Designing Future In-Vehicle Assistants: Insights from User Imaginations and Experiences

With the integration of electrification, intelligence, and autonomous driving technologies, personalized intelligent vehicle agents (IVAs) have gradually become the focus of innovation. However, research on the relationship between intelligent in-car interaction initiated by IVAs and user characteristics is still insufficient. This study explores how user characteristics affect their expectations of IVAs’ interactive scenarios, functions, appearance, and social attributes. Through a questionnaire-based pilot study, as well as a think-aloud experiment and semi-structured interviews based on an experimental car, we evaluated different users’ expectations for IVAs’ interaction on their characteristics and summarized expectations into four categories: scenarios, functions, appearance, and social attributes. The results show that users’ characteristics lead to changes in their preferences for the physical appearance and social attributes of IVAs. Based on these findings, we propose future IVAs’ interactive design, emphasizing the importance of multi-modal interactions based on smart materials and personalized and implicit interactions based on artificial general intelligence (AGI).


INTRODUCTION AND BACKGROUND
In recent years, the integration of electrification, intelligence, and autonomous driving technologies has propelled a revolutionary shift in the automotive industry.This transformation is not only evident in technological progress but also in the redefinition of mobility and human-computer interaction(HCI).The field of Intelligent Vehicle Agents (IVAs), in particular, has become a focus of innovation, with rapid advances in sensing, artificial intelligence, and interaction technologies, which academia and industry believe have the potential to reshape the way consumers interact with their vehicles [13].Despite some achievements in autonomous driving [4], cognitive assistance [27], and affective interaction [15,21], the integration of IVAs in the in-car interactive design is still in its early stages.Based on the existing literature, we concentrate on IVAscentered in-car intelligent interaction and further explore how HCI researchers and designers should design future in-car interactions according to the diverse user needs and characteristics.A specific gap emerges in the absence of research illuminating the relationship between in-car intelligent interaction designs initiated by IVAs and user characteristics.To fathom the user characteristics and requirements relevant to IVAs' interactive design, we must understand the main design modules of IVAs in the in-car interaction.As the primary interface of IVAs' interaction, IVAs' visual expressions typically fall into five categories: humanoid, mythical creature, animal, robot, and abstract entity [3].However, the relationship between these forms and users' characteristics across various interaction scenarios or functions still requires further discussion [12].On the other hand, recently, IVAs have instigated an escalating number of in-car multimodal interactions.It is still unclear how users' needs and characteristics affect the way IVAs provide interaction and social attributes.With these evolving trends, we aim to unveil the relationship between user characteristics and their shaping of IVAs' interactive scenarios, functions, appearance, and social attributes.To address the above problems, this paper embarks on probing users' perceptions and preferences for IVAs through two experimental studies.Initially, a quantitative pilot study involving 161 participants extracts preferences for IVAs' visual expressions and interactions across diverse interaction scenarios.Subsequently, a think-aloud experiment engages 12 participants in semi-structured interviews conducted within an authentic vehicular setting, unraveling user requirements and expectations regarding in-car intelligent interactions instigated by IVAs.By analyzing quantitative and qualitative experimental data, we hope to propose preliminary insights into designing future IVAs based on user characteristics, offering diverse interactive scenarios, functions, appearance, and social attributes.
The core contributions of this work are as follows: • This study is the first to explore the impact of user characteristics on the design of IVAs in terms of interactive scenarios and functions, and appearance and social attributes.Our study extends the HCI community's understanding of how users with different characteristics expect future IVAs.Our study shows that users' long-term and short-term characteristics lead to changes in their preferences for the appearance and social attributes of IVAs in different scenarios and functions, which will provide guidelines for the next generation of IVAs to personalize interactions for users.• Existing in-car interaction modalities will be challenged.This research helps address the limitations of existing in-car interaction modalities and discusses potential directions for future IVAs based on our results.• This study provides a comprehensive framework for understanding user behavior and preferences.It helps in the initial exploration of IVAs' design in actual environments, providing valuable insights for researchers, designers, and engineers working in the HCI community in automotive and related fields.

RELATED WORKS
Recently, we have seen a trend towards promoting in-car interactions in the form of interactive terminals such as IVAs [13].IVAs are viewed by researchers as the embodiment of computerized agents in in-car interactions.In academia, the study of the function and attributes of IVAs has always been a trending topic.These intelligent agents can assist in executing driving or non-driving tasks, which were traditionally performed by drivers and passengers.However, academics have argued that people expect to see more humanizing qualities in intelligent agents than just tool-like attributes.From this, academia has deduced four categories of research objectives for IVAs: Driving Task, Cognitive Process, Affective Process, and Overall Experience and Subjective Evaluation.Although this classification serves as a construct for understanding the utilities of IVAs, these categories are not mutually exclusive.This is because the tasks and processes that a driver engages in are interrelated.Specifically, Driving Task refers to the functions of the IVAs during driving, including assisting the driver to understand the functions and characteristics of the system, as well as interpreting the system status and driving behavior intentions of the autonomous vehicle.The research on these tasks aims to decrease the burden of driving tasks and support drivers in fulfilling tasks, including advanced detection, notification of upcoming events, and route planning.For example, Antrobus et al. [1] and Hofmann et al. [6] have investigated the effectiveness of IVAs as a driving partner to decrease the burden of driving tasks.IVAs also support non-driving-related tasks to achieve a driver's safe status [6,26].Cognitive Process: This refers to how IVAs influence the cognitive processes of drivers, and how they assist drivers in processing and understanding information in the driving environment.The research on these tasks aims to minimize driver distractions, maintain and awaken the driver's attention toward the driving situation, reduce drivers' fatigue, and increase situational awareness.For instance, Jeon et al. [8] and Jonsson et al. [9] have attempted to reveal the direct effects on the cognitive status of the driver and process without conducting or supporting the tasks in place of the drivers.Affective Process: This refers to how IVAs influence the emotional processes of drivers, and how they assist drivers in processing and understanding emotional responses in the driving environment.Researchers have tried to investigate the applicability of IVAs as a regulator of the drivers' affective status and process.Many of them have focused on mitigating the drivers' negative affective states, particularly anger, which is a widely known risk factor hindering driving safety.Overall Experience and Subjective Evaluation: This refers to how IVAs can be designed to improve the overall driving experience for drivers and passengers, as well as their subjective evaluation of the vehicle and its interface.This includes IVAs assisting the driver in understanding the functions and characteristics of the system, interpreting the autonomous vehicle's system state and driving behavior intent, and enhancing the user's positive experience (such as trust, acceptance, and subjective perception).They help to understand system functions and features and interpret the system state and intent of IAV's driving behavior.For example, Hauslschmid et al. [5] and Koo et al. [10] emphasized the role of IVAs in conveying and interpreting system states.These academic advances and research directions provide important insights into the function and attributes of IVAs.However, it is worth noting that most of these studies are based on traditional in-car environment interaction.Currently, when in-car interactions are becoming increasingly intelligent, how to meet the changing needs of users in multi-modal and personalized interactions, and how to conduct in-depth and systematic research on the design of IVAs characteristics is still an issue that has not been fully explored.

PILOT STUDY
We conducted a pilot study to investigate the preference for the basic model and graphical styles of IVAs' design among users in different interaction tasks.The study was conducted through a questionnaire that covered users' basic information, and preferences for different IVAs' designs in four interaction tasks.

Interaction Task Simulation
According to the previous study [13], we categorized in-car interaction into four types of tasks: Driving-Related Tasks, Affective Processes, Cognitive Interaction, and Overall Experience.The users were asked to read the description of each task, rating five basic models of IVAs, and then selecting their favorite graphical styles for these IVAs models.The settings of the pilot study are as follows: 1. Description of Each Task: • Driving-Related Task: Please imagine that you are sitting in a car with a driving assistance system, and the intelligent vehicle agent (IVA) on the car communicates the vehicle's condition in time.You do not need to control the steering, throttle, and brakes, and you can even carry out simple personal activities in the cabin.

Procedure
As shown in Fig. 1, in the pilot study, the procedure included three steps: a) Users were required to provide demographic data, B)read the task script (as shown in 3.1) to understand IVAs' interaction, C)and select preferred IVAs' forms based on these tasks.After collecting the questionnaire results, we analyzed various user groups' preferences for IVAs in different interaction tasks.
• Basic Model: As shown in Fig. 2, we found that older participants preferred the human-like IVAs to the preference of younger participants.Two-thirds of users over 40 years old chose human-like IVAs for driving-related tasks.On the other hand, young people have a more even preference.• Graphical Style: As shown in Table 1, users tend to have the same graphical style preferences for IVAs with different appearances in different tasks, which may mean that the future IVAs' design does not need to change the style according to the change of the scene.
Through the pilot study, we found that young people seem to have very different preferences for the basic model of IVAs in various tasks.These different preferences may provide valuable design directions for future personalized IVAs.However, since our experiment did not conduct further user interviews, we are unable to discuss in depth the specific reasons behind these different preferences and the expectations of users for IVAs.On the other hand, since the entire IVAs scene is only given by the simple text description in the questionnaire, it is difficult to ensure that all users are involved in the scene to answer.So we rented an authentic car with IVAs as the experimental car and conducted an exploratory study to further explore our question.

DESIGN EXPLORATION
Based on the results of the Pilot Study, we conducted a think-aloud study based on an authentic car.The experimental car is a new sedan introduced in the last few years, which includes voice recognition, gesture control, and touchscreen technology, as well as advanced IVAs technology.It allows the driver to control multiple functions of the vehicle through simple interaction with IVAs, from navigation and media.The experimental car also features an advanced driving assistance system, which helps the driver maintain a safe distance from vehicles in different traffic conditions, maintain lanes, and assist in driving.The experimental car has a certain level of interactivity that provides the basis for encouraging the user to imagine the future IVAs in our experiment.

Experiment Setup
The think-aloud experiment was conducted in an authentic car.
Interviews were carried out in the car.To ensure the experiment's safety and reliability, the vehicle was parked in an underground garage.Two research members experimented: the principal investigator was in charge of procedural guidance and interviews, while the assistant investigator handled video and audio recordings.
The experiment commenced upon entering the underground garage, with the assisting investigator initiating video and audio recording to continuously document participant reactions.Once inside the experimental car, the assisting investigator sat in the rear seat to limit the disruption to the conversation between the principal investigator and the participant.
The driver-participants were asked to personally drive the experimental car out of the parking spot, navigate around the garage, and return, after which they were interviewed while seated in the driver's seat.For passenger-participants, the principal investigator drove the experimental vehicle following the same route, while the participant was seated in the front passenger seat and interviewed after parking.
We obtained qualitative experimental data in the form of semistructured interviews to explore participants' user needs and their expectations for in-car interactions of future IVAs.

Participants
We created a recruitment poster detailing the research questions, objectives, compensation, and contact information.We distributed it through WeChat and websites to attract participants.To categorize participants effectively, we asked them to specify whether they were drivers or passengers when applying.
As shown in Table 2, 12 participants were recruited, including 6 drivers (with substantial driving experience) and 6 passengers (with little or no driving experience).The participants' ages ranged from 23 to 33, with 7 males and 5 females.Their occupations included company employees, academic researchers, and students.They originated from various provinces in China, but all were currently residing in Beijing.

Procedure
Participants were first introduced to the experiment procedure in the reception area, where they signed an informed consent form and filled out a basic information questionnaire.After preparation, the two investigators led participants to the garage.Driver-participants received the keys on their way to the garage, while passengerparticipants were led onto the vehicle by the principal investigator.
As shown in Fig. 3, besides the pre-screening session, the in-car experiment was comprised of three main parts: Using the "think aloud" method (where participants verbalize their thoughts in real-time while performing tasks [28]), participants completed tasks relevant to their roles as either driver or passenger.Tasks for drivers included: driving out of the parking spot and around the garage, activating navigation to the airport, reversing back into the parking spot, querying and reporting vehicle status (fuel level, speed, RPM), finding nearby parking, and playing specific music.Tasks for passengers included: playing specific music, conversing with the voice assistant of the car, locating nearby recommended restaurants or attractions, and adjusting the seats.questions in the interview are all related to the design of IVAs' interactive scenarios, functions, forms, and social attributes.Open-ended questions encourage participants to provide detailed and personal answers, helping researchers explore the motivations, emotions, values, and other factors behind different users' responses.For the "experience interview", the principal investigator asked questions based on observed issues during participants' task completion, aiming to understand the causes of operational difficulties and encouraging participants to express expected IVAs' functions.In the views on methods of IVAs and multimodal interaction section, the experimenter inquired about participants' opinions on IVAs' basic settings, position, interactive settings, scenarios, and requirements for personalization and exclusivity.The "personal attitudes" section explored participants' acceptance level of new technology and character traits to aid the research team in designing user personas.

Free Exploration.
Participants were encouraged to freely explore the experimental car using the "think aloud" method and describe their process.

Data Analysis Method
All think-aloud interviews were conducted in Mandarin and recorded by the assistant.The recordings were then transcribed verbatim into Chinese characters using voice transcription software.For qualitative analysis, we employed thematic analysis.To ensure coding reliability, two team members independently encoded the recorded material.A group of four team members conducted multiple rounds of comparison and discussion to agree on the initial coding structure.Subsequently, the two coders recoded the material based on this structure to ensure nothing was overlooked.After several iterations, the research team identified four key design elements of IVAs, each with multiple layers of logic.To protect participant identities, this report uses 'D' before participant numbers for drivers and 'P' for passengers.

Users' Expectations for Future IVAs
According to user interviews in the think-aloud study, we found that users' expectations for future IVAs can be summarized into four categories: scenario, function, appearance, and social attributes (Fig. 4).Each of them also subdivides multiple subcategories to provide a more comprehensive image of future IVAs.Crucially, these four categories also exhibit interconnections, contributing to a holistic understanding of user expectations.

Scenario and Function.
Based on the insights garnered from user interviews in the think-aloud study, a distinct pattern emerged, categorizing users' anticipated applications of IVAs into four daily scenarios: rest, regular commute, travel, and traffic jam.
For example, in the context of relaxation, users foresee IVAs contributing to their repose by facilitating seat adjustments and introducing pleasant scents.During regular commutes, IVAs are expected to alleviate monotony by broadcasting news and initiating engaging conversations.For travel scenarios, users envision IVAs livening up the atmosphere and capturing joyful moments.In the context of traffic jams, users anticipate IVAs playing light-hearted jokes and videos to divert attention from negative emotions like anxiety.
Notably, we observed an escalating inclination among users for IVAs to shape the overall in-car atmosphere under distinct scenarios.For instance, D1 anticipates IVAs fostering a sleep-conducive ambiance through music, temperature adjustments, and fragrances during relaxation periods.Conversely, P11 envisions IVAs sustaining an energetic atmosphere during prolonged road trips and offering emotional solace during traffic jams.This evolving demand for tailored atmospheres might be attributed to an overarching shift wherein users increasingly desire IVAs to enable interactions encompassing the entirety of the in-car space, transcending the traditional user-device interaction paradigm.As succinctly expressed by P9, "I envision the car's ambiance responding to me, creating a sensation that IVAs embody the car itself, orchestrating an interplay of the car's light and sound effects to engage with me." Interestingly, this interaction requirement for the overall scene also makes some users expect IVAs to have stronger multi-modal data recognition capability.As P2 says, "I want IVAs to recognize my emotions at any time, so it can freely change the light and sound effects in the car to match my needs in these situations." On the other hand, there has been a notable surge in users' inclination towards seeking entertainment and engagement from IVAs across diverse scenarios.This inclination appears to stem from an escalating duration of time spent within cars, prompting a growing desire to transform what might otherwise be perceived as unproductive time into meaningful interactions with IVAs.Drawing from our subjects, it's postulated that a significant number of urban commuters, particularly those in their thirties, allocate a substantial portion of their daily routine-sometimes hours-to commuting, whether as drivers or passengers.This protracted commuting time fuels a yearning to engage with IVAs so as to achieve additional functions or emotional satisfaction during this period in their daily driving.To exemplify, D1 articulated, "I spend three hours commuting every day, and it feels like a waste of time.I wish IVAs could arrange suitable podcasts and audiobooks for me to make these hours more meaningful." Additionally, P7 shared her desire for more emotional interactions with IVAs during these scenarios and periods, such as chatting, and teasing, to pass the time.
Moreover, our observations illuminated that users would put forward some interesting views on the specific interaction functions of IVAs in different scenarios based on some of their characteristics.For instance, most drivers have expressed their opinions on IVAs' Drive-Related Function.While most expressed a desire for IVAs to assist and optimize their driving behaviors, many exhibited hesitancy, even during the conceptual phase, to fully trust IVAs to drive.This signifies that the promotion and application of IVAs related to driving might require a phase of technology dissemination.As D4 said, "I frequently use the automatic drive in my car because it's widely used and mature, but I wouldn't allow IVAs to take over driving outright." Conversely, they made it clear about their aspiration for IVAs to invigorate their alertness when it detects signs of driver fatigue.As D5 said, "I hope that when IVAs detect prolonged hours of driving, its appearance becomes livelier and more vibrant to awaken the driver's attention and ensure vigilance." Indeed, it suggests that drivers' needs for interaction features in driving scenarios are often closely tied to safety.Additionally, drivers remain sensitive and conservative about any interaction that is beyond their control, such as automatic driving.
For Non-Driving-Related interactions, both drivers and passengers exhibited a more adaptable and open-minded stance toward the design of IVAs' features.Interestingly, in various scenarios, many users indicated that their functional needs for IVAs did not pursue adding explicit smart devices such as displays.Instead, they sought subtle yet caring interactions from IVAs, delivered succinctly and implicitly.D8 elucidated, "I'm not fixated on cramming high-tech elements like large screens into my car.The essence of my car experience lies in the ride and driving experience." Echoing this sentiment, P12 expressed, "I don't gravitate towards overly extravagant features like pervasive electronic screens.My preference for IVAs is its ability to fulfill functions and cater to users using restrained resources." This suggests a new direction for future IVAs' interaction.When the ever-increasing display no longer constitutes the primary user pursuit, future IVAs beckon the exploration of alternative intelligent hardware to enrich the in-car experience.

Appearance and Social Attributes.
Through the thinkaloud study, we also discerned that users' preferences for IVAs' appearances are related to the social attributes that IVAs present during the interaction.Previous research has indicated that participants interacting with IVAs exhibiting distinct appearances might perceive different social attributes [24].In many HCI studies, social attributes are usually considered to refer to social traits such as warmth and capability exhibited by IVAs during the interaction process [11].Although these qualities are more of a dynamic feeling that occurs in an interaction, they play a crucial role in the user's judgment of the intelligence displayed by the interaction object and the interaction experience.Our study has the same findings as the pilot study, it remains challenging to identify a single IVAs visual expression universally suitable for a given scenario.We find that users' preferences for certain IVAs' visual expression will change with different scenes.For instance, users favoring animal-like IVAs place greater value on the warmth of IVAs.As articulated by P12, "Abstract IVAs lack emotional resonance for me, while human-like and robot-like IVAs are more professional and distant.I prefer cute IVAs, such as a puppy." Similarly, users inclined toward human-like IVAs with a professional appearance assign a higher value to IVAs' capability and interaction efficiency.As P9 said, "Interacting with human-like IVAs feels akin to conversing with a human.It has vitality, and I can communicate with it more naturally." Moreover, within the distinctive sphere of in-car interactions, IVAs' appearance encompasses not only basic visual expressions but also design elements like changeability and mobility during interactions.Acceptance of IVAs' changeabilty predominantly correlates with users' personalities.Open and extroverted users frequently seek changeability in IVAs to introduce novelty to interactions.P10 emphasized, "I'm more open to new technology, so I expect IVAs to be a little more variable.While on autopilot, it could transform into a professional-looking human-like assistant.When playing music or radio, IVA becomes a cute puppy.Despite these changes, I still recognize them as the same agent." Others try to understand IVAs as flexible and realistic personification characters who can change accessories with the setting.As P11 said, "I would prefer IVAs to maintain its core image but modulate its voice tone based on the passengers' identities.For instance, when there are children or elderly people in the car, IVA's tone is gentle." However, introverted and sensitive users lean towards IVAs retaining a consistent form to provide inner stability.P2 articulated, "Cats hold a special place in my heart, and I desire my IVA to embody a kitten perpetually.If it is a physical IVA, I would hold it and garner a deep emotional connection to internally stabilize myself." These results indicate that in future IVAs' interactions, in addition to maintaining the appearance that meets the needs of the brand, designers must also consider users' long-term characteristics such as their personalities, to provide users with a more immersive interaction experience.
Another interesting social attribute that users focus on is IVAs' identity cognition within the interaction process.With the development of technology, the user's expectations for future IVAs have gradually evolved from the original voice-based communication to visual engagement, it is plausible that multi-modal interaction technology could enable IVAs to traverse the entire vehicular space.Consequently, diverse perceptions of IVAs' identity have arisen, encompassing two perspectives: IVAs as a communication intermediary between users and the car, or IVAs as a direct representation of the car itself.The former viewpoint implies users communicate their needs to IVAs, which then prompts the car to perform corresponding functions, which represents the prevailing form of IVAs in today's automobile market.For example, D5 believes that IVAs is an independent product in the car, similar to the smart speaker in the car, conveying the needs of the user to the car.D6 perceives IVAs' temperament and style as aligning with the car, yet believes its essence remains distinct from the car-an agent merely conveying instructions.In contrast, the latter perspective envisions direct user-vehicle interactions, representing the future ideal for IVAs' development.These advanced IVAs would engage users in real-time through AI, multimodal interfaces, implicit interactions, and other technologies.D1 envisages accomplished IVAs to simulate interaction with the car, infusing it with vitality.P12 positions IVAs as the car's cognitive center, capable of evolving.When people change a new car, IVAs, or the brain also migrate together, deepening the user's bond with the new car.D8, conversely, perceives IVAs as an extension of the car.Despite possessing an independent image, IVAs' actions and words still symbolize the car's interactions with users.These perspectives are similar to the explicit and implicit interaction paradigms in HCI [23].As shown in Fig. 5, in the past, limited by in-car interaction technology, most interactions around IVAs were carried out explicitly, such as users instructing IVAs to play music.Insights from our study suggest that the upcoming generation of IVAs interactions may heavily feature implicit interactions, enabling IVAs to interact directly with users, transcending the "messenger" role between user and car.

Impact of User Personalized Characteristics on IVA Design
Through qualitative analysis of users' expectations for IVAs, we note that when users' expectations for future interactions become more personalized and implicit, it is difficult for future IVAs design to adhere to the conventional approach of creating a one-size-fitsall IVA based on our early user research [3].Instead, designers must increasingly account for users' short-term and long-term attributes, which can affect the implementation of potential incar interactions, before and during car usage.As shown in Fig. 6, the study outlines a preliminary design framework rooted in the relationship between user characteristics, revealing insights into IVAs' interactive scenarios, functions, appearance, and social attributes.
From think-aloud interviews, it is evident that diverse users' needs for personalized IVAs' interactions predominantly stem from two types of characteristics: short-term characteristics (e.g., changes in the user's real-time emotional state.)and long-term characteristics (e.g., the user's long-term view on the perception of things).Among them, changes in short-term characteristics will have more influence on how IVAs are specified when designing IVAs to determine the specific interaction scenarios and functions required by users.On the other hand, the long-term characteristics will have more influence on how to make IVAs provide interaction appearance and social attributes that meet users' needs when designing IVAs.

5.2.1
Short-term Characteristics.The influence of short-term characteristics on personalization design is graphically represented in Fig. 6.Our findings indicate that numerous users, both drivers and passengers, disclosed how their real-time emotional fluctuations while in the car directly impact their desired scenarios and functions for IVAs.For instance, D3 said that he would feel irritable and angry when he was stuck in traffic, and he hoped that IVAs could perceive his emotions and employ interaction to pacify his feelings.D1 and D4 noted their boredom during commutes, particularly when waiting for traffic lights, and hoped IVAs could engage in conversations to uplift their spirits.This demand shows that when users are in a low mood, they expect IVAs to optimize their emotional experience through chat, companionship, and other functions based on the scenario.As articulated by P12, "For a familiar car, I'll develop an emotional bond with IVAs.When I am overwhelmed by work pressure, I hope to get reassurance from IVAs." On the other hand, when users experience positive emotions, IVAs can enhance the overall pleasant experience of driving and riding through empathy and other ways.As D4 suggested, "If IVAs can actively adjust the seat, temperature, and other riding environment to my usual mode upon entering the car, it will make me feel very comfortable" Furthermore, P11 proposed that IVAs could contribute to harmonious multi-person travel scenarios by participating in conversations to create a convivial atmosphere.These potential impacts underscore the necessity for future interaction designers to meticulously address how IVAs can proficiently discern users' real-time emotional shifts within the vehicle.Extensive research exploration is warranted to elucidate the link between users' emotional variations across diverse functional scenarios and the resultant interaction outcomes of each IVA.

5.2.2
Long-term Traits .The influence of long-term characteristics on personalization design is depicted in Fig. 6, wherein we conducted a preliminary compilation of users' long-term characteristics that may wield influence over the future IVAs' interactive design.Our findings reveal that the long-term characteristics significantly impact the design of IVAs' appearance and social attributes.
Users' openness to new technologies will affect the perception of IVAs' Predictability and Trust.For instance, some users exercise a cautious attitude toward novel features, placing higher demands on IVAs' predictability.D1 abstains from utilizing automatic pilot and automatic car backing and maintains a lack of trust in newer functionalities, such as reversing cameras.Conversely, D8 eagerly embraces new technology advancements in vehicles, displaying a pronounced interest in the visual appearance of the car.Also, users' willingness to cooperate with new technologies also affects users' trust and growth expectations towards IVAs.People who have high willingness are usually friendly and compassionate and will have higher cooperation tendencies and growth expectations of IVAs.P12 expressed the hope that IVAs can continuously learn from driving behaviors and provide corrective feedback to cultivate proper driving habits.D3 envisions IVAs' mobility, moving to different positions as needed by passengers, serving both front and rear passengers instead of just the driver.P11 hopes that IVAs can change their tone according to different scenes and types of people in the car.For example, when there are children or elderly people in the car, IVAs' voice and tone will be more gentle for such groups.On the contrary, individuals with low willingness tend to exhibit skepticism and distrust towards IVAs, prioritizing selfdetermination and control.D3 and D6 express reservations about IVAs autonomously performing functions while driving, valuing the driver's control over the vehicle.
The intentions of the user for the vehicle will affect the IVAs' Capabilities and Security design.For instance, P9 prefers technology in the car that enhances driving comfort and communication, favoring meaningful functionality over superficial technological effects.D4 and D8 express reservations about IVAs interfering with the driving process, especially among the elderly who may find IVAs' movements cumbersome and unfamiliar.D8 suggests customizing the number of IVAs for the front and rear rows, each with distinct driving modes, catering to passenger needs without disturbing the front row.
The extraversion of the user will affect the social characteristics of IVAs' growth and warmth displayed during the interaction process.Extroverted individuals may lean towards IVAs for communication and entertainment.P11 envisions IVAs connecting with nearby cars during a traffic jam, initiating discussions to foster brief social interactions.P2 hopes that IVAs can move and activate the atmosphere in the car.For example, IVAs can appear as multiple cats on the seats and interact with passengers, so as to enrich in-car engagement.Conversely, people with low extroversion prefer to drive alone and do not want to be interrupted by IVAs while driving.D3, an introvert, prefers to drive alone, quietly listening to music.The introverted P12 wishes to have independent, highly customized IVAs so that when she gets in the car, she can naturally form a tacit understanding of interaction, reducing the need for extensive conversation.

DISCUSSION
In this section, by reflecting on the qualitative and quantitative findings, we discuss how we provide personalized and intelligent interactions for users when designing future IVAs.

Strategies for In-car Personalization and Implicit Interaction with IVAs
The major challenge in the design of future IVAs is what technical means should IVAs use to accurately grasp the different interaction rules brought about by the long-term and short-term characteristics of users, and customize the interactive experience according to the specific needs of different users.With recent advances in artificial general intelligence (AGI), the combination of large language models and materialized agent technology seems to give IVAs the ability to recognize and respond to users' vague interaction intentions.Moreover, these advancements empower IVAs to facilitate personalized interactions by harnessing memory mechanisms.Then, we will discuss the future of in-car interactions in terms of personalized and implicit interaction design, in conjunction with today's AGI technology.
6.1.1AGI driven Personalized Implicit Interaction.The development of AGI has enabled IVAs to recognize and respond to the user's ambiguous interaction intentions [2], which means that in the future, combined with the deployment of multi-modal sensors in the car, IVAs can provide more adaptive interaction experiences.As shown in Fig. 7, if the user indicates that she is preparing to attend a friend's wedding, IVAs can first identify the real-time emotion of the user according to the sensor, determine the user's interaction scenarios and functional requirements, and make corresponding personalized feedback based on the long-term characteristics of the user.For example, it can create a customized atmosphere of preferred color and music for the user to relax.Adjusting the multi-modal interaction device on the seat provides the user with a warm hug and so on.In addition, the user can also express some fine-tuning instructions during the interaction, and IVAs can immediately make adjustments and fine-adjust them into new rules.
6.1.2AGI driven Personalized IVAs Form.In addition, through our research, we found that users' needs for the form of IVAs and the corresponding services vary with their long-term and shortterm characteristics.This implies that future designs of IVAs need to focus more on the variability of social entities.Advances in generative AI provide an interesting opportunity for IVAs to automatically adjust their appearance based on different text inputs [14].This provides a valuable opportunity for future IVAs to be designed by recognizing the characteristics of each user interaction and adjusting the appropriate actions, appearance, and even morphology.Therefore, we believe that this change will challenge how to quantitatively assess user characteristics during in-vehicle interactions and finely correlate these characteristics with various design parameters of IVAs.We expect that future research will further advance this quantitative design and bring more colorful interactive experiences to in-vehicle users through AI technologies.
6.1.3AGI-based Interaction Collaboration between User and Designer.In addition, we believe that the future interaction design should be a collaboration of designers and users.Although a lot of interaction design in the past relied on user participation [15, 21, 29], in the future environment of complex and changing in-car interaction, it is difficult for designers to quickly track and iterate on the precise interaction for each user.However, the large language model and the concept of embodied agents [7] give IVAs the ability to produce their interactive solutions according to the vague needs of the user [2].It means that in the future, in-car interaction designers can integrate the brand tonality and style into the interaction rules of IVAs in advance during the development process of IVAs.In this way, IVAs can design interactive solutions for users while keeping consistent with the brand image of the car company.At the same time, users can fine-tune the base model, which is already aligned with the brand, to achieve a refined personalized interactive experience.

Reflections and Visions of In-car Multi-modal Interaction
Through the think-aloud study, we are also seeing more and more expectations from users for multi-modal interactions in the car.While traditional in-car interactions relied more on light and sound interaction, in the era of electric vehicles, the display increasingly appeared in the car for user interaction.While some car manufacturers are trying to use concepts such as robots and agents as a follow-up vision for displays, it still does not get rid of the sound and visual interaction with the central control screen as the core.Our research reveals that users' expectations of multimodal interactions are not limited to these modalities and that they are increasingly beginning to expect IVAs to initiate interactions from non-central control screens, such as automotive interior trim.It coincides with the increasing focus on smart materials in the research field of HCI.If interaction technology continues to enhance a wide array of automotive components, it is conceivable that future IVAs will possess the capacity to deliver tailored multi-modal interactions to individual users, spanning various sensory modalities.These multimodal interactions will not be limited to sound or vision, but will also include tactile and olfactory modalities.Then, we will review the users' suggestions for multimodal interactions in this study and discuss the future development and vision of in-car interactions in combination with current smart materials.

6.2.1
Human-centered Material Interaction.The existing IVAs interaction mainly relies on the input and output of traditional electromechanical control, for example, the common input devices in cars are buttons, knobs, microphones, and output devices are lights and sound.However, as the in-car IVAs evolve from multimodal interaction to human-centered interaction design, more input and output variables can be taken into account using the technology of smart materials.For example, with the fluid material [18], we can use the changes of temperature [25], light [25], and resonance [22] caused by human interaction to give feedback to color [19], temperature [31], and touch [20].These ever-expanding interactive driving mechanisms will enable more possibilities for IVAs' interactions.In the future, in-car interactions may become closer to ubiquitous computing, from the integrated display as the center of interaction with IVAs to automotive interior trim.

Fusion of Traditional
Craft and Multi-modal Interaction.In addition to using smart material to expand the interactive method of the future IVAs, its combination with traditional craft also opens up possibilities.It means that IVAs are increasingly free to provide interactive scenarios that match the local aesthetic and style according to the cultural characteristics of local consumers.The combination of traditional crafts is a prospective research direction of HCI, a growing number of scholars are exploring the technological appeal that cultural crafts can offer [30].For example, with thermotropic pigment, designers can combine traditional Chinese painting techniques with dynamic interactive painting to create a new experience for users [17]; with dynamic information expression of fluid, designers can also combine the fiber weaving process with interactive art, bringing dynamic interaction to users with vitality [16].In the future, the field of in-car interaction design is likely to evolve and incorporate various aspects, including traditional craft, multi-modal automotive interior trim, material interaction, and more.This evolution aims to make in-car interaction increasingly universal and localized, offering poetic solutions integrating technology and cultural aroma.

LIMITATIONS AND FUTURE WORKS
While the present study provides insights into users' perceptions and future imaginings of Intelligent Virtual Assistants (IVAs), it is essential to acknowledge certain limitations.All participants in this study were from the same country and city, and their numbers were limited.This localized sample introduces a potential bias, as the research outcomes may be influenced by the regional culture and individual backgrounds of the participants.Future studies will expand upon the current results, involving participants from diverse backgrounds.
In addition, the impact of users' long-term characteristics seems to be related to their personality traits.Our future research might be able to use the Big Five personality theory to match users' personalities quantitatively (e.g., openness, conscientiousness, extraversion, agreeableness, and neuroticism), thus further researching how to design IVAs via users with different personality traits [32].

CONCLUSION
This work surveys users' perceptions and imaginations about future IVAs.Specifically, we attempt to explore the relationship between user characteristics during in-car interactions and the design of their interaction scenarios, functions, forms, and social attributes for IVAs.To explore the issue, 161 participants were surveyed about their basic preferences for IVAs' interactions through a pilot questionnaire.Subsequently, we conducted a think-aloud study based on an experimental car.Using recorded and semi-structured interviews, we assessed the user characteristics of 12 participants and their expectations for future IVAs.Through comprehensive qualitative and quantitative analysis, we summarize users' expectations for future IVAs into four categories, scenarios, functions, appearance, and social attributes, and discuss the potential design relationship between them and users' characteristics.In addition, by deeply discussing the impact of these relationships on future IVAs, we propose several design strategies for future IVAs' interaction from two aspects: smart materials-based multimodal interaction and AGI-based personalized and implicit interaction.

Figure 1 :Figure 2 :
Figure 1: The Flow of the Pilot Study

Figure 3 :
Figure 3: The Flow of our Think-aloud Study

4. 3 . 2
Interview.After the car returned to the designated parking spot, a three-part interview was conducted.The interview sections were: experience interview, views on methods of IVAs and multimodal interaction, and personal attitudes.Semi-structured interviews allow experimenters to explore specific topics in depth based on participants' responses, maintaining some flexibility.The

Figure 4 :
Figure 4: Four design categories of user expectations for future IVAs

Figure 5 :
Figure 5: Future trends of personalization and implicitness in IVA design

Figure 6 :
Figure 6: Design suggestions based on the relationship between user characteristics and user's expectations of interactive scenarios, functionality, appearance, and social attributes of IVAs.

Figure 7 :
Figure 7: Design suggestions based on the relationship between user characteristics and their design of interaction scenarios, functionality, form, and social attributes of IVAs.

•
Affective Processes: Please imagine that there is an IVA in your car that can recognize your emotions, communicate with you according to your emotions, and provide different modes of service in response to your emotional changes.•Cognitive Processes: Please imagine that there is an IVA in your car that can provide real-time feedback based on your driving status or that of your driver, and provide driving status warning services such as distraction warning.• Overall Experience: Please imagine that there is an IVA in your car, which is responsible for helping you check the weather, navigation routes, and other basic functions.

Table 1 :
Participants' Preference for Graphical Style

Table 2 :
The Overview of Participant Demographics