunities for benchmarking [76], scaffolding [23], and validating [50] robotic accessibility. Another needed aspect is the development of overarching design frameworks for sidewalk robots that include accessibility recommendations [2]; this is also paramount to nurturing a culture of accessibility amongst roboticists (as described in Section7.2). This will take time and as noted in Section5.3 the lack of overarching design frameworks can be a challenge, especially for start-ups who potentially lack the awareness, bandwidth, and resources to emphasize access in design and development. As such, we saw robotic practitioners advocate for enhanced guidelines and regulations to promote accessibility.

Proactive public policy can do much to ensure that society reaps the greatest benefits from new technology while reducing possible harms  [17]. Our qualitative data from PwMD reveal that current regulations concerning sidewalk robots may be overly broad and, in some cases, excessively prescriptive. For instance, PwMD’s perception of the robot as lacking communication and other necessary interaction functionalities to fully adapt to the complexities of the sidewalk (refer to Section4.1.1) can challenge the sidewalk robot’s current classification as a "pedestrian." Some robotic behaviors such as inadvertently blocking access or malfunctions could also violate U.S. mandates that require public sidewalks and services to be universally accessible  [1]. Thus, beyond simply defining the maximum weight, size, and speeds of the robot [4], we argue future public robot regulations should go further and more comprehensively attend to the interactions and effects—intended or unintended—of sidewalk robots to ensure accessibility for all.

Additionally, the synergistic relationship between public robotic policies and urban planning remains under-explored. The rise of public robots could radically reshape the distribution, utilization, and dynamics of urban spaces. We have seen recent smart city practices start to consider establishing a dedicated line on the sidewalk for mobile robots to operate  [79]—our robotic practitioner interviewees also mentioned similar ideas (refer to Section5.3). However, from an accessibility lens, salient questions arise: Will reallocating public space to robots exacerbate existing accessibility constraints on public sidewalks? To what extent can accessibility challenges be alleviated? And, fundamentally, should urban planning cater to robots, or should robot designs adapt to current urban environments?