Current Status and Challenges in Usability Evaluation Research on Human–Machine Interaction in Intelligent Vehicles: A Systematic Review

Featured Application

This study systematically analyzes the automation levels, task scenarios, and interaction modalities that are currently the focus of intelligent vehicle human–machine interaction usability evaluation research. Based on this analysis, it reveals the current application status and challenges in three key aspects of this field. These findings hold significant value for promoting research and technological advancement in this field.

Abstract

Usability evaluation is a crucial stage in optimizing human–machine interaction design for intelligent vehicles. The academic community has accumulated substantial research findings on this topic. However, the global landscape, core focus areas, and potential challenges of these findings remain unclear and require systematic review and analysis. This is essential for understanding progress in the field and clarifying future research and technological development directions. To address this research need, we followed the PRISMA guidelines and conducted a systematic review of 158 applied research cases in the field of usability evaluation for intelligent vehicle human–machine interaction. We focused on analyzing the automation levels, task scenarios, and interaction modalities examined in these cases. Based on this analysis, we examined the current development status and potential challenges and identified future development recommendations for this field. Our research reveals that current usability evaluation studies of intelligent vehicle human–machine interaction exhibit a contradictory pattern of “coexisting development and limitations.” On the one hand, the research scope and depth continue to expand with technological advances. On the other hand, related research topics show obvious lag, with particularly insufficient research on non-driving tasks and multimodal interaction topics. The potential challenges arising from this research status seriously constrain industry development and require practitioners to actively address them. In this regard, our research findings provide a valuable reference for practitioners in understanding the current research progress and planning future development directions.

1. Introduction

Intelligent vehicles, through the integration of advanced sensors and artificial intelligence technologies, are gradually evolving into intelligent next-generation mobile spaces and application terminals [1,2,3,4]. These transportation vehicles, commonly referred to as intelligent and connected vehicles (ICVs) or autonomous vehicles, have the core goal of enhancing driving safety, comfort, and traffic efficiency [5,6,7]. However, while this technological revolution is reshaping the driving experience [8,9,10,11], it also brings unprecedented challenges to human–machine interaction design [4,12,13,14,15,16,17]. To address these challenges, conducting effective usability evaluations in the development process of intelligent vehicles has become a crucial stage for ensuring successful human–machine interaction design [18,19,20,21,22,23]. For this reason, the field of intelligent vehicle human–machine interaction usability evaluation has attracted attention from the academic community and generated diverse and rich research cases [24]. However, each research case often focuses on specific technologies or application scenarios, which has resulted in a “fragmented” state.

In this context, to provide clearer guidance for research development and technological practice in this field, it is particularly important to systematically review and analyze the current research progress in intelligent vehicle human–machine interaction usability evaluation. Specifically, this research need stems from the following motivations: (1) Clarifying current research progress can reveal research hotspots, trends, and neglected research topics within the field, thereby providing a reference for future research and technological development. (2) Integrating and consolidating scattered research cases can both extract valuable practical experience and expose common problems facing the entire field, thereby improving future research.

In recent years, scholars have attempted to review this field using literature surveys. For example, Su et al. summarized changes in objects and tasks for autonomous vehicle human–machine interaction usability evaluation, providing guidance for selecting evaluation methods in this field [25]. However, their analysis dimensions were relatively macro-level and did not deeply explore the application status of different automation levels, task scenarios, or interaction technologies (i.e., interaction modalities). Forster et al. analyzed 42 papers published at the AutoUI conference before 2018 and reported the field’s attention to different levels of driving automation accordingly [24]. However, the limitation of their work lies in its single data source (limited to the AutoUI conference) and lack of exploration of the application status of task scenarios and interaction modalities. Albers et al. reviewed 16 papers on the usability evaluation of human–machine interfaces for conditional automated driving (SAE L3 level) and summarized the practices and best-practice recommendations of these papers in test cases [26]. However, their study only focused on L3-level driving automation and did not systematically examine diverse task scenarios and interaction modalities.

Overall, although previous research findings provide a valuable reference for practitioners to understand the research status in this field, their studies often have limitations in terms of the time span, coverage scope, and discussion topics. In this regard, previous research collectively points to an urgent research gap that needs to be addressed: the academic community currently lacks a comprehensive, multi-dimensional systematic review that depicts a panoramic view of the current research on intelligent vehicle human–machine interaction usability evaluation research. This research gap makes it difficult for practitioners, especially those new to the field, to accurately judge the development trajectory of and potential innovation points in the field, thereby affecting the direction and value of research work.

To resolve the contradiction between the above research needs and research gaps, and thereby help practitioners in this field better conduct usability evaluation research, this study used systematic review methods to conduct an in-depth investigation and analysis of the current research status of intelligent vehicle human–machine interaction usability evaluation. Specifically, we first extensively collected relevant applied research literature in this field. Subsequently, through rigorous screening and data extraction processes, we conducted a detailed analysis of 158 research cases that met the inclusion criteria, focusing on the research topics (including automation levels and task scenarios) and interaction modalities (i.e., interaction technologies) of these cases. Based on the analysis results, we answered three research questions (RQs):

RQ1: What is the current development status of usability evaluation research on intelligent vehicle human–machine interaction? (i.e., What types of driving automation levels and task scenarios does this field focus on? Which interaction modalities are emphasized?)

RQ2: What are the potential challenges in current usability evaluation research on intelligent vehicle human–machine interaction? (i.e., What research limitations and deficiencies exist in this field regarding driving automation levels, task scenarios, and interaction modalities?)

RQ3: What are the development recommendations for future usability evaluation research on intelligent vehicle human–machine interaction? (i.e., To address potential challenges, what research directions and optimization recommendations urgently need exploration in the future?)

By answering the above research questions, we bring the following contributions to the field of intelligent vehicle human–machine interaction usability evaluation: (1) This study is the first to systematically reveal the knowledge structure, research hotspots, and potential gaps in this field from a comprehensive perspective of automation levels, task scenarios, and interaction modalities, contributing new knowledge to the field. (2) The research results help practitioners (especially novices) quickly understand the overall landscape of the field and more precisely position their work, thereby improving the effectiveness of research design. (3) By analyzing the challenges and shortcomings of existing research, this study lays the foundation for establishing more comprehensive and forward-looking usability evaluation frameworks and research agendas in the future.

2. Materials and Methods

In this section, we briefly introduce the systematic review method used in this study. The detailed implementation procedures of this method have been organized in Supplementary File S1 of the Supplementary Materials. This arrangement not only addresses the length constraints of the main text but also, more importantly, enhances the transparency of the entire research process by providing an independent and comprehensive methodological record.

2.1. Research Framework

We adopted a systematic review methodology to conduct this study. Systematic review, as a structured investigation method, can effectively integrate and analyze dispersed research materials, thereby exploring specific research questions beyond individual studies [27]. Additionally, through its rigorous and systematic research process, systematic review can minimize subjective bias to the greatest extent, thus improving the reliability of analytical results.

To ensure the standardization and rigor of our research, we followed the PRISMA guidelines and examined research cases from the same field to establish the methodological framework for this study [28,29,30,31]. As shown in Figure 1, our research process is divided into three progressive stages: acquiring literature, arranging literature, and analyzing literature.

2.2. Phase 1: Acquiring Literature

In the literature acquisition phase, we aimed to comprehensively obtain literature materials relevant to the research questions. The following paragraphs present the specific operational procedures, important notes, and execution results.

• Operational Procedures

We collected literature through two pathways: online databases and gray literature. First, we conducted systematic searches across ten major international academic databases, including Web of Science, Taylor & Francis Online, and Elsevier ScienceDirect. These databases cover primary global scientific research resources (see Supplementary Materials—Supplementary File S1—Table S2), ensuring the authority and reliability of the literature. Additionally, to ensure the breadth of literature collection and address potential omissions in database searches, we also collected gray literature. Gray literature sources included literature discovered through citation networks, literature identified during full-text reviews, and literature from our daily accumulation.

2.

Important Notes

Setting appropriate search terms and search expressions was crucial during the literature collection process. Therefore, to ensure the quality of literature collection, we implemented the following measures: (1) Based on research questions, themes, and pilot searches, and considering terminological diversity (such as synonyms and singular/plural forms), we identified search terms/phrases (see

Table 1. Keywords used in literature search.

Group A	Group B	Group C 1
Usability	Evaluation Assessment	Intelligent Vehicle, Intelligent Connected Vehicle, Smart Vehicle; Intelligent Driving, Smart Driving; Intelligent Cockpit, Smart Cockpit; Intelligent Car, Smart Car, Intelligent Connected Car; Autonomous Vehicle, Automated Vehicle, Automatic Vehicle, Self-Driving Vehicle; Autonomous Driving, Automated Driving, Automatic Driving; Autonomous Car, Automated Car, Automatic Car, Self-Driving Car

¹ Note 1: The search formulas for these keyword combinations and their search results are provided in Supplementary Materials—Supplementary File S1—Table S2.

). (2) Considering that some literature might not directly use common terms like “human-machine interaction” but instead describe specific tasks (such as “driving takeover” or “voice interaction”), we adopted a broad keyword combination strategy that was not limited to specific terminology. Although this increased the screening workload, it maximized the completeness of literature collection. (3) To ensure comprehensive literature retrieval, we did not impose publication year restrictions. This was because the emergence and development timeline of related research topics is difficult to define precisely. (4) We constructed search expressions using Boolean operators and wildcards and applied limiting conditions such as subject areas and article types to balance precision and recall rates. (5) We utilized academic paper network platforms such as Inciteful and Connected Papers to rapidly discover relevant literature.

3.

Execution Results

Through the above literature collection strategy, we initially acquired 1938 articles. Among these, 1795 articles were sourced from database searches and 143 articles from gray literature. These articles required more detailed examination before determining their inclusion in this study.

2.3. Phase 2: Arranging Literature

After completing the literature collection, we entered the literature arrangement phase. During this phase, we systematically processed the literature materials in three core stages. First, we conducted literature screening based on predetermined criteria to exclude irrelevant studies. Second, we performed thematic coding on the qualifying literature to extract key information. Finally, we conducted quantitative analysis on the coded data. These procedures were designed to ensure the reliability and accuracy of this research.

2.3.1. Literature Screening

The purpose of literature screening was to select core literature directly relevant to our research questions from the preliminary literature obtained during the literature acquisition phase. The following presents the specific operational procedures, important notes, and execution results.

• Operational Procedures

We implemented literature screening in two steps. First, we reviewed the titles and abstracts of all literature to preliminarily exclude literature that did not meet the inclusion criteria. Second, we conducted full-text readings of literature that passed the title and abstract screening to thoroughly assess the alignment between the research content of these articles and our research questions.

2.

Important Notes

Setting screening criteria and reaching consensus opinions were crucial during the literature screening process. Therefore, to ensure the quality of literature screening, we implemented the following measures: (1) Two researchers (i.e., the second and third authors) strictly followed pre-established inclusion and exclusion criteria for literature screening (see

Table 2. Inclusion and exclusion criteria followed during literature screening.

No.	Item 1	Inclusion and Exclusion Criteria 2
1	Language	Inclusion: Only English-language literature was included. Exclusion: Non-English literature was excluded.
2	Type	Inclusion: Only journal articles and conference papers were included. Exclusion: Literature from non-academic sources (not peer-reviewed) such as market reports, news articles, white papers, and working papers was excluded. Additionally, books and dissertations were also excluded; although they may contain relevant research, their comprehensive content makes them inconvenient for rapid reading and analysis.
3	Permission	Inclusion: Only literature with accessible full text was included. Exclusion: Literature with copyright restrictions was excluded.
4	Domain	Inclusion: Literature focusing on “HMI (human-machine interaction)” as the research subject was included. Exclusion: Literature outside the “HMI (human-machine interaction)” research field was excluded.
5	Subject	Inclusion: Literature focusing on in-vehicle interaction in passenger vehicles as the research subject was included. Exclusion: Literature discussing topics such as public transportation vehicles, traffic accidents, traffic network systems, automated driving roads, and intelligent transportation was excluded.
6	Divergence	Temporarily include disputed literature: For research literature for which relevance cannot be clearly determined, it is recommended to retain it initially and make decisions after subsequent discussion, thereby avoiding potential selection bias.
7	Redundancy	Exclusion of duplicate literature: Before conducting full-text review, duplicate literature should be excluded. Duplication here not only refers to general duplication but also includes literature with overlapping content published by the same author, even when these publications use different titles.
8	Content	Inclusion: This study primarily included applied research literature focusing on HMI design practice evaluation, for example, feasibility validation of novel interaction design solutions, comparative evaluation of multiple interaction design solutions, and user interaction performance and experience studies under specific driving scenarios. In other words, this study prioritized research cases related to actual project development. Exclusion: This study excluded fundamental research literature primarily focused on theoretical research, such as studies aimed at constructing theoretical frameworks, exploring basic principles and patterns of human–machine interaction, validating psychological models, constructing behavioral models, and investigating related influencing factors.
9	Method	Inclusion: Literature that conducted empirical analysis was included. Exclusion: Articles that were narrative reviews, comparative studies, survey research, or other types of reviews were excluded. Additionally, studies that only discussed interaction design concepts or user interaction experiences were also excluded.
10	Quality	Inclusion: Literature with relatively well-designed research methodology was included. Exclusion: Research literature that did not provide clear and detailed usability evaluation methods was excluded.

¹ Note 1: We recommend using criteria 1–6 during the title and abstract screening stage of literature, which helps rapidly identify literature relevant to the research question. For the full-text review stage, criteria 6–10 should be applied. ² Note 2: Detailed explanations of these inclusion and exclusion criteria can be found in Supplementary Materials—Supplementary File S1—Table S3.

) to ensure decision consistency. (2) After researchers independently performed literature screening, they engaged in face-to-face discussions to reduce personal bias. (3) Disagreements were resolved through “team consensus,” with industry expert arbitration sought when necessary to ensure decision accuracy. In other words, we required researchers to reach absolute consensus through in-depth discussions and excluded literature for which consensus could not be achieved. (4) For literature that could not be included, we documented detailed exclusion reasons in a shared Microsoft Excel spreadsheet to ensure transparency of the screening process. (5) During the full-text review process, we also examined cited literature to discover gray literature that might have been missed by database searches and included it in the full-text review.

It is worth emphasizing that this study only included applied research literature focusing on the evaluation of human–machine interaction design practices in intelligent vehicles. This is because, compared to fundamental theoretical research, applied research literature is more aligned with industry practices in terms of research objectives, content, and methods (see Table 2—Row 8 explanation), which makes it more helpful in directly providing actionable guidance for practitioners.

3.

Execution Results

After literature screening, 158 core articles were ultimately included in the analysis (related materials can be obtained by contacting the corresponding author). Subsequently, we conducted structured coding of the research cases reported in these articles to provide data support for answering the core questions of this study.

2.3.2. Literature Coding

After identifying the 158 core literature pieces, we proceeded to the literature coding phase. In this phase, we aimed to systematically extract key information relevant to the research questions from the core literature. The following paragraphs present the specific operational procedures, important notes, and execution results.

• Operational Procedures

Following the predetermined data extraction and classification protocol (see

Table 3. Data extraction and classification protocol used in literature coding (see Supplementary Materials—Supplementary File S1—Table S5 for related terminology definitions).

Data Entry	Extraction Items	Terminology Classification of Related Extraction Items
		Primary Classification	Secondary Classification
Data 1 (for addressing Questions 1–3)	Automation Level	Driving Assistance (L0–L2)	Level 0: no driving automation (emergency assistance); level 1: driving assistance (partial driving assistance); level 2: partial driving automation (combined driving assistance)
		Conditional Driving Automation (L3)	Level 3: conditional driving automation
		Advanced Driving Automation (L4–L5)	Level 4: high driving automation; level 5: full driving automation
Data 2 (for addressing Questions 1–3)	Task Scenario	Primary Driving Task	Autonomous driving task; driving takeover task; advanced driver assistance task
		Secondary Driving Task	Vehicle navigation task; instrument cluster system; driving condition prompt system; warning system; head-up display system; vehicle imaging system
		Non-driving Task	Virtual intelligent voice assistant; function-setting task; vehicle entertainment task; vehicle intelligent robot; communication and social task; other tasks (some task scenarios that are difficult to categorize, such as interface layout, font design, screen size, personalized interface design, etc., also called edge tasks)
Data 3 (for addressing Questions 1–3)	Interaction Modality 1	Input Modality	In-touch screen; in-steering wheel key; in-central control key; in-gestures; in-speech; in-multimodal
		Output Modality	Out-indicating light; out-acoustic; out-speech; out-central control interface; out-dashboard interface; out-HUD interface; out-haptic vibration; out-haptic temperature feedback; out-olfactory; out-electronic rearview mirror monitoring; out-robot facial expression; out-other interface (some interaction modalities that are difficult to categorize or overly complex, such as head-mounted displays); out-multimodal

¹ Note 1: It should be emphasized that interaction modalities need to be classified in a balanced manner. The traditional approach of categorizing modalities by sensory categories is not conducive to analysis. Specifically, first, if the classification is too coarse (e.g., classifying central control keys and steering wheel keys uniformly as “keys,” or classifying central control interfaces and HUD interfaces uniformly as “visual”), it will result in different operation methods even when using the same interaction modality, which is not conducive to in-depth analysis. Second, overly fine classification (such as subdividing central control keys into mechanical keys, touch keys, knobs, etc.) makes it difficult to compare interaction modalities across different vehicle models. Therefore, we have chosen a classification method that can both distinguish different operation methods and facilitate comparison across different vehicle models to better support HMI usability evaluation.

), two researchers (the second and third authors) independently reviewed and coded the core literature, recording the required data in detail in a shared Microsoft Excel electronic spreadsheet.

2.

Important Notes

To ensure the objectivity, accuracy, and consistency of the coding results, we implemented the following measures: (1) We developed a structured data extraction and classification protocol and provided training to all researchers. (2) During data recording, we preserved the original expressions from the literature to avoid paraphrasing biases, thereby ensuring coding precision and traceability. (3) The two researchers conducted multiple rounds of cross-checking and focused discussions, carefully comparing their respective coding results and effectively resolving questions and disagreements. (4) For cases where literature contained unclear descriptions, the two researchers engaged in careful discussion based on contextual information before jointly determining the final extracted content. (5) Additionally, we invited industry experts to arbitrate and review the coding results. Most importantly, through expert consultation, we not only validated the effectiveness of the coded data but also supplemented several practical experiences that had been overlooked in existing literature but possess significant potential value. (6) We employed multiple rounds of “team discussions” to ensure the reliability and validity of the coding, actively promoting and ensuring consistency among researchers.

3.

Execution Results

Through the above literature coding strategy, we structured and recorded the consensus coding data in a shared electronic spreadsheet. These data will provide the foundation for subsequent quantitative and analytical work. It should be noted that not every research case could present complete target data, as some research cases had missing data. Therefore, in subsequent statistical analysis, we only processed the valid data that were actually obtained.

2.3.3. Literature Counting

After obtaining the structured coding data, we proceeded to the literature-counting phase. In this phase, we aimed to conduct statistical analysis and visualization processing of the relevant data. Specifically, we used Origin 2025 software to create relevant charts (see below). These visualization charts intuitively reveal the current status of intelligent vehicle human–machine interaction usability evaluation research across related research topics, providing clear evidence for the comparative analysis, phenomenon identification, and conclusion extraction in this study.

2.4. Phase 3: Analyzing Literature

After completing the structured organization and data analysis of the literature information, we proceeded to the literature analysis phase. In this phase, we conducted in-depth analysis of and discussions on the relevant chart information, thereby grasping the progress and limitations of this field. The following paragraphs present the specific operational procedures, important notes, and execution results.

• Operational Procedures

Literature analysis was primarily conducted through three steps: internal discussions, expert reviews, and comprehensive synthesis. First, we compared and discussed different practical cases (i.e., internal discussions), identified common characteristics in existing research, and pointed out existing challenges and deficiencies. Second, to ensure the reliability of the research, we invited industry experts to review and discuss the relevant texts, charts, and preliminary research findings (i.e., external discussions). Finally, we integrated the results from internal discussions and expert reviews to form an in-depth understanding of the research questions.

2.

Important Notes

Literature analysis should adhere to the principles of objectivity, comprehensiveness, critical thinking, and comparative analysis. Specifically, it is important to maintain an objective stance, avoid subjective speculation and personal bias, and give equal attention to and fairly evaluate all practical methods. We applied systematic thinking to comprehensively analyze the connections and differences among different practical cases, thereby gaining a deep understanding of the advantages and disadvantages of various evaluation tools.

3.

Execution Results

Through the above literature analysis measures, we ultimately formed a comprehensive understanding of the current development status, potential challenges, and development recommendations for intelligent vehicle human–machine interaction usability evaluation research.

3. Results and Discussion

Through a systematic review of 158 studies on the usability evaluation of human–machine interaction in intelligent vehicles, we mapped the research landscape of this field (see below). Specifically, Figure 2 presents the distribution of current research across two core dimensions: automation level and task scenarios. Building on this foundation, Figure 3 further reveals the temporal evolution trends of research intensity across different task scenarios. Additionally, Figure 4 reports the types, frequencies, and proportions of interaction modalities examined in different task scenarios.

Based on these comprehensive research maps, we report and discuss the relevant research findings in this section according to the sequence of research questions. These research findings can help practitioners comprehensively understand the research progress in this field and provide a reference for their research work.

3.1. Current Development Status of Usability Evaluation Research on Human–Machine Interaction in Intelligent Vehicles (Addressing Research Question 1)

In this section, we address Research Question 1: What is the current development status of usability evaluation research on human–machine interaction in intelligent vehicles? Overall, with the continuous evolution of automated driving technology (i.e., changes in human–vehicle driving roles), the task scenarios and interaction modalities of concern in this field continue to expand, triggering many new and specific human factors issues. The following sections will elaborate on the current status of usability evaluation research on human–machine interaction in intelligent vehicles, focusing on the main research directions, research priorities, and existing stage-specific characteristics.

3.1.1. Research Status 1: Existing Human–Machine Interaction Usability Evaluation Research Primarily Focuses on L0–L2 Driving-Assistance Scenarios

As shown in Figure 2, to facilitate the analysis of existing research focus areas, we constructed a three-level driving-automation classification framework based on international standards (ISO/SAE PAS 22736:2021) [32], industry standards (such as SAE J3016-2021, GB/T 40429-2021) [33,34], and critical transitions in driver roles: L0–L2 (driving assistance), L3 (conditional driving automation), and L4–L5 (advanced driving automation). This classification framework not only reflects technological evolution but also corresponds to the core contexts for human–machine interaction usability evaluation, i.e., the transition from “continuous supervision” to “readiness to take over” and then to “complete handover” within human–machine systems.

Based on this classification framework, we found that current human–machine interaction usability research primarily focuses on L0–L2 driving-assistance scenarios. As illustrated in Figure 2, half (50.57%, 88/174) of the research cases are dedicated to evaluating human–machine interaction usability at the L0–L2 level. In comparison, research targeting L3 conditional driving automation and L4–L5 advanced driving automation accounts for 25.86% (45/174) and 22.41% (39/174), respectively. This research finding corrects the previous research viewpoint of Forster et al., who argued that L3 was the most commonly studied automation level [24]. The reason for this discrepancy may lie in the fact that Forster’s data sources were mainly limited to specific academic conferences (such as AutoUI).

Overall, the concentration of current research on L0–L2 levels is driven by two major factors. First, from a market perspective, L2-level driving-assistance systems (such as adaptive cruise control and lane-keeping assist) represent the most widely deployed technology in production vehicles [35]. Their large user base has exposed issues such as mode confusion, over-reliance, and misuse in daily operation [35,36,37,38,39,40,41], which create the most urgent and prevalent safety challenges. Second, from an interaction perspective, drivers and vehicles at the L0–L2 stage operate in a state of “human-machine co-driving” collaboration [42,43,44]. This high-frequency, dynamic interaction process provides rich and critical research scenarios for human–machine interaction usability studies.

3.1.2. Research Status 2: Current Research Focuses on Primary Driving Tasks, While Non-Driving Tasks Receive Limited Attention

As shown in Figure 2, to facilitate the analysis of the focus areas in existing research, we categorized current task scenarios in intelligent vehicles into three levels based on industry experience and the core contexts of current human–machine interaction design and evaluation (detailed explanations and subdivided task scenarios are provided in Supplementary Materials—Supplementary File S1—Table S5): primary driving tasks, secondary driving tasks, and non-driving tasks. Specifically, primary driving tasks directly relate to driving safety and core vehicle control (such as autonomous driving and driving takeover) [25,45,46,47,48]. Secondary driving tasks provide support for primary driving tasks by offering decision-making information (such as navigation, traffic information, and vehicle status data) [47,48,49]. Non-driving tasks are detached from driving operations themselves [25,45,47,48,50], focusing on the comfort, entertainment, and office needs of drivers and passengers (such as media control and environmental adjustment).

Based on this classification framework, we identified a significant distribution characteristic in the current research field: research resources and attention are highly concentrated on primary driving tasks, while the exploration of non-driving tasks remains relatively limited. As shown in Figure 2, among the 174 research cases analyzed, studies focusing on primary driving tasks accounted for 43.68% (76/174), secondary driving tasks followed closely with 31.03% (54/174), and non-driving task research had the lowest proportion at only 25.29% (44/174). Additionally, this imbalance becomes more pronounced when examining subdivided task scenarios. Specifically, the four subdivided task scenarios with the highest current research intensity are almost all closely related to driving safety and core control authority (cumulative proportion of 53.44%): driving takeover tasks (22.41%, 39/174), autonomous driving tasks (10.92%, 19/174), advanced driver assistance tasks (10.34%, 18/174), and vehicle navigation tasks (9.77%, 17/174). By contrast, non-primary driving tasks such as vehicle imaging systems, communication and social tasks, vehicle intelligent robots, instrument cluster systems, and function-setting tasks have a cumulative proportion of less than 9% (8.62%, 15/174).

In summary, the academic community’s core agenda primarily revolves around the safety and reliability of “human-machine co-driving.” This conclusion supports the previous survey findings of Albers et al. [26]. Driving takeover and advanced driver assistance are key components for handling potential hazards and ensuring driving safety [51,52,53]. The greater attention invested in these areas reflects the industry’s cautious attitude toward fundamental safety assurance in the progression toward higher levels of automated driving. However, the insufficient attention to non-driving tasks in existing research is also noteworthy (see discussion of “Potential Challenge 1” below).

3.1.3. Research Status 3: Each Automation Level Triggers Specific Human Factors Issues

As shown in Figure 2, human–machine interaction usability evaluation research presents different focuses and topics across different automation levels (the width of connecting lines in the figure represents the association strength and research intensity of research topics):

• At L0–L2 levels (driving assistance), drivers remain the absolute primary agents for vehicle control and environmental monitoring. Accordingly, current human–machine interaction usability evaluation research focuses on optimizing the effectiveness of task scenarios such as advanced driver assistance systems, warning systems, vehicle navigation tasks, and head-up display systems. In other words, the usability goal at the L0–L2 levels is to provide effective assistance to drivers focused on driving [54] to ensure that information delivery is both efficient and safe.
• When automation levels reach L3 (conditional driving automation), human–machine interaction conflicts concentrate on the critical point of “human-machine co-driving.” [55] Accordingly, the vast majority of current human–machine interaction usability evaluation research focuses converge on driving takeover as the core task (82.22%, 37/45). How to design efficient and reliable takeover requests to ensure drivers can safely and promptly regain control from the system has become the most critical and challenging human factors problem at this stage [56,57].
• When automation levels enter L4–L5 (advanced driving automation), the driver’s role completely transforms to that of a passenger. Accordingly, current human–machine interaction usability evaluation research priorities also shift to two aspects: first, human–machine interaction surrounding autonomous driving tasks themselves, e.g., how to effectively communicate the vehicle’s driving intentions and decision-making rationale to passengers [58]; second, maintaining passengers’ situational awareness of the vehicle and surrounding environment through driving condition prompt systems [59]. However, it is noteworthy that although passengers are liberated from driving tasks, current research on non-driving tasks such as virtual intelligent voice assistants, vehicle entertainment, and communication and social tasks remains relatively limited. This gap poses potential constraints on realizing the value of advanced automated driving (see discussion of “Potential Challenge 1” below).

Overall, the reason why the research focuses evolve with the automation levels is that improvements in driving automation levels trigger the redistribution of driving responsibilities, subsequently generating new human factors issues. Accordingly, practitioners need to pay attention to conducting specialized usability research on task scenarios under relevant automation levels (see discussion of “Development Recommendation 2” below).

3.1.4. Research Status 4: The Scope of Task Scenarios in Usability Evaluation Research on Human–Machine Interaction in Intelligent Vehicles Is Expanding (Breadth) and Diversifying (Depth)

As shown in Figure 3, since 2017 (which coincides with the widespread adoption of intelligent vehicles), usability evaluation research on human–machine interaction in intelligent vehicles has not only increased significantly in quantity but also become increasingly diverse and complex in terms of the task scenarios involved. This is specifically manifested in the following five aspects:

• As indicated by the blue circles in Figure 3, since 2017, usability evaluation research focusing on primary driving task scenarios has begun to receive intensive attention, including driving takeover tasks, advanced driver assistance, and autonomous driving tasks. Although research on these primary driving tasks started relatively late, their growth momentum is strong, occupying a significant proportion (43.68%, 76/174) in a short period. This rapid convergence of research topics essentially means that as automated driving levels improve, designing safe, efficient, and trustworthy human–machine interaction for primary driving tasks has become a frontier issue in the current field of human–machine interaction usability evaluation.
• As indicated by the yellow circles in Figure 3, since 2017, usability evaluation research focusing on secondary driving task scenarios has also begun to receive intensive attention, for example, vehicle navigation tasks, warning systems, head-up display systems (HUDs), and driving condition prompt systems. This trend indicates that the academic community has begun to systematically study navigation, HUD, and other functions as an independent “secondary driving” task cluster to better support primary driving tasks.
• As indicated by the red circles in Figure 3, since 2017, the number of usability evaluation studies focusing on non-driving tasks has continued to rise, such as vehicle entertainment tasks, virtual intelligent voice assistants, other tasks, and function-setting tasks. This trend is closely coupled with the development of intelligent vehicle technology [60,61], where technological progress has not only enriched vehicle functions but also transformed the driver’s role. Therefore, how to enable drivers to safely and conveniently engage in non-driving activities during the driving process has become a new research hotspot.
• As shown in Figure 3, although current research has long focused on “traditional” task scenarios such as vehicle navigation tasks, vehicle entertainment tasks, virtual voice assistants, and vehicle function settings (such as air conditioning, windows, seat adjustment, etc.), before 2017, related research was scattered and failed to attract widespread attention. However, after 2017, research around these traditional tasks became more concentrated, with significantly increased numbers. This reveals that against the backdrop of intelligent cockpits (such as large touch screens and multimodal interaction) becoming mainstream configurations, the interaction methods for traditional task scenarios are being reshaped [62]. Therefore, the academic community urgently needs to re-examine and optimize the usability of these traditional interaction tasks with new evaluation perspectives to adapt to new hardware platforms and user expectations.
• As shown in Figure 3, in recent years, some studies have begun to focus on other tasks that were previously overlooked, namely, edge tasks or atypical tasks. Examples include the usability of touchscreen size, interface layout, and key design [63,64,65,66]; the usability of handwriting input box size in vehicle information systems [67]; the usability of T-shaped panel layout design [68]; the interface layout design of vehicle information systems [69]; the font design of central control interfaces [70]; and adaptive vehicle system design [71]. This shift reflects the improvement in user expectations for human–machine interaction usability, transitioning from meeting basic functions to pursuing ultimate, detailed experiences. Additionally, this also marks that human–machine interaction design concepts are further deepening toward “user-centered” approaches, emphasizing that optimizing the details of non-core functions, in addition to technological innovation, can also enhance overall user satisfaction and product competitiveness.

Overall, research on the usability evaluation of human–machine interaction in intelligent vehicles shows expansion trends in terms of both breadth and depth. This conclusion supports the previous survey findings of Su et al. [25], who noted that usability evaluation targets for automated vehicle human–machine interaction are becoming increasingly diverse. This diversity presents a scenario where traditional interaction interfaces and emerging interaction interfaces are developing concurrently. This trend not only reflects the academic community’s increasing attention to human–machine interaction usability issues driven by both technological progress and market demand but also means that future research needs to timely adjust usability evaluation research to meet increasingly complex evaluation needs (see discussion of “Development Recommendation 4” below).

3.1.5. Research Status 5: Usability of Vehicle Navigation Tasks Remains a Consistently Important Research Topic

As shown in Figure 3, compared to other task scenarios, vehicle navigation tasks represent one of the earliest research topics to receive attention in the field of human–machine interaction usability evaluation, and the field has maintained sustained research interest in this area. This finding reveals the foundational position of vehicle navigation tasks in the development of intelligent vehicle human–machine interaction and also indicates that with ongoing technological advancement and evolving user needs, vehicle navigation tasks will continue to be a focal point of innovation and research.

The developmental characteristics of vehicle navigation tasks benefit from two aspects. First, in terms of interaction frequency and accompaniment to driving tasks, vehicle navigation possesses inherent necessity and universality [72,73]. Specifically, unlike driving takeover or advanced driver assistance system (ADAS) tasks, which are triggered only under specific conditions, vehicle navigation spans the vast majority of complete driving journeys, representing a high-frequency, long-duration, continuous interaction task [74]. Particularly in urban environments or unfamiliar areas with variable road conditions and high decision-making pressure, the performance of navigation system human–machine interaction directly relates to driver task efficiency and subjective experience [75]. Second, with technological development, vehicle navigation is no longer a simple information presentation tool but plays the critical role of “digital navigator” and “safety co-pilot.” Therefore, the clarity, timeliness, and accuracy of vehicle navigation information delivery have crucial impacts on optimizing route planning, reducing driving anxiety, preventing operational errors, and ensuring driving safety [76,77,78]. Accordingly, practitioners should continue to focus on usability research for vehicle navigation tasks (see discussion of “Development Recommendation 5” below).

3.1.6. Research Status 6: Current Research Tends to Use Multimodal Interaction When Exploring the Usability of Driving Takeover Tasks While Highly Concentrating on Single-Modal Optimization and Application for Secondary Driving Tasks and Non-Driving Tasks

Based on industry experience, current application situations, and the necessity for in-depth analysis, we categorized the interaction modalities addressed in current usability evaluation research on human–machine interaction in intelligent vehicles into two major categories: unimodal and multimodal. Each category is further subdivided into various input and output modalities (detailed explanations are provided in Supplementary Materials—Supplementary File S1—Tables S4 and S5). Here, modality refers to a method and medium for transmitting information, used to interpret the sender’s state or intention [79,80,81,82]. Unimodal interaction refers to interaction that centers on only one modality for an input or output task, while multimodal interaction refers to the combined use of two or more modalities for an input or output task [48,80,82,83,84,85].

According to this classification, we found (as shown in Figure 4) that existing research tends to explore the usability of unimodal interaction (71.55%, 171/237), while research on multimodal interaction usability is relatively limited (27.85%, 66/237). However, when we conducted a cross-analysis using task scenario types as variables, we discovered the following two trends:

• In research related to secondary driving tasks and non-driving tasks, as many as 87.31% (117/134) of studies focus on unimodal interaction, while research exploring multimodal interaction usability is limited, accounting for only 12.68% (17/134). This indicates that existing research tends to optimize single interaction modalities when handling non-primary driving tasks such as infotainment and navigation settings.
• In contrast to the aforementioned trend, when exploring the usability of driving takeover tasks, existing research leans toward multimodal interaction. Specifically, multimodal interaction research on driving takeover tasks accounts for 57.38% (35/61), exceeding unimodal interaction research (42.63%, 26/61). The reason for this interaction modality application trend lies in the fact that the core objective of driving takeover tasks is to ensure that drivers can quickly and accurately receive and understand critical information, and multimodal interaction is key to achieving this goal [8,86,87,88]. Specifically, multimodal interaction fully utilizes the synergy and complementarity of information, enabling the more reliable transmission of critical information to drivers [88,89,90]. This not only enhances system fault tolerance but also effectively alerts drivers and helps them quickly integrate information to establish comprehensive situational awareness.

It should be emphasized that the above status reflects the current research focus rather than concluding that secondary driving tasks and non-driving tasks do not need or are not suitable for multimodal interaction. On the contrary, this may represent a research gap worthy of future in-depth exploration (see discussion of “Potential Challenge 3” below).

3.1.7. Research Status 7: From the Perspective of Modality Types, Existing Research Has Explored Diverse Interaction Modalities

As shown in Figure 4, current usability research on human–machine interaction in intelligent vehicles demonstrates significant diversity in the application of interaction modalities. Specifically, existing research covers 17 different interaction modalities (modality names are listed in Table 3 above), which are applied both individually and in multimodal combinations. Through an in-depth analysis of these interaction modalities, we found several noteworthy research status and development trends as follows:

• In terms of output modality applications, current research presents the following characteristics. First, visual output modalities represented by HUD interfaces, central control interfaces, and dashboard interfaces are the most widely applied. Among these, the frequent application of HUD interfaces is particularly noteworthy, as it fully demonstrates its unique advantages in reducing driver gaze deviation from the road and enhancing driving safety [91]. Second, among non-visual modalities, auditory feedback (such as speech and acoustic signals) has gained widespread recognition due to its advantages in reducing driving distraction risks [92]. Additionally, haptic vibration as an auxiliary feedback method has also received attention, which indicates that in driving contexts with high visual and cognitive load, timely haptic cues can effectively supplement information and enhance the immediacy and reliability of interaction [93,94].
• Beyond the aforementioned mainstream output modalities, some studies have begun to explore the application value of emerging interaction technologies. For example, some researchers have attempted to use changes in indicator light colors and flashing frequencies to convey warning or status information [95,96,97,98]. Meanwhile, some studies have started focusing on special modalities such as olfactory or haptic temperature feedback [99], aiming to relieve driving fatigue and help drivers maintain alertness through the release of specific odors or the provision of temperature stimulation. Additionally, other research has explored the feasibility of replacing traditional optical rearview mirrors with electronic rearview mirrors [100]. It should be noted that although these emerging modalities show certain application prospects, their applicability, stability, and user acceptance in real vehicle environments still require more comprehensive empirical validation.
• In terms of input modality applications, current research presents the following characteristics. First, touch screens, as one of the primary input devices, reflect the high dependence of current intelligent vehicles on information integration and operational flexibility in central control interaction through their high-frequency application. However, although touch screens possess rich interaction capabilities and visual expressiveness, their distraction risks during driving are equally noteworthy. Second, speech interaction has become a current research focus due to its convenience in freeing hands and being suitable for multitasking. Additionally, natural or physical interaction modalities such as gestures, steering wheel keys, and central control keys also play indispensable roles in specific scenarios [101,102], which proves their rationality and value.

Comprehensively, visual and speech modalities are currently the primary modalities of human–machine interaction in intelligent vehicles, while non-audiovisual interaction modalities such as haptic vibration, gestures, haptic temperature feedback, and olfactory modalities have also gained attention as auxiliary means. This application status is primarily driven by both technological progress and user needs. Specifically, on the one hand, with the continuous maturation of related technologies such as HUD, speech recognition, and touch screens, vehicle interaction systems can achieve more diverse and complex information display and input methods. On the other hand, as users’ demands for richer interaction experiences increase, some innovative interaction modalities have also begun to be applied in human–machine interaction systems for intelligent vehicles.

3.1.8. Research Status 8: Fragmented Research on Multimodal Interaction Usability in Intelligent Vehicles, Particularly for Driving Takeover Tasks

As shown in Figure 4, research on multimodal interaction usability in the intelligent vehicle field exhibits significant “fragmentation” and “divergence” characteristics. Taking the “driving takeover” scenario as an example, our investigation found that existing research has generated as many as 25 different multimodal combination schemes for conveying takeover requests. These schemes consider multiple channels, including visual (such as dashboard icons, central control screen animations, ambient light flashing), auditory (such as warning sounds, voice announcements), and haptic (such as seat vibration, steering wheel vibration) modalities, but their combination approaches vary widely and lack typical paradigms.

The reasons why current multimodal interaction usability research is fragmented and divergent are multifaceted. First, multimodal interaction in intelligent vehicles, as an emerging research topic, remains in the early stages of broad exploration regarding design paradigms, thus exhibiting “divergent” exploration characteristics [57]. In this context, many studies are driven by “technical feasibility” rather than “core user needs” [103]. Although the rapid iteration of interaction technologies provides practitioners with a rich “toolbox,” it also leads to the path dependency of multimodal interaction on specific technologies and fragmented outcomes [86]. A more fundamental problem lies in the fact that the academic community has not yet established widely accepted theoretical frameworks and standardized evaluation systems for multimodal interaction that are specifically tailored to vehicle environments [104,105,106]. This causes individual studies to often proceed based on different assumptions and evaluation dimensions, forcing practitioners to design personalized interaction schemes for specific tasks.

In summary, although existing research has explored rich multimodal interaction schemes, the essence is that current multimodal interaction usability research on intelligent vehicles lacks systematic guidance. This research status not only results in explorations without practical value but also constrains the development progress of the entire industry (see discussion of “Potential Challenge 4” below).

3.2. Potential Challenges in Usability Evaluation Research on Human–Machine Interaction in Intelligent Vehicles (Addressing Research Question 2)

In this section, we address Research Question 2: What potential challenges exist in current usability evaluation research on human–machine interaction in intelligent vehicles? Overall, the field currently faces four key potential challenges that constrain the advancement of usability research and the development of industry application practices to varying degrees. The following sections elaborate on each of these potential challenges.

3.2.1. Potential Challenge 1: Insufficient Attention to Non-Driving Tasks in Existing Research, Which Poses a Potential Constraint on the Value Realization of Advanced Automated Driving

As indicated in the above “Research Status 2 and Research Status 3,” current usability evaluation research on human–machine interaction in intelligent vehicles shows relatively low attention to non-driving tasks, particularly non-driving tasks in L4–L5 advanced driving automation scenarios. This research status reveals that the academic community’s response to the evolving role of autonomous vehicles shows some lag.

Specifically, as driving responsibility is completely transferred from human drivers to automated systems, the core attributes of vehicles will undergo a qualitative change, evolving from a single “driving tool” to a “third living space” after home and the workplace [48,107]. In this emerging space, users’ activity focus will shift toward non-driving tasks. However, existing research on users’ potential needs, behavioral patterns, and new interaction paradigms within this emerging space is still in its infancy, which creates a significant research gap.

This research gap directly constrains the “value realization” of advanced automated driving technology. This is because, as automation levels increase, the driver’s role transforms from “operator” to “monitor,” and even to “passenger,” with time previously focused on driving being largely freed up [3,48,108,109]. In this technological context, how to effectively utilize this “leisure time” will become the key factor determining the quality of future in-vehicle experiences [110]. In this regard, non-driving tasks as solutions will no longer be optional “secondary needs” but will directly define core elements of product differentiation, user satisfaction, and brand loyalty [48]. Therefore, if the academic community continues to lack research on interaction design, situational awareness, and service integration for non-driving tasks, the time value released by advanced automated driving will be difficult to convert effectively, which could potentially undermine its social acceptance. In response to this potential challenge, future research should strengthen usability studies on non-driving tasks (see the discussion in “Development Recommendation 1” below).

3.2.2. Potential Challenge 2: For L3 Conditional Driving Automation Scenarios, Existing Research Overly Focuses on Driving Takeover Tasks, Which May Mask Systemic Safety Risks

As shown in Figure 2, research in the academic community on L3 conditional driving automation scenarios has almost entirely converged on driving takeover tasks. This research landscape reveals a potential “tunnel effect”: while the academic community focuses on human factors performance and human–machine interaction strategies in takeover scenarios, it may have narrowed our overall understanding of L3 automated driving safety issues.

In fact, L3 conditional automated driving is not solely composed of isolated takeover events [111,112] but also depends on the interaction lifecycle throughout the entire usage process. Within this lifecycle, “non-takeover” interaction stages are equally critical for ensuring ultimate safety, for example, the activation and deactivation of the automation system, the continuous understanding of operational status, the clear communication of functional boundaries, the attention management of drivers during non-driving tasks, and even the user experience after completing driving takeover.

This reveals that there are insufficient usability studies on “non-takeover” human–machine interactions, which not only creates significant knowledge gaps but also leads to potential safety hazards. Specifically, neglecting “non-takeover” human–machine interaction can easily foster a “local optimization” safety cognitive fallacy, i.e., mistakenly believing that solving the driving takeover problem ensures the overall safety of L3 systems. However, system safety often depends on its weakest link. If drivers develop too much trust due to flawed human–machine interaction design during “non-takeover” phases or experience the continuous degradation of situational awareness due to prolonged disengagement from driving tasks, then even if the human–machine interaction system can accurately and timely issue takeover commands, drivers may be unable to respond effectively due to inadequate preparation.

In summary, failures in “non-takeover” human–machine interaction can undermine “takeover” quality and even trigger dangers. In response to this, future research should expand its focus from single driving takeover tasks to the entire interaction lifecycle of L3 conditional automated driving (see the discussion in “Development Recommendation 3” below).

3.2.3. Potential Challenge 3: Significant Gaps in Usability Research on Multimodal Interaction in Secondary and Non-Driving Tasks Will Constrain the Overall Development of Intelligent Vehicles

As indicated in “Research Status 6” above, current research shows limited exploration of multimodal interaction usability in secondary driving tasks and non-driving tasks (accounting for only 7.33%, 17/232). This application research status contradicts the development needs of the intelligent vehicle industry. Specifically, the intelligent vehicle industry is committed to transforming intelligent cockpits into immersive “third living spaces” that integrate entertainment, office work, and social functions [113,114]. The realization of this vision highly depends on the natural, flexible, and efficient human–machine interaction experience created by multimodal interaction technology [79,80,115,116,117]. However, multimodal interaction usability research, which serves as the application foundation, seriously lags behind and fails to provide the industry with the necessary design principles, evaluation paradigms, and empirical data support, creating a severe research gap [116].

This research gap brings multiple risks. First, without multimodal interaction usability guidance, newly added functions in vehicle systems may severely weaken the core competitiveness of products due to confusing interaction logic and excessive operational load [118]. Second, when the optimization of the multimodal interaction experience lacks practical evidence, technological innovation will fall into the trap of “innovation for the sake of showing off technology,” failing to translate into real user value. Over time, this will lead to waste of innovation resources and directly delay the evolution of intelligent vehicles toward “third living spaces.”

In summary, the absence of multimodal interaction usability research is not simply a matter of research-direction selection but also a potential risk concerning whether intelligent vehicles can achieve sustainable development in the future. Future research should focus on strengthening multimodal interaction usability research to meet the development needs of intelligent vehicles (see the discussion in “Development Recommendation 6” below).

3.2.4. Potential Challenge 4: The Absence of Usability Standards for Multimodal Interaction in Intelligent Vehicles Hinders Industry Development

As indicated in “Research Status 8” above, existing research on multimodal interaction usability shows clear fragmentation and inconsistency (i.e., research conclusions lack consistency), particularly for driving takeover scenarios. This phenomenon superficially reflects the diversification of interaction methods brought by technological progress [110], but it reveals a serious challenge at a deeper level: in the field of multimodal interaction usability for intelligent vehicles, a recognized set of design standards or basic paradigms has not yet been established. In other words, compared to the standardized steering wheel, current research has not yet established a unified or standardized multimodal interaction protocol for intelligent vehicles [109,119].

This research gap may trigger a series of negative effects ranging from industrial ecosystems to user safety. First, at the industrial level, the absence of usability standards for multimodal interaction in intelligent vehicles directly leads to significant research and development redundancy and innovation efficiency degradation. This not only causes the unnecessary extension of development cycles and increased validation costs but also impedes the collaborative efficiency of the industrial chain [83]. Second, at the user level, the inconsistency of multimodal interaction logic increases users’ learning costs and imposes unnecessary cognitive load. In high-load driving scenarios, any operational errors caused by interaction inconsistency may constitute potential and non-negligible traffic safety risks [82,120]. In response to this, future research should focus on addressing the systemic challenges caused by the absence of usability standards for multimodal interaction in intelligent vehicles (see the discussion in “Development Recommendation 6” below).

3.3. Development Recommendations for Usability Evaluation Research on Human–Machine Interaction in Intelligent Vehicles (Addressing Research Question 3)

In this section, we address Research Question 3: What are the development recommendations for future usability evaluation research on human–machine interaction in intelligent vehicles? Overall, future research needs to explore emerging interaction scenarios under advanced driving automation in depth and work toward building systematic and standardized usability frameworks and practical guidelines to address the growing technological complexity and user needs. The following sections elaborate on each of these development recommendations.

3.3.1. Development Recommendation 1: Future Research Should Strengthen Usability Studies of Non-Driving Tasks in L4–L5 Advanced Driving Automation Scenarios

As indicated in “Potential Challenge 1” above, although non-driving tasks are not necessary and sufficient conditions for safe driving [36,121], as vehicle intelligence levels increase (especially at L4–L5 levels), non-driving tasks become crucial in shaping in-vehicle experiences and enhancing user satisfaction. However, current research remains primarily confined to primary driving tasks and secondary driving tasks (see Figure 2), lacking attention to non-driving tasks. This research gap directly constrains the “value realization” of advanced automated driving technology.

In response, to advance the forward-looking development of human–machine interaction usability in intelligent vehicles, we propose the following recommendations. First, future research should direct more attention toward non-driving task scenarios, integrating multidimensional perspectives from human–machine interaction, psychology, sociology, and design studies to jointly explore the unlimited possibilities of future intelligent vehicles. Second, beyond traditional indicators such as safety and efficiency, future evaluation systems need to incorporate additional dimensions, including users’ emotional experiences, engagement, comfort, and satisfaction, to more comprehensively assess the quality of human–machine interaction design. Additionally, forward-looking research focused on high-potential non-driving scenarios should be conducted. For example, innovative interaction solutions should be developed and validated for specific application scenarios, such as in-vehicle collaborative work, multiplayer immersive gaming, and personalized health monitoring.

3.3.2. Development Recommendation 2: Future Research Needs to Conduct Specialized Usability Studies for Task Scenarios at Each Automation Level

As indicated in “Research Status 3” above, each transition in driving automation level is accompanied by profound changes in driver roles [110], responsibilities, and capabilities, thereby generating specific human–machine interaction issues. This evolution in terms of research focus is also confirmed in the evaluation dimensions of related task scenarios. Taking vehicle navigation tasks as an example, at L0–L2 levels (driving assistance), some studies on human–machine interaction usability evaluation focus more on performance indicators such as whether information is easily identifiable and whether it interferes with driving [39,122,123]. However, at L4–L5 levels (advanced driving automation), some studies on the usability evaluation of vehicle navigation tasks shift more toward trustworthiness and overall user experience [124,125]. This indicates that as automation levels increase, the focus of human–machine interaction usability evaluation transitions from “ensuring driving performance” to “building human-machine trust and enhancing comprehensive experience” [109,110].

This demonstrates that, considering the shifts in research focus and evaluation dimensions, practitioners must abandon the “one-size-fits-all” research approach. In other words, future human–machine interaction usability evaluations cannot follow a universally applicable set of criteria [126]; rather, an evaluation framework must be established that is deeply coupled with automation levels, task scenarios, and even real-time contexts. Practitioners need to not only develop specialized evaluation indicators for task scenarios at specific automation levels but also explore the consistency and evolutionary patterns of human–machine interaction usability across levels to ensure that users can obtain seamless, coherent, and trustworthy interaction experiences when switching between different automation modes.

3.3.3. Development Recommendation 3: For Task Scenarios Under L3 Level, Future Research Should Expand Its Scope from Single Takeover Tasks to the Complete Interaction Cycle of L3 Automated Driving

As indicated in “Potential Challenge 2” above, the limitation of current research lies in equating the human factors safety issues of L3 automated driving almost entirely with takeover request (TOR) problems. This “tunnel effect” has led to insufficient attention from the academic community to interaction design and human factors performance during “non-takeover” phases, creating systematic knowledge gaps and potential safety risks.

To address the above challenge, future research should expand its scope from single takeover tasks to the complete interaction cycle of L3 automated driving [105,127]. The research focus needs to shift toward building a holistic human–machine interaction framework that systematically explores safety strategies for both “takeover and non-takeover” phases [128]. For example, in future research, practitioners should work to develop and validate interaction interfaces that can clearly communicate system status, capability boundaries, and confidence levels; explore interaction solutions for maintaining appropriate situational awareness and engagement levels among drivers; and design an orderly “advance notice-handover” process for when the system anticipates that it will soon exceed its operational design domain, rather than waiting until the last moment to issue an emergency takeover request.

3.3.4. Development Recommendation 4: Future Research Needs to Promptly Adjust Usability Evaluation Work to Meet Increasingly Complex Evaluation Demands

As indicated in “Research Status 4” above, research on the usability evaluation of human–machine interaction in intelligent vehicles is experiencing a rapid development phase characterized by dual growth in both quantity and topics [129]. Technological innovations represented by automated driving have not only expanded new task scenarios but also transformed traditional task scenarios [44]. Additionally, growing market demands and user expectations are driving research in deeper and more refined directions [129]. These trends collectively pose challenges to the applicability and effectiveness of existing usability evaluation methods.

To address the above challenges, future research needs to make systematic adjustments in evaluation work. First, practitioners should closely monitor the development status of the human–machine interaction usability field in intelligent vehicles, recognizing that new tasks and changes bring new usability problems, and continuously expand usability evaluation research. Second, with updates in in-vehicle hardware and interaction technologies, it is necessary to re-examine the usability of traditional tasks such as navigation and environmental control under current technological contexts. Additionally, to improve research reproducibility and result comparability, the academic community should promote standardization work for human–machine interaction usability evaluation [21,130,131,132], including unified test scenarios, tools, and procedures.

3.3.5. Development Recommendation 5: Future Research Should Continuously Focus on How to Design and Evaluate Vehicle Navigation Tasks

As indicated in “Research Status 5” above, vehicle navigation tasks play a fundamental and important role in driving, which makes them a consistently prominent research topic in usability research on human–machine interaction in intelligent vehicles. Looking ahead, the research focus on vehicle navigation tasks is continuously deepening and expanding with the evolution of cutting-edge technologies. It is foreseeable that research on this task scenario will continue as a hotspot and present the following key development trends:

• With the increasing maturity of technologies such as augmented reality head-up displays (AR-HUDs), high-precision speech recognition, and natural language processing (NLP), the presentation and interaction methods of vehicle navigation information are triggering a paradigm revolution [133,134]. The focus of future usability research on vehicle navigation tasks will shift from traditional two-dimensional screen visual optimization toward the deep integration of visual, auditory, and even haptic feedback, aiming to build seamless, immersive, and highly intuitive multimodal interaction experiences.
• Vehicle navigation systems are transforming from independent functions into the “data brain” of intelligent vehicles, capable of providing data support for advanced driver assistance systems (ADASs) and obtaining real-time traffic information through vehicle-to-everything (V2X) communication [135]. Therefore, how to clearly integrate navigation, driving assistance, and external environment information on the interface to ensure that drivers can clearly and accurately understand the vehicle’s comprehensive operational status and decision intentions represents a key design and evaluation challenge.

3.3.6. Development Recommendation 6: Future Research Should Establish a Standardized Framework or Practical Guidelines for Multimodal Interaction Usability in Intelligent Vehicles to Meet Development Needs

As indicated in “Potential Challenge 3 and Potential Challenge 4” above, current usability research on multimodal interaction in intelligent vehicles faces a systematic challenge: limitations in application scenarios and the absence of usability standards. These research challenges constrain the development of human–machine interaction in intelligent vehicles.

To address current challenges, future research should not only expand the task scenarios applicable to multimodal interaction but also integrate relevant theoretical and empirical findings to establish a scientific and standardized multimodal interaction usability framework. The establishment of this framework can provide a unified benchmark for academic research, enhancing the comparability and cumulativeness of results, while also providing reliable design and validation guidance for industry to reduce development costs and improve system usability.

Specifically, we recommend that this framework include at least four levels: task scenarios, theoretical models, design guidelines, and evaluation systems. First, an effective usability framework should begin with the systematic organization and integration of core theories in the field, such as multiple resource theory, which explains the cognitive resource allocation (e.g., VACP model) and critical path analysis (CPA) used for task flow analysis [136,137,138], thereby establishing a solid theoretical foundation. Second, by referencing and refining Oviatt’s multimodal interaction design principles [139], Wechsung’s multimodal interaction evaluation framework [140], and Ma et al.’s multimodal interaction design methods for secondary driving tasks [116], a set of operational design guidelines and pattern libraries specifically for intelligent vehicle scenarios (such as driving takeover and infotainment system operation) should be developed. Additionally, this framework must clearly define its applicable specific task types and contexts, systematically covering all types of tasks from primary driving to secondary driving and non-driving tasks to ensure its comprehensiveness and practicality. Finally, to ensure the reliability of usability evaluation, it is necessary to develop a comprehensive evaluation methodology that includes subjective (such as standardized questionnaires), objective (such as task completion time and eye movement data), and physiological (such as heart rate) indicators.

4. Limitations of the Research

We must acknowledge that this study has limitations. First, the core limitation lies in the breadth of the literature coverage. Although we implemented a systematic literature search strategy that included cross-database searches across multiple professional databases and supplemented these with gray literature collection, we could not completely capture all relevant literature. In particular, due to subscription restrictions to SAE industry databases and other objective barriers, this study was unable to systematically and comprehensively incorporate literature from industry communities. This may have made our analysis more focused on academic research while providing insufficient coverage of the latest practices in industry. Additionally, during the literature screening, coding, and analysis phases, the process was inevitably influenced by the research team’s knowledge background and personal experience, which constitutes potential subjective bias.

To examine and address these limitations, we continuously searched for subsequently published relevant literature throughout the research process (with the most recent search conducted through June 2025). This approach not only supplemented our statistical data but also, more importantly, validated our research viewpoints. Specifically, the supplementary searches and analyses revealed that new research cases did not present viewpoints or recommendations that differed from those in our study. The primary impact of these new cases was updating the absolute percentages of relevant statistical data, without changing the relative importance of the statistical data or the research viewpoints derived from them. This indicates that our research findings have achieved theoretical saturation and demonstrate good stability in our conclusions.

Therefore, despite the limitations regarding literature coverage breadth, we still believe this study can provide a valuable reference for practitioners seeking to understand the current state of research on intelligent vehicle human–machine interaction usability evaluation. Additionally, we plan to conduct systematic updates of this research every five years to ensure the continued validity of our conclusions.

5. Conclusions and Future Work

This study employed a systematic review methodology to conduct an in-depth analysis of the current application status and challenges in usability evaluation research on intelligent vehicle human–machine interaction. Specifically, this research summarized and organized 158 applied research literature studies, with a focus on analyzing the automation levels, task scenarios, and interaction modalities addressed in these studies, and constructed a knowledge map for this field accordingly. Based on the knowledge map, this study sequentially discussed the development status, potential challenges, and development recommendations in the field of usability evaluation research on intelligent vehicle human–machine interaction. This systematic knowledge integration provides clear guidance and an empirical reference for future research innovation and technological applications in this field, offering significant value for promoting continuous innovation in this domain.

5.1. Research Conclusions

Overall, the current research on the usability evaluation of intelligent vehicle human–machine interaction presents a complex landscape characterized by “coexisting development and limitations.” On the one hand, the scope and depth of research continue to expand with technological advances. On the other hand, the research topics exhibit obvious lag and fragmentation characteristics. The following content summarizes the main conclusions formed by this study regarding the relevant research questions (RQs):

Regarding Research Question 1, “What is the current development status of usability evaluation research on intelligent vehicle human–machine interaction?” our main conclusions are as follows: (1) Existing research primarily focuses on L0-L2 level driving-assistance scenarios, with the evaluation emphasis concentrated on human–machine interaction related to primary driving tasks, while attention to secondary driving tasks and non-driving tasks remains relatively low. (2) It is noteworthy that each level of automated driving technology poses different requirements for drivers’ cognitive and operational capabilities, thereby generating human–machine interaction issues with hierarchical characteristics. (3) From a temporal development perspective, the task scenarios covered by current research continue to broaden in scope and become increasingly diverse in type, with the usability evaluation of vehicle navigation tasks consistently being an important topic in this field. (4) Current research tends to evaluate the usability of multimodal interaction when exploring driving takeover tasks while focusing more on the usability of unimodal interaction when evaluating secondary driving tasks and non-driving tasks. (5) It is worth noting with caution that against the backdrop of increasingly diverse interaction modality types, multimodal interaction usability research presents a relatively scattered state, lacking systematic integration.

Regarding Research Question 2, “What potential challenges exist in current usability evaluation research on intelligent vehicle human–machine interaction?” our main conclusions are as follows: (1) Existing research is highly concentrated on driving-related tasks, while insufficient attention is paid to non-driving tasks, which are becoming increasingly important in advanced automated driving scenarios (L4–L5), which affects the realization of technological market value. (2) For L3 conditional driving automation scenarios, current research exhibits a tendency to over-focus on driving takeover tasks. This singular usability research perspective may mask potential systemic safety hazards by overlooking the complex dynamic interactions of human–machine systems. (3) There are obvious gaps in the usability research on multimodal interaction in secondary and non-driving tasks, which will constrain the overall development of future intelligent vehicles. (4) The academic community has not yet formed unified multimodal interaction usability standards and methodologies, which has become a bottleneck hindering standardized technological development and innovation.

Regarding Research Question 3, “What are the development recommendations for future usability evaluation research on intelligent vehicle human–machine interaction?” our main conclusions are as follows: (1) There is an urgent need to expand research on interaction usability when users perform non-driving tasks (such as office work and entertainment) in advanced automation scenarios. (2) Future research needs to conduct more targeted and contextualized usability research and evaluation for typical task scenarios under different automation levels. (3) The research scope for L3 automated driving should expand from the single “takeover” node to a more complete interaction cycle, comprehensively examining user behavior and perception throughout the entire process of monitoring, receiving requests, executing takeover, stable control, and returning control authority. (4) As the integration and complexity of vehicle technology continue to increase, future research needs to synchronize with technological development, dynamically adjusting and optimizing usability evaluation frameworks, indicators, and tools to meet increasingly complex evaluation needs. (5) The usability of vehicle navigation tasks consistently remains key in affecting driving safety and experience. Future research should continue exploring optimal presentation methods and interaction modes for navigation information under different automation levels. (6) Future research should construct a standardized usability evaluation framework for intelligent vehicle multimodal interaction to provide unified and effective guidance for academic research and industrial practice.

5.2. Future Work

After clarifying the current “content” of usability evaluation research on intelligent vehicle human–machine interaction in this study, future research should focus on evaluation “methods.” Through this survey, we also found that despite the rich evaluation practices available, the field still lacks systematic organization and comparative analysis of existing usability evaluation models, tools, user cases, environments, participant characteristics, measurement indicators, and testing equipment (such as eye trackers and driving simulation platforms with different fidelity levels). These knowledge gaps limit practitioners’ ability to select and design optimal evaluation solutions based on specific requirements.

To address this gap, we plan to apply the systematic literature review methodology used in this study to conduct an in-depth comparative and critical analysis of evaluation methods in this field. The ultimate goal of this future research is to develop a comprehensive practical guideline that reveals the strengths, limitations, and applicable scenarios of different evaluation methods, thereby providing decision support for practitioners and promoting the standardization and improvement of evaluation quality across the entire field.