1 Introduction

The era of automated vehicles (AVs) is upon us, promising to revolutionize our mobility and transform the in-vehicle experience fundamentally. These vehicles, capable of automated operation through environmental sensing and responsive actions, hold the promise of reshaping transportation dynamics [1,2,3].

Among these innovations, shared automated vehicles (SAVs), employing AVs in a collective framework, emerge as an economically viable option, potentially making travel more affordable while curbing congestion, emissions, and the land footprint of parking spaces through higher capacity utilization in comparison to privately used vehicles [4, 5]. However, despite their manifold advantages, challenges persist, particularly regarding the absence of a human driver [6,7,8]– a source for a sense of security [9].

While current SAVs operate with onboard operators–at least in Europe– due to legal constraints, the future landscape envisions fully automated SAVs for public use. Yet, the acceptance of these vehicles may hinge on addressing the perceived insecurity stemming from the lack of a human driver. To bridge this gap, Schuß et al. [10] suggest a Conversational Agent (CA) functioning as a Digital Companion (DC) to emulate certain driver functions, although specifics remain open.

Recognizing the cultural diversity in perceptions of automation trust and security needs, our study, rooted in a German institution, delves into contrasting cultural landscapes. We aim to explore how different countries, delineated by the Hofstede framework [11] and varying perceptions of security in public transportation (PT) systems, shape users’ views on SAV security needs and their expectations from a DC. Germany, statistically exhibiting high personal security in PT but lower perceived security [12], stands in contrast to South Korea, where extensive CCTV coverage creates a perceived high security in PT [12,13,14]. Additionally, we incorporate perspectives from regions where PT faces security challenges. For instance, Latin American PT users encounter safety concerns [15], particularly related to gender-based violence and personal safety, presenting a compelling basis for a cross-cultural comparison between Colombia, Germany, and Korea.

Our work involves analyzing security needs with Colombian participants and conceptualizing a DC with German and Korean participants. Employing participatory design techniques in co-creation workshops, we developed both robot- and human-like DCs to enhance perceived security in SAVs, and evaluate them in our main study including participants from Colombia, Germany, and Korea.

2 Related work

2.1 Advantages of shared automated vehicles

Jeon et al. affirm that the primary perceived benefit of Automated Vehicles (AVs) is safety, attributed to the notion that 94% of vehicle accidents stem from human errors [16, 17]. AVs, equipped with collision avoidance systems and greater data access, hold promise in reducing accidents [18].

In the realm of automated PT, SAVs are poised to revolutionize shared mobility by addressing economics, wherein vehicles often remain empty 95% of the time, according to Sperling [19, 20]. Chen forecasts competitive pricing for SAVs compared to conventional shared mobility, potentially making AVs financially viable [4, 21, 22].

SAVs offer multifaceted benefits, including reduced vehicle ownership, traffic congestion, parking space demand, and emissions [4, 5]. However, their lack of a driver poses significant challenges, particularly concerning in-vehicle security. Studies indicate that personal security concerns might impede SAV adoption [23,24,25]. Notably, gender-based differences in security perceptions emerge, with women expressing higher levels of anxiety and fear regarding shared mobility [7, 26].

Proposed solutions to mitigate the absence of a driver include Digital Companions (DCs) to alleviate the feeling of being unobserved [10, 27]. These DCs could potentially improve in-vehicle security and crime prevention, as suggested by Sanguinetti et al. [26]. It is imperative to research the design of these DCs due to the critical role they could play in enhancing in-vehicle security and crime prevention, addressing the absence of a human driver and fostering a sense of safety among passengers.

2.2 Designing conversational agents and digital companions

In the landscape of conversational agents (CAs), Niess et al. categorize Digital Companions (DCs) into active and passive types, emphasizing the contextual importance of their roles, thereby suggesting a shift in human-DC relationships [28]. Burmester et al. stress the significance of security, feedback, and memory for DCs in Shared Automated Vehicles (SAVs) operating as public transportation [29]. The absence of human drivers in SAVs presents challenges in service and social dynamics, considering the diverse roles typically fulfilled by human drivers in these vehicles [30, 31]. These roles span from psychologist to supervisor [31] underscoring the diverse requirements in human-DC relationships as they evolve from hierarchical to multifaceted social dynamics [32]. When it comes to the appearance of DCs, anthropomorphism, attributing human-like characteristics to technology, plays a pivotal role in establishing user trust and satisfaction with automation [33,34,35]. In this context, cultural differences significantly influence perceptions and preferences regarding CAs and automated systems, underscoring the need for culturally adapted DCs, particularly in security roles within SAVs, to effectively address diverse user needs and foster trust across varied cultural contexts [36,37,38,39,40]. Therefore, the acknowledgment of cultural nuances becomes crucial in designing DCs tailored to different cultural inclinations, especially in the context of security roles in SAVs, potentially transcending cultural boundaries to promote a sense of security and trust.

2.3 Culture set in focus: Colombia, Germany, and South Korea

In this work, we prioritize diverse cultural inclusion to mitigate a Western-centric outlook, aiming to minimize biases. Our selection of cultures is primarily guided by the security levels within their public transportation systems, assuming these experiences significantly influence attitudes toward SAVs as a form of automated, shared transport [41], as well as on Hofstede’s cultural dimensions which is one of the most cited and used framework in the field of cross-cultural research [42,43,44] based on extensive empirical research [43,44,45]. Hofstede’s findings highlight Asian and Latin American cultures (e.g., South Korea, Japan, China, Colombia) characterized by high power distance and collectivist values, emphasizing hierarchical structures and prioritizing group harmony over individuality [11]. These cultures tend to accept automation akin to authority, correlating with their high power distance, affecting their attitudes and acceptance of automated systems [46, 47].

Colombia faces substantial security challenges in its public transportation, marked by gender-based violence, harassment, and high crime rates [15, 48, 49]. In contrast, South Korea’s transportation system boasts a perceived sense of security, attributed to lower crime rates and government initiatives like the u-City project, aiming to integrate technologies across urban aspects for enhanced living experiences [50, 51]. South Korea’s reputation for safety extends to women travelers, ranking among the top 20 safest countries for solo female travel [13].

Hence, our choice to include these diverse countries into our research was purposeful: South Korea embodies a context where PT is widely perceived as secure [13], Germany stands as a model with strong security protocols but a visible disparity between real and perceived safety [52], and Colombia acts as a representation of areas grappling with (perceived) insecurity within their public transportation networks [15].

Based upon related work we pose the following research questions:

  • RQ1. How should a digital companion for shared automated vehicles be designed from the perspective of people from Colombian, German, and South Korean?

  • RQ2. Does a digital companion enhance perceived security in shared automated vehicles?

  • RQ3. Which concept for a digital companion is prefered across Colombia, Germany, and South Korea?

To investigate these research questions we followed a human-centered design (HCD) [53] approach and included potential users in the early design stages when developing our DCs through a series of interviews and co-design workshops to answer RQ1. Subsequently, we conducted an online survey involving participants from Colombia, Germany, and South Korea to evaluate these concepts and answer RQ2 and RQ3.

3 Material and methods

3.1 Pre-studies: interviews and co-design workshops

We conducted in-depth interviews with participants from Colombia, aiming to delve into their perceptions of security within PT and extract insights applicable to security considerations in SAVs. Following this, we facilitated co-design workshops employing participatory design techniques [54, 55] involving participants from Germany and Korea. The objective of these workshops was to extract design implications for DCs that enhance security perceptions in SAVs, aligning with the preferences of potential users. Due to the constraints posed by the COVID-19 pandemic, conducting the co-design sessions in Latin America was unfeasible at the time. Consequently, we will elaborate on the findings from the interviews with Colombian participants and the outcomes of the co-design workshops involving German and Korean participants in the subsequent sections.

Interviews with colombian participants

We conducted interviews via Zoom with 8 participants (4 men, 4 women, 0 diverse, 0 n/a; from 27 to 46 years; \(M = 33.3 \); \(SD = 6.2 \); \(Mdn = 33 \) years) from Bogotá in Colombia. The participants for the interviews were recruited via snowball-sampling by our acquaintances in the city of Bogotá and the aim of the study was to explore the general attitude and security needs of the participants in SAVs. The interviews have been conducted in Spanish. All sessions were video and audio recorded and transcribed verbatim afterwards. The material was analyzed using qualitative content analysis [56, 57] and the results have then been translated into English. We applied inductive coding and refined the themes and codes in an iterative process until the final themes and codes were identified. The analysis was carried out using MAXQDA [58]. In this section we will focus on the results relevant for the design of our DC.

Our results confirm that security is of superior importance for people’s mobility in Bogotá. In this context the bus driver plays an important, although ambiguous, role. They would stop and close the door if something happens in the bus, or they would close the door and pass by people and do not let them in the bus for security reasons: “Normally, in Latin America, the driver also functions as a kind of controller or security of what happens inside the vehicle. I don’t know how it [SAV] will make up for that lack of security of the shared vehicle.”, [P01]. This is in line with previous research from Europe where participants are not used to traveling without a driver either and were expressing concerns due to sharing trips with strangers [6, 7, 10, 59]. In other cases, the driver may not be able to do anything (“But many times they don’t want to risk anything happening to them either.” [P06]). Contradicting to that, the drivers themselves were seen as a cause of risk for accidents or even for crime and harassment, which corresponds to previous research on ride-sharing services [60] pointing at passengers particularly worrying about the driver “being creepy” [61], distracted, or causing an accident. This is particularly interesting, as related work emphasizes that the absence of a driver in SAVs is potentially problematic for passengers and could lead to the rejection of these vehicles. However, it needs to be evaluated whether this will hold true for Colombia, as well.

Co-design workshops with German and South Korean participants

In total, 19 participants, 9 Koreans (4 women, 5 men, 0 diverse, 0 n/a; from 19 to 35 years; \(M = 24.56 \); \(SD = 4.90 \); \(Mdn = 25 \) years) and 10 Germans (5 women, 5 men; from 16 to 64 years; \(M = 29.30 \); \(SD = 17.72\); \(Mdn = 30 \) years) took part in the co-design workshops. Four workshops, each including three to six participants, were conducted. One workshop exclusively accommodated Korean participants, while another only involved German participants. The remaining two workshops were mixed sessions, involving both Korean and German participants, to foster cross-cultural discussions and to allow for a broader spectrum of insights regarding potential divergent needs and concepts regarding the DC. The participants encompassed primarily students or individuals interested in the suject and had varying degrees of familiarity with SAVs. With the exception of the German-exclusive workshop conducted in German, all workshops were conducted in English. German participants were recruited through our university networks and in Korea, we collaborated with local research institutions and used social media for recruitment. All participants attended the workshops voluntarily after signing a consent form. We used Zoom as the platform and Miro, a digital whiteboard tool, to facilitate collaborative engagement among participants. Ensuring proficiency in using Miro, a preparatory self-recorded video tutorial was disseminated to all participants prior to the sessions.

The workshop structure included four exercises, an introductory round, a break, and allocated time for feedback. Following recommendations from Lee et al. [62, 63], an introductory round was incorporated as a warm-up activity. The initial exercise centered on exploring the challenges surrounding SAVs and participants were encouraged to freely express their personal opinions towards SAVs without the constraints of right or wrong answers. They were asked to brainstorm ideas regarding their individual security needs within an SAV and prompted by questions such as “What kind of DC would offer you security?” and “Where within the SAV would it ideally be placed?”. Utilizing sticky-notes on the Miro board, participants collected their thoughts, facilitating a deeper immersion into the topic. Subsequently, participants were grouped into teams of two or three for the next exercise, where they crafted moodboards depicting their ideal DC within an SAV. Equipped with an empty frame on the Miro board and an array of images, colors, and words, the teams selected elements they associated with the DC, dragging these selections onto the designated frame. This approach aimed to infuse a sense of playfulness and motivation among participants, particularly catering to the comfort and motivation of participants from East Asia in interview settings [63]. Upon completion, each team presented their choices and discussed the varying opinions and discussions that arose within their respective teams.

Fig. 1
figure 1

Synthesized moodboard created by participants during the co-design workshops (Image sources: Internet)

Full size image

During the subsequent exercise participants were sketching their envisioned design of the DC for SAVs within a 20-minute timeframe. Once completed, the drawings were placed onto the Miro board, and all participants presented their designs and were able to comment on the other designs. The workshop concluded with a brief group discussion and additional elements that were considered as central for the DC were collected. Each workshop terminated with an opportunity for participants to offer feedback. Each workshop lasted approximately 150 min.

Findings from the co-design workshops

The completed Miro boards and video recordings of the workshops were analyzed using qualitative content analysis [56] to categorize the findings across all workshops. In the subsequent sections, we report on the most significant results. Some quotes were translated form the respective languages into English.

Across all participants, the most frequently cited challenge, irrespective of nationality, pertained to overall security. For German participants, this subject predominantly circled around in-vehicle security: “What happens actually if, for example, in [the SAV] somehow there is a fight or so? Because no one can resolve it [...], because normally it’s the bus driver [...] who does that” [P03, German participant]. However, Korean participants voiced concerns beyond in-vehicle security, encompassing apprehensions about personal information privacy and cybersecurity.

Moreover, discussions arose concerning challenges linked to the fear of losing control, a concern primarily expressed by German participants. Participants underlinded the importance of retaining control over the vehicle, highlighting the desire to halt the vehicle: “So how do I get it to stop?” [P01, German participant]. Simultaneously, there was a recurring emphasis on the need of receiving system updates: “How do I know where it is going and if it will take me to where I need to go?” [P04, German participant].

With a DC for SAVs, participants primarily associated the characteristics reliable, trustworthy, friendly, providing security, and comforting. In total, we gained nine moodboards from the workshops, leading to a compilation of images, colors, and words that received multiple selections (see Fig. 1). A consistent choice across all teams was an image depicting a young girl holding hands with a robot, symbolizing the interface between human and digital realms: “I picked this one because I want [the] digital companion to be friendly” [P15, Korean participant]. For Korean teams, this illustration denoted a portrayal of a friendly robot. Another recurring choice among mixed teams was a photograph of someone resting their head on another person’s shoulder, akin to the image of two hands clasping: “Those [...] basically say the same I guess like trust and comfort” [P11, German participant]. Additionally, an image of a phone with connected headphones symbolized entertainment, while pictures of headphones and computer monitors were also repeatedly selected. A photograph depicting two girls engaged in martial arts represented notions of fun and cultural diversity. Similarly, an image of waves correlated with the concept of reliability, whereas the presence of stormtroopers from Star Wars was interpreted diversely. One participant associated it with technology and robots, while another linked it to “law and order and obedience” [P18, Korean participant]. Furthermore, the recurring selection of Superman images aligned with associations of security.

Regarding the color white was selected representing “openness and space” [P11, German participant]. Black was used for contrast or to indicate security and trust. Additionally, the colors blue and purple were prevalent among the mixed teams “because it’s something new, something modern [...]. It’s kind of a futuristic color scheme in some way” [P12, German participant]. The light blue color was picked for its bright and softness. Apart from that, blue was associated with safety and comfort: “blue symbolizes like the safety and comfort and especially the lighter colors kinda make us feel calm” [P16, Korean participant].

The German participants, especially those in the German workshop, tended to draw a human-like DC (HDC). Two German participants envisioned the DC as a Voice Assistant or Graphical User Interface (GUI) without any visual appearance. One participant sketched a rather abstract appearance and additionally an animal-like version for children. Another participant imagined the DC as a car with a face. Most participants in the Korean workshop sketched a robot-like DC (RDC). One participant imagined the appearance of the DC to be a bus with a face. In the mixed workshop, one Korean participant drew the DC as a cat, while another drew a genie. Based on the sketches of the German participants, a visual appearance of the DC was particularly important to them. Nevertheless, most German participants who commented on this question preferred to see and hear the DC: “I would [...] say both because then I would have more the feeling [like] there is someone who is really driving like a real driver” [P08, German participant]. In contrast, the Korean participants prioritized acoustic over visual communication, but would also prefer to see and hear the DC: “the more the better” [P19, Korean participant]. However, two participants were concerned about being disturbed by the DC. Therefore, they suggested in certain situations to only use sound without a visual appearance of the DC. Both the German and Korean participants had very diverse opinions on whether there should be one DC for all passengers or if everyone should have a personal DC and some liked to have both. The German participants preferred a HDC: “I want [it] to look like real humans” [P04, German participant]. Most Koreans, on the other hand, preferred a RDC: “sometimes they try to make it like humanized [..] but it’s kinda creepy [...] so if you are not gonna make it perfect just go with the robot side cause I’m okay with machine like [or] robot-like appearances” [P14, Korean participant]). Participants did not express a preference regarding the voice of the DC. Yet, two Germans stated that the voice should be as realistic and organic as possible. The few Koreans who responded to the question had very contrary opinions. One participant said the DC should sound like a robot while another said it could sound like a human: “it doesn’t need to sound like a robot I think it can [...] sound like a human that is in a call center” [P18, Korean participant).

3.2 Conceptualization of the digital companions

Building upon the findings from the co-design workshops and relevant research, we derived two distinct concepts for a DC tailored for SAVs: one featuring a robot-like design and the other resembling a human-like figure. As the security challenges inherent in SAVs appear notably similar across Colombian, German, and Korean contexts, the envisioned roles and personalities for the DC remain akin. Consequently, our aim was to design two agents perceived as both competent and warm [64], given the relevance of these traits in addressing security concerns. Additionally, we aimed to implement empathy within our companions, as advocated by [65], shaping their behavior to be passive, emphasizing a caring and cautious disposition [28]. Nonetheless, our two concepts differ mainly in terms of appearance and vocal characteristics.

Robot-like digital companion

As one of the main requirements was that the DC should be friendly, we first made a few sketches to explore how to design a friendly-looking robot. A kind face and especially large eyes contribute to an overall friendly look and a cute appearance also made the robot look friendlier [66]. The Korean participants in the workshops wanted a protective but also obedient DC and frequently associated an image of stormtroopers from Star Wars with the DC for SAVs. Similarly, Lee and Sabanović [67] suggest that Koreans envision robots as animate and social, but hierarchically subservient to humans. As participants in the co-design workshop choose the image with the Star Wars stormtroopers for their moodboards because they represented law and order and obedience, we equipped the RDC with an armor to lean on this picture. In addition, it seemed important to show that the DC is supervising the situation. By drawing upon previous work [68], we included shining ears with antennas as an implicit cue indicating that the robot is also listening to the users. The final design was created in Adobe Illustrator and we named this version of the DC “Bo” (see Fig. 2).

Fig. 2
figure 2

Final design of the robot-like DC “Bo“: Shining ears effect (left) and complete appearance (right); Image source: own illustration

Full size image

Human-like digital companion

We designed three culturally adaptable designs for the HDC with two key motives. Firstly, there have been observations from social critics highlighting concerns regarding the predominantly white appearance and blue eyes commonly found in robots and avatars, raising ethnic perspective issues [69]. Our aim was to mitigate any potential racial biases in the human-like portrayal of our HDC. Secondly, the cultural representation of our HDC might sway participants’ perceptions. Thus, we sought to sidestep any possible influences stemming from individuals’ self-identification with their respective cultures.

Yet, it remained crucial for the HDCs to maintain comparability across the three cultures encompassed in our study. Consequently, we made slight adjustments to their appearances to align with the respective cultures (refer to Fig. 3). During the workshops, participants emphasized the significance of a serious and professional demeanor in the DC, coupled with a sense of approachability. To address these expectations, the HDC was fashioned with a casual dress shirt. We opted for a light blue hue as this color was popular among the workshop moodboards. Additionally, we added a pair of glasses, to add competence and trustworthiness to the DC’s appearance. Similar to the RDC, itis crucial to note that the HDC assumes a supervisory role. Thus, akin to the RDC’s ears, we equipped the HDCs with luminous headphones, symbolizing their attentive listening to users. The HDCs were designed utilizing Ready Player Me [70].

Fig. 3
figure 3

Final designs of the human-like DCs (from left to right): Colombian (“Roberto”), German (“Bruno”), Korean (“James”). Image source: Ready Player Me/ Animaze by FaceRig

Full size image

Voice and animation

The voices of the DCs were generated using the Audio Content Creation tool provided by Microsoft Azure. Initially, we selected suitable voices from the tool’s available options. Drawing from insights garnered from the workshops, we opted for an organic voice for the HDC, emphasizing natural speech with a pleasant tone to infuse a human-like quality. For the RDC, we chose a voice slightly less realistic yet still human-sounding. The script for the DC’s speech was integrated into the program, and subsequent recordings of both voices were captured using OBS Studio.

Moving on to animation, the RDC’s design was crafted using Adobe Illustrator, and Adobe Character Animator facilitated its animation. Conversely, the two HDCs were animated using Animaze by FaceRig [71]. Both tools relied on face and head tracking for expressing facial gestures, coupled with audio-based lip synchronization for mouth movements. The internal recording tools embedded within these programs were employed to capture the animations.

3.3 Scenario-based online experiment for evaluating the human and robot digital companion

In our main study we evaluated the two developed concepts, the RDC and HDC, comparing them with a baseline concept without a DC. To make the concepts tangible, we implemented three scenarios on an SAV in Virtual Reality (VR) using Unity (see Fig. 4). All three scenarios depicted situations in which one might feel insecure or confused. These scenarios were presented to our participants as videos in an online questionnaire and we followed a within-subjects design with every participant experiencing all three versions (baseline, robot-like digital companion = RDC, human-like digital companion = HDC). We opted for an online questionnaire due to the cross-cultural focus of our work so we would be able to conduct the study with participants from Colombia, Germany, and South Korea. To enhance immersion, the scenarios were recorded from the participants’ perspective.

The study was conducted in accordance with the ethical guidelines stated in the Declaration of Helsinki [72]. Participants took part voluntarily, were obliged to provide their written informed consent, and had the opportunity to abort the study at any time without reasons.

Fig. 4
figure 4

VR setting of the SAV in Unity (Image source: Screenshot of Unity Scenario)

Full size image
Fig. 5
figure 5

VR setting of the three scenarios in Unity (from left to right): scenario 1 with RDC, scenario 2 with German HDC, scenario 3 with Korean HDC (Image source: Screenshot of Unity Scenarios)

Full size image

The spoken and displayed text in the videos was translated from German into Spanish and Korean and corresponding subtitles were added to the videos to improve accessibility. A pilot study with three participants was conducted to ensure that the content of the videos was comprehensible. We used the following three scenarios:

  • Scenario 1: Alone with Stranger In the first scenario, the participant is alone within the SAV, accompanied by a stranger (see Fig. 5). The role of the other passenger was deliberately designed to evoke a sense of discomfort in the participant. Positioned opposite the participant, the co-passenger maintained a constant gaze throughout the ride, exhibiting unnaturally rigid movements— leaning back and forth while nervously tapping their knee with their fingers. In scenarios involving a DC, the ride starts with the DC warmly welcoming the passengers, introducing itself by name, and emphasizing its availability for any inquiries throughout the journey. This intentional introduction aimed to immediately draw the participant’s attention to the DC’s presence, alleviating the sense of being alone with a stranger. Additionally, to provide a sense of security, the DC informs the passengers after a brief period that two more passengers will get on board shortly.

  • Scenario 2: Alone with Group of Men In the second scenario, the participant within the SAV is surrounded by a group of men (see Fig. 5). The behavior of the group aims to evoke a sense of insecurity in the participant: one standing man continuously singing and performing dance moves throughout the ride, while the two seated men loudly laugh, make exaggerated gestures, and wave with their hands or make victory poses. To amplify the atmosphere, the sound of a YouTube video with a boisterous group of people [73] served as a backdrop, portraying the other passengers as excessively noisy. Similarly, within the scenarios involving a DC, the ride starts with the DC offering the same introductory announcement as in scenario 1. As the noise in the SAV persists, the DC intervenes by requesting the other passengers to lower their voices. As the request is disregarded, the DC issues a second admonishment and warns to report to the control center if the situation does not improve. This intervention was intended to reassure the participant that the situation was under supervision and, if necessary, would be taken care of, thereby fostering a sense of security.

  • Scenario 3: Sudden Stop In the third scenario, the participant is alone on the SAV as it suddenly stops for no apparent reason (see Fig. 5). The visual effect of the vehicle stopping was created by pausing the \(360^\circ \) video environment. Again, the concepts with DC started with the same welcoming as the previous two scenarios. After the SAV stops, the DC informs the participant about why the SAV stopped and that the ride will continue soon to let them know that they do not need to worry. After some time, the DC explains that the problem has been fixed and that the ride will continue.

3.3.1 Participants

In total, 136 participants, 29 Koreans (15 women, 13 men, 0 diverse, 1 n/a; from 19 to 67 years; \(M = 29.14\); \(SD = 11.0\); \(Mdn = 25\) years), 50 Germans (23 women, 26 men, 0 diverse, 1 n/a; from 18 to 78 years; \(M = 35.6\); \(SD = 19.13\); \(Mdn = 24.5\) years), and 57 Colombians (30 women, 27 men, 0 diverse, 0 n/a; from 19 to 73 years; \(M = 31.89\); \(SD = 15.37\); \(Mdn = 24\) years) participated in our main study. Participants were recruited through university mailing-lists, social media channels, and word-of-mouth and participated voluntarily. To motivate more people to participate, three gift cards per cultural group worth app. 15 US dollars were raffled.

While most Korean participants stated to have completed a bachelor’s degree (48.3%), most Colombian and German participants specified a high school degree or equivalent as their highest degree of education (Colombians = 42.1%, Germans = 48.7%). Most Korean participants reported traveling by car at least once per week (31.0%) or several times per week or daily (20.7%). Similarly, many German participants stated that they travel by car at least once per week (30.8%) and most of them several times per week or daily (33.3%). Most of our Colombian participants indicated to travel by car several times per week or daily (45.6%). The majority of Colombian and Korean participants reported using PT several times per week or daily (Colombians = 63.1%, Koreans = 58.6%). Some German participants used PT several times per week or daily (33.3%) but there were also many German participants that reported using PT less than once per month (25.6%).

Since a 6-item short version of the State-Trait Anxiety Inventory Trait Measure (STAI-T) [74] was used to measure trait anxiety, the scores were divided by 6 and multiplied by 20 to give a range from 20 to 80, making the scores compatible with the original STAI-T scores, as suggested by Graff et al. [75]. All men, Colombian (\(M = 39.13\); \(SD = 5.66\), \(Mdn = 40.00\)), German (\(M = 41.04\); \(SD= 9.94\), \(Mdn = 40.00\)), and Korean (\(M = 43.59\); \(SD = 10.04\), \(Mdn = 40.00\)), showed above-average anxiety traits (Mean of men from 30 to 59 years = 34.59; CI = \(+/- 5.5\)). The Korean women (\(M = 46.44\); \(SD = 9.47\); \(Mdn = 43.33\)) and Colombian women (\(M = 43.44\); \(SD = 8.77\); \(Mdn = 40.00\)) also scored high on trait anxiety (Mean of women from 15 to 29 years = 35.65; CI = ± 5.7). The trait anxiety scores of the German women (\(M = 41.82\); \(SD = 12.84\); \(Mdn = 40.00\)) tended to be average, yet rather high (Mean of women from 30 to 59 years = 36.85; CI = ± 5.7) [76].

3.3.2 Procedure and measurements

The study was conducted using an online survey and the recorded VR scenarios were integrated in the questionnaire. We had a mixed design approach with the between-groups design condition “culture”. Participants were split into three groups— Colombians, Germans, and Koreans. Each participant experienced the three concepts as the within-groups design condition (baseline = BL, robot-like digital companion = RDC, human-like digital companion = HDC) in a randomized order. Within each condition, participants engaged with the three VR scenarios and provided evaluations. To keep the process within limits and reduce participants’ cognivite workload, the scenarios were presented consecutively, minimizing the need for respondents to fill out the questionnaires more than three times in total (for the conditions BL, RDC, HDC). However, the baseline was consistently displayed as the first scenario. This decision aimed to prevent the reveailing information that could potentially influence participants’ interpretations, particularly in the third scenario, where the objective was to maintain participants’ unawareness regarding the cause of the SAV’s sudden stop. As the three scenarios were combined into one video for each concept, the sequence of scenarios remained unchanged throughout the study.

There is no established and validated questionnaire specifically tailored to assess security concerns within the context of AVs. Security, is a multifaceted construct, including aspects of anxiety, threat perception, risk assessment, trust, and perceived control [77]. To address these dimensions, we incorporated these constructs alongside state anxiety, representing the corresponding emotional response associated with feelings of in/security. To gain further insights into participants’ sense of security, five self-constructed items evaluating attitudes towards SAVs were included using a 7-point Likert scale ranging from “strongly disagree” to “strongly agree”: “I feel threatened by other passengers on the automated shuttle bus.” (threat), “I trust the automated shuttle bus.” (trust), “I feel like I am in control of the situation when I am riding on an automated shuttle bus.” (control), “I think it is risky to ride on an automated shuttle bus.” (risk), “I do not feel secure sharing an automated shuttle bus with other passengers.” (security). Assessment of state anxiety within the presented scenarios was conducted using a validated 6-item short version of the State-Trait Anxiety Inventory State Measure (STAI-S) questionnaire [78], which we provided in the respective languages German, Spanish, and Korean.

Additionally, we used the standardized User Experience Questionnaire Short Versions (UEQ-s) [79] in German, Korean, and Spanish [80] to evaluate the UX of each of the concepts, as we expected that security and UX are intertwined, e.g. a feeling of insecurity will influence the UX in a negative way. After being presented with all three concepts, the participants were asked which one they liked the best and why. The questionnaire was provided in German, Korean, and Spanish.

Procedure

The questionnaire started with a welcome screen that included a briefing, a consent form, and an introduction to the topic of SAVs. Next, the participants were asked about their nationality. This question was asked at the beginning as based on this, the participant was later shown the Asian, European, or Hispanic-looking version of the HDC (“James”, “Bruno”, “Roberto”). The participants were then presented with the three videos (one video for each concept: baseline, RDC, HDC) each of which showed all three scenarios in one piece. We opted against showing each scenario-DC combination (baseline, RDC, HDC) separately, as our participants would have had to fill out the main part of the questionnaire nine times leading to cognitive overload and excessive time consumption. Still, with three variations of the situations we aimed to make our results more stable. After each video, the participants had to fill out the short version of the STAI-S to measure their level of anxiety, the five self-constructed items regarding the attitude towards SAVs, and the UEQ-s. Subsequently, the participants had to choose which of the three concepts they liked the best and why.

The study was concluded with demographic questions. As the level of a person’s anxiety influences the perceived security [77], we included six items on trait anxiety from a validated short version of the State-Trait Anxiety Inventory Trait Measure (STAI-T) [74]. Since current research is not conclusive on whether having experienced any sort of crime has an influence on perceived security [81], we did not consider this aspect. An overview of the study design can be seen in Fig. 6.

Fig. 6
figure 6

Study procedure (top) and sequence of videos and conditions (bottom)

Full size image

4 Results

For the quantitative results, descriptive and inferential statistics were calculated using JASP 0.16 [82] and mixed ANOVAs were conducted. Results are reported as statistically significant if \(p < 0.05\). According to Shapiro-Wilk’s Test, the measures did not follow a normal distribution. As there is no appropriate non-parametric test for the presented study design, a mixed ANOVA was applied, despite being a parametric test. In this regard, Blanca et al. [83] provide empirical evidence for the robustness of the F-test to non-normality under a wide variety of conditions. In addition, Levene’s test of equality of error variances indicated that the assumption of homogeneity of variances was violated, \(p =.05\). Mauchly’s test of sphericity revealed significant values, \(p =.05\), for the state anxiety, threat, trust, control, risk, security, and UX and a Greenhouse-Geisser correction was applied to these measures. In the following we will report significant Bonferroni corrected p values from our post-hoc tests.

Table 1 Descriptive statistics: Median (Mdn), Mean (M), and Standard Deviation (SD) for anxiety, threat, trust, control, risk, security for Colombian (COL), German (DE), and South Korean (KOR) participants
Full size table

First, we report the results for anxiety, then for threat, trust, risk, control, and security, and the overall UX. We finish with the digital companion preferences from the open-ended questionas at the end of the questionnaire. A more detailed comparison of the resulting scale means, as well as medians and standard deviations, are shown in Table  1 and descriptive plots including 95% confidence intervals of the results are shown in Fig. 7.

Fig. 7
figure 7

Boxplots of dimenstions anxiety, threat, risk, trust, control, and insecurit for the three study conditions (baseline (BL), robot-like companion (RDC) and the human-like companion (HDC) and the between-subjects factor culture

Full size image

4.1 Anxiety

Regarding the state anxiety values it is important to note that raw test values were utilized, ranging from 10 (indicating the lowest level of anxiety) to 80 (indicating the highest level of anxiety).

The average scores of the scenarios with RDC as well as the scenarios with HDC show lower values of anxiety than in the baseline scenarios. Post-hoc tests revealed significant differences in the comparison of baseline with RDC, \(t(df) = 4.33, p <.001, d = 0.36\), baseline with HDC, \(t(df) = 8.75, p <.001, d = 0.73\), and RDC with HDC, \(t(df) = 4.42, p <.001, d = 0.37\), with the HDC receiving better security scores than baseline and RDC. Interestingly, no significant differences in the preferences of the baseline or DCs were found between Germany and Korea, but between Colombia and Germany, \(t(df) = -5.19, p <.001, d = -0.85\), and between Colombia and Korea, \(t(df) = -5.14, p <.001, d = -0.99\). While all three countries feel the least anxiety with the HDC, this effect is higher in Colombia than in Germany and Korea.

4.2 Threat, trust, control, risk, and insecurity

When analyzing the results regarding the attitude towards SAVs, it is important to note that for the dimensions threat, risk, and insecurity low values are considered positive, while trust and control aim for high values.

The feeling of being threatened by other passengers on the SAV was lower with both DCs in comparison to rides without a DC. Significant differences were seen between baseline and HDC, \(t(df) = 5.41, p <.001, d = 0.42\), baseline and RDC, \(t(df) = 2.87, p =.013, d = 0.22\), and RDC and HDC, \(t(df) = 2.54, p =.035, d = 0.19\). Again, we found significant differences between Colombia and Germany, \(t(df) = -2.77, p =.019, d = -0.47\), and Colombia and Korea, \(t(df) = -2.72, p =.022, d = -0.54\), while Germany and Korea did not show any significant differences. The preference of a human companion is more pronounced among Colombians than among Germans or Koreans.

Trust was statistically significantly higher for the RDC, \(t(df) = -3.17, p =.005, d = -0.23\), and the HDC, \(t(df) = -5.88, p <.001, d = -0.43\), compared to the baseline and between the RDC and the HDC, \(t(df) = -2.72, p =.021, d = -0.20\). While Germans and Colombians rated trust higher for the HDC, Koreans preferred the RDC over the human one. However, we found no statistical differences between culture for the dimension of trust.

The dimension control of the situation was statistically significantly higher for both the RDC, \(t(df) = -5.82, p <.001, d = -0.50\), and the HDC, \(t(df) = -7.27, p <.001, d = -0.63\), in comparison with the baseline. There were no significant differences between the two companions. Colombians and Koreans felt more in control with the human companion while Germans rated the robot higher.

The feeling of risk was statistically significantly higher in the baseline condition when compared with both, the RDC, \(t(df) = 2.59, p =.030, d = 0.22\), and HDC\(t(df) = 3.04, p =.008, d = 0.25\). However, no significant difference was found between the two companions and between cultures. Yet, Koreans felt higher risk with the HDC than with the RDC.

With respect to insecurity, a trend can be seen towards a lower rating for the scenarios with DC than the scenarios without DC. Yet, we only found significant differences between the baseline and the HDC, \(t(df) = 3.72, p <.001, d = 0.31\). Germans and Colombians considered the HDC as more secure than the RDC, and Germans considered the RDC as insecure as the baseline. Koreans felt similarly secure with both companions.

4.3 User experience and overall preference

In terms of the UX, both DCs scored higher than the baseline regarding pragmatic, hedonic, and overall quality (see Fig. 8).

For pragmatic quality we found significant differences between baseline and RDC, \(t(df) = -2.25, p =.016, d = -0.19\), baseline and HDC, \(t(df) = -5.92, p <.001, d = -0.42\), and RDC and HDC, \(t(df) = -3.12, p =.006, d = -0.22\). The HDC received the highest scores. The differences between Colombia and Korea were significant, \(t(df) = 2.50, p =.040, d = 0.50\), whereas there were no significant differences between Colombia and Germany and Korea and Germany.

Regarding hedonic quality differences between baseline and RDC, \(t(df) = -3.14, p =.006, d = -0.24\), and baseline and HDC, \(t(df) = -5.21, p <.001, d = -0.39\), were significant. The companions were evaluated more positively than the baseline but their differences were not significant. However, Colombia showed significant differences in comparison with Germany, \(t(df) = 2.58, p =.033, d = 0.44\), and Korea, \(t(df) = 2.86, p =.015, d = 0.57\), as their more positive rating of the HDC had a bigger effect. No differences were found between Germany and Korea.

Fig. 8
figure 8

Boxplots of UEQ-s scale: overall UX, pragmatic UX, and hedonic UX (\(-\) 3 = low; 3 = high) for the three conditions (baseline (BL), robot-like companion (RDC) and the human-like companion (HDC) and the between-subjects factor culture

Full size image

As a result the overall quality of UX was perceived as significantly different between baseline and RDC, \(t(df) = 3.53, p =.001, d = -0.24\), baseline and HDC, \(t(df) = -6.63, p <.001, d = -0.46\), and RDC and HDC, \(t(df) = -3.09, p =.006, d = -0.21\), with the HDC receiving the best ratings. Germany did not show differences with Korea and Colombia but we found significant differences among Colombia and Korea, \(t(df) = 2.96, p =.011, d = 0.60\).

Consistent with the results from the previous dimensions, 72.1% of all participants liked the SAV with the HDC best compared to 16.9% who chose the RDC and only 11.0% who would rather have no companion on board (see Table 2). Interestingly, 27.6% Koreans liked the RDC best (see Table 3) which reflects part of our findings from the pre-study as well as their lower risk and higher trust and security ratings of the RDC. Germans were indifferent in their preference of baseline over RDC, and Colombians and Koreans seem to prefer having either a RDC or HDC over the baseline concept. Here the gap among Koreans is the largest with only 6.9% choosing to have no companion.

Table 2 Descriptive statistics: preferences for digital companions across cultures
Full size table
Table 3 Descriptive statistics: preferences for digital companions between cultures
Full size table

4.4 Understanding digital companion preferences

In the following, we present the synthesized analysis of free text fields regarding the most liked version of the DC. Topics are presented with their number of mentions (n) and number of Colombian, German, and Korean participants who mentioned them. All quotes were translated form the respective languages into English. The most important reason for preferring the HDC was its human likeness (27, 14 Colombians, 9 Germans, 4 Koreans) as it provided security (21, 14 Colombians, 4 Germans, 3 Koreans) and trust (16, 7 Colombians, 6 Germans, 3 Koreans): “It generates the sensation of accompaniment and being watched by a person, giving security and confidence.”, [Colombian participant]. Participants felt that the human version would be able to provide quick and adequate help (16, 7 Colombians, 7 Germans, 2 Koreans) in case of emergency: “You feel safer because you have the feeling that someone would really intervene in case of danger.” [German participant]. Participants indicated that they would feel more comfortable (8, 1 Colombians, 3 Germans, 4 Koreans) and monitored (8, 4 Colombians, 3 Germans, 1 Koreans) with a HDC.

On the other hand, participants considered the RDC to be cute (5, 2 Colombians, 1 Germans, 2 Koreans). Others mentioned that the RDC appeared as more neutral and not too personal (2, 1 German, 1 Korean): “robots are more neutral, objective, and advanced than humans” [German participant]. While one German participant preferred the RDC because the HDC seemed too unrealistic (“the human seemed to fake. With the robot at least you know it’s AI”, another German participant considered it to be too realistic: “The human companion would be a little too real for me. Here I would only feel like someone is staring at me the whole ride” [German participant].

The few participants who liked the SAV without DC the best gave different reasons for this. One commented: “I don’t think a companion other than a real companion would feel like a companion, so I think it’s better not to have one” [Korean participant]. Another one stated: “I don’t want interactions when I’m on the bus” [German participant]. A third person preferred the SAV without DC “because I do not believe that one of these digital helpers on the screen [will be able] to handle a really dangerous situation. On the [other] hand both torture [my] nerves” [German participant].

5 Discussion and implications

In our discussions of the results, we first address RQ.2, evaluating the efficacy of DCs in enhancing security within SAVs. Subsequently, we delve into RQ.3, exploring cross-cultural preferences for the human-like and robot-like DCs. Finally, we conclude by presenting recommendations for security measures that we derived from our study’s findings.

5.1 Enhanced ride experience with digital companions, human digital companions are prefered over robots in self-driving vehicles

In general, our results demonstrate notable improvements compared to the baseline when employing a DC. A DC inside the vehicle significantly reduces feelings of anxiety, threat, and risk while elevating trust, control, overall UX, as well as pragmatic and hedonic qualities. Specifically regarding security, significant effects were observed only with the HDC in contrast to the baseline.

Altogether, participants showed a preference for the HDC over the RDC, though statistically significant effects were established for anxiety, threat, trust, pragmatic UX, and overall UX. This inclination towards the HDC contrasts with the prior workshop findings, contradicting the anticipated preference for robots among Korean participants. Several factors may explain this contradiction between the main study and workshop outcomes. Primarily, research suggests that the appearance of DCs should align with their designated roles [28, 32], emphasizing the contextual relevance in their design. Our main study participants were presented with visualized SAV scenarios implemented in VR, enabling a better grasp of contextual situations compared to our workshop participants who solely relied on imaginative scenarios. The different contextual perceptions might have influenced the envisioned appearance of the DC. For instance, individual feedback highlighted a preference for the HDC due to its perceived authority. Workshop participants might not have considered this characteristic, given the absence of situations requiring authority. In contrast, our main study participants encountered a scenario where the DC demonstrated authority (scenario 2). This aligns with Johal et al.’s findings [32], suggesting that a cute appearance in robots might hinder acceptance in roles requiring authority.

Given the nuanced perceptions influenced by contextual experiences, future DC designs should account for the roles they undertake. Incorporating contextual scenarios or simulations could offer a more holistic understanding of the DC’s potential functionalities and appearances. Furthermore, designing DCs with adaptable appearances or behaviors that align with varied situations may enhance their acceptability across diverse cultural preferences. A balance between perceived authority and approachability is crucial, suggesting a need for multifaceted designs that exude both competence and warmth. Integrating these findings into future co-design workshops or participatory design sessions may yield more comprehensive insights into DC preferences and enhance their acceptance in diverse cultural contexts.

5.2 Implications for digital companions including cultural differences

The workshop results suggest a consensus between Koreans and Germans regarding the desired traits of a DC for SAV— both cultures value a friendly and sympathetic one. However, distinctions emerge in their preferences concerning the hierarchical position of the DC in relation to humans: Koreans lean towards a DC that displays obedience to humans, whereas Germans favor a supportive, reliable, and serious DC. This contrast could stem from cultural differences delineated by Hofstede [84], where Korea tends to uphold a slightly hierarchical societal structure while Germany gravitates towards lower power distances as discussed in Chapters. The inclination in Korean society towards hierarchy may underlie the preference for a DC subordinate to humans which aligns with the findings of Lee and Sabanović [67] asserting Koreans’ preference for submissive robots. Conversely, Germany’s pursuit of egalitarianism may manifest in a DC perceived as an equal, existing to support humans. Moreover, Germany’s emphasis on punctuality and discipline [85] may drive the desire for a reliable and serious DC.

Although Koreans displayed a preference for the HDC, they expressed greater openness and trust toward robots compared to Colombia and Germany. Noteworthy differences emerged in anxiety, perceived threat, and hedonic quality, with significant differences between Colombia and Germany, as well as Colombia and Korea, but not between Germany and Korea. Differences were also significant between Colombia and Korea concerning the pragmatic and overall UX. Hofstede’s cultural dimensions [84] suggest a closer alignment between Colombia and Korea than either between Colombia and Germany or Korea and Germany. Surprisingly, Colombians favored the RDC less than both Germans and Koreans when compared to a human-like DC. This preference might be attributed to Colombia’s high power distance, potentially leading to the perception of the HDC as more authoritarian than the RDC. Conversely, the qualitative interviews preceding our main study and related research suggest that human drivers might not be viewed as equally trustworthy in Colombia as they are, for instance, in Germany, potentially influencing their preference for the RDC over the HDC.

This contrast becomes more intriguing when considering Colombia’s apparent lesser openness to technologies, potentially influencing participants to opt for a concept closer to today’s PT. This finding gains significance when contemplating more abstract and disembodied forms of a DC prompting inquiries about the potential relationship between the level of anthropomorphization and Colombians’ acceptance of a system accommodating human drivers. Furthermore, it raises questions about whether Koreans might be more inclined toward adopting object-like solutions. Contrarily, qualitative insights underscore that the preference for the robot-like version aligns with the futuristic essence of self-driving shuttle buses, suggesting that as time progresses, robots may amass more trust among users.

These findings suggest that the design of a DC should consider cultural nuances and should aim for a balance between familiarity and futuristic elements to cater to varying cultural predispositions. One solutions could be adaptable interfaces allowing users to personalize their DC’s appearance and interaction style. For instance, a modular design could offer varying levels of anthropomorphism, allowing users from different cultures to choose between human-like or abstract representations and interfaces could allow users to adjust facial features or choose between humanoid or abstract avatars. The DC behaviors and dialogue scripts should be adaptable and align with specific cultural preferences and language variations, gestures, and conversational styles could be tailored to each culture. The DC’s responses and mannerisms could be adjusted to match the cultural norms of respect and authority in high power distance cultures, while maintaining a friendly demeanor.

Despite that, we recommend to implement gradual familiarization strategies for the introduction of futuristic elements. For example, the DC could start with more familiar attributes, such as resembling a kiosk or a familiar object, gradually integrating humanoid traits as users grow more accustomed to automation. We propose to design the DCs with adaptive learning capabilities to understand user preferences over time. These companions could evolve their behavior and appearance based on user interactions and feedback, gradually gaining trust and acceptance. An example might include a DC learning user preferences for safety announcements or personal assistance.

5.3 Comprehensive security measures beyond user interfaces in shared automated vehicles

While prioritizing functionalities for security and comfort in DCs is essential, it is crucial to note that some participants emphasized limitations. For instance, despite suggestions to include emergency buttons and interactive screens for security updates, a subset of participants expressed concerns that a user interface alone might not suffice in critical situations or emergencies. Their feedback underlines the need for multifaceted solutions that extend beyond interface design, possibly including physical emergency mechanisms or direct communication channels with emergency services integrated into the vehicle itself as proposed by Schuß et al. [10]. This insight highlights the importance of integrating diverse safety measures and ensuring redundancy in emergency protocols, going beyond interface-based solutions to enhance real-time response capabilities. Thus, physical emergency mechanisms need to be integrated within the SAV itself, such as easily accessible emergency stop buttons or handles that directly alert the vehicle’s control center or emergency services. The SAVs should be equipped with integrated communication systems allowing passengers to directly contact emergency services or the vehicle’s control center in urgent situations, such as medical emergencies or unusual situations, ensuring swift response and assistance. These features could be supplemented by integrating biometric identification features like fingerprint or facial recognition, and incorporating remote monitoring capabilities for swift intervention in suspicious situations. Sensor-based technologies within the vehicle that can detect anomalies or distress signals among passengers, triggering automated alerts for immediate attention. Automated protocols within the SAVs could respond to specific emergency situations, such as automatic rerouting to a safe location or activating emergency lighting and audio signals to attract attention. Predefined protocols could be established to guide the DC’s response in threatening situations, such as alerting other passengers, locking doors, or initiating emergency procedures. We argue that Integrating a combination of these measures alongside UI functionalities ensures a robust security ecosystem within SAVs, addressing the participants’ concerns and enhancing overall emergency preparedness and response capabilities. In this context, concerns about security, privacy, and employment implications SAVs were previously discussed [86] and addressing participants’ simultaneous privacy concerns, balancing security and privacy demands presents a significant challenge for the design of future systems.

6 Limitations and future work

Our main study was conducted using VR scenario videos, which offered a good understanding of the situations but cannot fully replicate real-world interactions. Participants’ preferences might differ when they actually engage with the DCs in real-life situations. Each participant in the main study viewed one RDC and one HDC, so preferences might be influenced not only by the robot-like or human-like nature but also by individual preferences for the specific DCs they encountered. While our main focus was on cultural differences in DC preferences, we did not explicitly consider gender differences, which could play a significant role, especially within different cultural contexts. Future work could look into this aspect to provide a more nuanced understanding. The statistical significance of our findings is limited due to the sample size of 125 participants across three cultural groups. Future studies should aim to include a broader range of age groups, encompassing categories like school kids, college students, people from different socio-economic statuses and education levels, and older adults.

This work draws upon our previous research from Schuß et al. [10, 31] where a (human) teleoperator on a screen was proposed to account for security issues that might arise due to the absence of a driver in SAVs. Future research should explore diverse implementation formats for DCs beyond our artificial teleoperator visualized on a large display inside the vehicle. Investigating private implementations, such as on personal devices like smartwatches or smartphones, presents an intriguing avenue. Regarding DC appearances, our study specifically adopted a human-like representation, yet alternative approaches, including non-human appearances, warrant exploration. Additionally, further investigations into DC behavior across a spectrum of SAV scenarios are vital. Understanding preferences and requirements across everyday situations, security-critical scenarios, and emergencies is crucial in meeting the needs of potential future passengers.

In our study we did not address specific design considerations such as the placement of display for the Avatar and microphone/speakers for communication. This is significant, as ensuring seamless interaction between the DC and passengers, while minimizing interference from co-passengers, is crucial for enhancing user experience and acceptance. Future research should consider this aspect.

We deliberately stayed with the term digital companion during the co-design workshops as we wanted to investigate how this particular form of a UI for SAVs should look like from our participants’ perspective. While this term directed participants towards an embodied representation, using a more technical term like “system” might have steered towards an abstract concept of a DC. Echoing related studies emphasizing anthropomorphization in designing caring conversational agents (CAs) [28], we contend that these traits hold particular importance in addressing security needs. However, research such as that by [87], which utilized a peripheral robotic object, suggests that more open approaches in co-designing DCs for SAVs could yield equally promising yet abstract designs. This presents an avenue worth exploring in future research endeavors.

Although Colombians were not part of the workshops, our interviews provided valuable insights for DC conceptualization, suggesting that our work contributes to understanding preferences and requirements for DC appearance. The evolving landscape of mobility, particularly with SAVs, lacks concrete solutions, indicating a significant need for further research in this area. Moreover, a cross-cultural approach remains vital to cater to diverse needs in future mobility solutions as discussed in Sect. 2.

Given our small sample size and exclusive focus on German participants, the outcomes lack generalizability. The results from Chapter 5.2 have indicated statistically significant differences in DC preferences among Asian, European, and Latin American participants, suggesting the need for wider participant diversity encompassing various gender identities, cultural backgrounds, and educational levels. Consequently, we advocate replicating this study with more diverse participant groups.

Additionally, further investigations into DC behavior across a spectrum of SAV scenarios are vital. Understanding preferences and requirements across everyday situations, security-critical scenarios, and emergencies is crucial in meeting the needs of potential future passengers.

7 Conclusion

This chapter delves into the design aspects of a CA, a digital companion intended to enhance security and UX for passengers using SAVs, focusing on the diverse cultural contexts of Colombia, Germany, and Korea. Through interviews and co-design workshops with participants from these cultures, we established specific requirements for the implementation of these digital companions within the realm of automated shuttle buses. Two distinct concepts emerged from these discussions: a human-like (HDC) and a robot-like companion (RDC).

These concepts were converted into prototypical designs, resulting in three culturally adapted versions of a HDC (“Roberto”, “Bruno”, and “James”) and one RDC. These designs were evaluated in a comprehensive main study (N = 136) employing an online questionnaire that featured VR scenarios presented as videos. The findings suggest that the presence of a DC notably enhances the sense of security and overall UX for Korean and German passengers traveling in SAVs. Interestingly, Colombians, Germans, and Koreans exhibited particularly positive perceptions of the human-like companion, with Colombians explicitly favoring it and Koreans demonstrating the highes openness towards the robot-like version.

Our study contributes at the intersection of SAV security, digital companions’ role in enhancing security, and cross-cultural preferences for such companions. Based on our findings we advocate for the human-like companions as a promising design approach to elevate perceived security within SAVs, ultimately augmenting experiences. Additionally, these companions hold the potential to address challenges associated with the absence of a human driver in these vehicles. At the same time it is central to consider hierarchical structures and cultural nuances, as demonstrated by differences in preferences between cultures when designing DCs. The hierarchical nature of certain societies, as observed in Korea, possibly influenced preferences for more obedient companions. Conversely, societies valuing equality and reliability, like Germany, may prefer companions as equals. These distinctions highlight the importance of tailoring companion designs to align with diverse cultural perspectives, emphasizing the significance of cultural adaptation in creating universally effective DCs for SAVs. In summary, our study introduces novelty in two key aspects. Firstly, we propose DCs within SAVs to address security concerns and evaluate them. Secondly, our cross-cultural approach fills a significant research gap, allowing for a comprehensive understanding of user preferences and perceptions across diverse cultural contexts.

8 Editorial Policies for:

Springer journals and proceedings: https://www.springer.com/gp/editorial-policies

Nature Portfolio journals: https://www.nature.com/nature-research/editorial-policies

Scientific Reports: https://www.nature.com/srep/journal-policies/editorial-policies

BMC journals: https://www.biomedcentral.com/getpublished/editorial-policies