Understanding perceived ride safety and trust formation in robotaxi services under day and night conditions

Abstract

This study investigates how passengers perceive ride safety and develop trust in Robotaxi services in the absence of human drivers, with a focus on differences between daytime and nighttime scenarios. Drawing on the Elaboration Likelihood Model (ELM), we examined central cues (system transparency, perceived ride control) and peripheral cues (brand reputation, in-vehicle comfort) influencing safety perception and trust. Based on data from 563 real-world users, results show that all four cues significantly enhance perceived ride safety, which in turn fosters user trust. Notably, nighttime users relied more on perceived ride control and brand reputation, while daytime users prioritized in-vehicle comfort. System transparency remained a consistently strong predictor across both settings. The trust pathway from perceived safety proved stable regardless of time context. Theoretically, this study extends ELM to autonomous mobility, clarifying the roles of central and peripheral cues in real-world use. Practically, it offers targeted recommendations: enhance perceived control and brand safety at night, improve cabin comfort during the day, and ensure high system transparency at all times. These findings support broader acceptance and sustainable consumption of Robotaxi services, contributing to safer and more inclusive urban mobility systems.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-25722-w.

Introduction

Autonomous mobility services, particularly Robotaxis, are rapidly transitioning from experimental trials to global commercial deployment, marking a major shift in smart city transportation systems and sustainable mobility^1,2. As an emerging green travel mode, Robotaxis offers notable advantages, such as reducing traffic accidents caused by human error³, enhancing overall traffic efficiency, mitigating urban congestion, improving travel accessibility⁴, and significantly lowering environmental pollution through route optimization and energy-efficient driving^5,6.

However, despite these advancements, the widespread adoption and continued use of Robotaxis remain limited, primarily due to persistent concerns over the safety and trustworthiness of driverless technology^7,8. Unlike traditional taxi services, Robotaxis completely eliminates the human driver as a perceived “safety agent”⁹.As such, understanding how passengers perceive riding safety and building trust in real usage scenarios becomes crucial to enabling broader societal acceptance and sustainable consumption of Robotaxi services.

Most previous studies on autonomous vehicles have focused on users’ pre-adoption attitudes, technology acceptance intentions, and initial trust formation^9–12. These studies often rely on hypothetical or simulated scenarios¹³, exploring how potential users assess risk, psychological barriers, and trust prior to actual experience¹⁴. While such research offers insights into early-stage acceptance barriers, it largely overlooks the mechanisms by which safety perception and trust are developed during real ride experiences². Prior evidence suggests that actual vehicle performance and real-time interaction in authentic contexts often exert greater influence on users’ trust and attitudes than imagined or anticipated experiences¹⁵. In Robotaxi contexts, the absence of a human driver compels users to rely solely on their cognitive evaluations and sensory experiences to assess safety⁹, underscoring the need for deeper theoretical and practical exploration of trust-building mechanisms in real-world use.

To address this research gap, this study adopts the Elaboration Likelihood Model (ELM), which posits that individuals form attitudes through either cognitively demanding central cues or intuition-based peripheral cues when processing persuasive information¹⁶.ELM has been widely used to investigate how consumers form attitudes toward emerging technologies and adopt sustainable products or services^17–19, making it particularly suitable for analyzing how users develop safety perceptions and trust in uncertain, high-risk environments²⁰. In the context of Robotaxi rides, users form safety perceptions through both central cues factors, such as system transparency and perceived ride control, and peripheral cues, including brand reputation and in-vehicle comfort^21,22. However, the current literature has yet to clearly establish how these two types of factors function in real-world contexts, nor how their influence on perceived riding safety and trust may differ in practice.

Moreover, this study places particular emphasis on the contrast between daytime and nighttime riding contexts. Prior research suggests that nighttime environments are typically associated with higher levels of uncertainty and perceived risk²³, which may alter how users attend to and rely on different informational cues^24,25. Accordingly, this study adopts a comparative approach to explore how the absence of a human safety agent results in divergent perceptions of safety and trust between day and night scenarios.

To address the above research gaps, this study investigates the following key questions:

1. In the absence of a human driver, where do passengers of autonomous taxis derive their perceptions of safety during actual rides?

2. How do these perceptions of safety subsequently influence passenger trust?

3. Do central and peripheral cues factors exert significantly different effects under daytime versus nighttime scenarios?

To answer the above research questions, this study adopts the ELM theoretical framework to systematically examine how system transparency, perceived ride control, brand reputation, and in-vehicle comfort influence perceived riding safety and trust in Robotaxi ride contexts. It further clarifies the distinct mechanisms and contextual differences through which central and peripheral cues factors operate under daytime and nighttime scenarios.

From a theoretical perspective, this study makes three contributions: (1) it is the first to apply the ELM framework to real-world Robotaxi usage, thereby extending the model’s applicability within autonomous mobility research; (2) it clearly differentiates the central and peripheral cues factors involved in shaping passengers’ perceived riding safety, uncovering how these two cues function differently in real-life experiences; and (3) it reveals how environmental conditions, specifically the contrast between day and night, moderate users’ information processing and safety perception, thereby deepening theoretical understanding of trust formation in autonomous driving contexts.

Practically, this study offers context-specific design and operational strategies for Robotaxi service providers and designers. It outlines how services should be optimized under different usage scenarios to strengthen user trust, acceptance, and continued usage, ultimately contributing to the sustainable development and intelligent transformation of urban transportation systems.

Literature review, conceptual model, and hypotheses development

Robotaxi ride services

As autonomous driving technology continues to mature, Robotaxis has emerged as a representative form of shared autonomous mobility and a key direction in the development of smart transportation systems²⁶. In recent years, Robotaxi services have gradually transitioned from pilot trials to commercial operations in numerous cities worldwide^27,28. Notable examples include Waymo and Cruise in the United States²⁹, Olli by Local Motors, jointly developed in Germany and the U.S., and Apollo Go by Baidu and WeRide in China²². These services commonly operate using Level 4 or above autonomous driving technologies, eliminating the role of a human driver and fully transferring driving control to the autonomous system.

Existing research on Robotaxi services primarily focuses on three domains: technological performance, user acceptance, and market deployment strategies. First, studies on technological performance explore how Robotaxis enhances driving safety and system stability through sensor fusion, algorithm optimization, and environmental perception^30–32. Second, research on market deployment investigates operational models, pricing strategies, and regulatory policies that facilitate large-scale adoption of Robotaxi services in urban settings^29,33–35). Third, studies on user acceptance examine key factors influencing individuals’ willingness to adopt Robotaxis, such as perceived risk, trust in technology, and social norms^36–38. However, most of these studies concentrate on users’ perceptions prior to actual use, focusing on anticipated trust or acceptance formed through imagination or expectation. They largely overlook the complex mechanisms through which perceived riding safety and trust are constructed during real ride experiences^9–12.

The experience of riding in a Robotaxi differs fundamentally from that of a traditional human-driven taxi. Previous research indicates that human drivers in conventional vehicles serve as “safety agents,” offering psychological reassurance through interpersonal interaction, driving behavior, and emergency handling⁹. When Robotaxis removes the human driver, users’ sources of perceived safety fundamentally shift. This can lead to a strong sense of loss of control, particularly in nighttime or otherwise uncertain scenarios, where such feelings may be amplified²³. Although some studies have recognized this issue, limited theoretical and empirical work has explored how users perceive riding safety during actual Robotaxi trips, which factors most strongly influence safety and trust, or how these mechanisms differ across daytime and nighttime settings.

In summary, while Robotaxi services have received considerable attention from technical and market perspectives, the formation of perceived riding safety during actual use remains underexplored. Therefore, the present study focuses on real-world Robotaxi riding scenarios to systematically examine how users develop a sense of safety in the absence of a human driver as a safety agent and to explore how such safety perceptions differ between daytime and nighttime contexts—ultimately aiming to deepen the understanding of trust formation during the riding experience.

Elaboration Likelihood Model (ELM)

The Elaboration Likelihood Model (ELM) is a theoretical framework that describes how individuals process persuasive information and form or change attitudes through two distinct information-processing routes: the central cues and the peripheral cues. The choice of processing cues depends on an individual’s motivation and cognitive ability¹⁶. Specifically, the central cues involve thorough and analytical processing, focusing on the substance and logic of the information. In contrast, the peripheral cues rely on superficial cues or heuristics, such as source credibility, brand image, or presentation style, which require less cognitive effort.

In recent years, the Elaboration Likelihood Model (ELM) has been widely applied to investigate how consumers process information and form behavioral responses toward various emerging technologies³⁹, demonstrating strong theoretical explanatory power and cross-context applicability. In the consumer domain, studies have shown that in-depth evaluations of product quality and functionality via the central cues significantly enhance stable purchase behavior, whereas peripheral cues such as brand image, environmental ambiance, and promotional elements help trigger initial interest and trust more quickly^40,41. In the context of AI recommendation systems, factors like information accuracy and system transparency—central cues elements—play a crucial role in shaping long-term trust, while interface friendliness and interactive experience—peripheral cues—are more effective in establishing initial trust¹⁹. Similarly, in the context of green consumption research, central cues factors such as environmental knowledge and ecological awareness drive sustained pro-environmental behavior, while peripheral cues like brand perception and perceived benefits are more influential in short-term green consumption decisions^17,18.

In the field of shared mobility, existing ELM-based studies have shown that platform services and information quality—as central cues factors—enhance users’ awareness of safety risks, while peripheral cues such as system quality and brand reputation influence their perceived risk and intention to continue using the service through heuristic processing²¹. However, these studies primarily focus on risk expectations prior to the ride experience and still lack an in-depth exploration of how users form safety perceptions and trust through information processing during actual rides.

In Robotaxi scenarios—where the traditional safety agent (i.e., human driver) is entirely absent—users are required to process information independently to construct a sense of safety. This makes them more likely to engage with both central and peripheral cues simultaneously. Central cues processing involves deliberate assessment of system transparency and perceived ride control, requiring cognitive effort to judge how the autonomous system behaves in situations like lane changes, obstacle avoidance, or yielding to pedestrians. In contrast, peripheral cues—such as brand reputation, in-vehicle comfort, ambient lighting, cleanliness, and the presence of surveillance cameras—allow users to make quick safety judgments with minimal cognitive investment^21,22.

Moreover, in scenarios where the traditional driver role is absent, passengers become particularly reliant on environmental information. The uncertainty and heightened risk perception associated with nighttime conditions may increase users’ sensitivity to safety cues, thereby altering their reliance on central versus peripheral cues information²³. Existing studies have also pointed out that different environmental contexts will affect the selection and mode of action of information processing pathways^17,18.

In summary, this study adopts ELM as its theoretical foundation to systematically examine how users form safety perceptions through central cues (system transparency, perceived ride control) and peripheral cues (brand reputation, in-vehicle comfort) during actual Robotaxi rides, and how these perceptions contribute to trust. Particular attention is given to the differences in these processing mechanisms under daytime and nighttime scenarios. This research not only extends the applicability of ELM to autonomous mobility services but also offers theoretical and practical guidance for context-aware service optimization in Robotaxi operations.

Conceptual model and hypothesis development

Building on the systematic review of the aforementioned literature, this study adopts the Elaboration Likelihood Model (ELM) proposed by Petty and Cacioppo¹⁶ to investigate the specific mechanisms through which perceived riding safety and trust are formed in the context of Robotaxi services. A conceptual model is developed, in which both central cues factors (system transparency and perceived ride control) and peripheral cues factors (brand reputation and in-vehicle comfort) jointly influence users’ perceived riding safety, which in turn affects their trust in the service (see Fig. 1).

Fig. 1.

Research model.

Although the classical ELM framework primarily focuses on dual-route mechanisms in attitude formation¹⁶, there remains a lack of systematic research examining how these two cues operate during real-world autonomous ride experiences. Furthermore, given the potentially significant environmental differences between daytime and nighttime scenarios, this study incorporates day–night context as a moderating variable to explore how environmental changes may shape users’ safety perceptions and the pathways through which trust is established. Based on this framework, the following hypotheses are proposed.

Central cues and perceived riding safety

In the context of Robotaxi rides, Perceived system transparency (PST) refers to the vehicle’s ability to provide users with timely and sufficient information regarding its decision-making process and current operational status⁴². Previous research has shown that when autonomous systems disclose more information about ongoing actions and the rationale behind them, passengers are better able to understand the vehicle’s intentions, which reduces uncertainty stemming from information asymmetry and enhances confidence in the system’s safety^12,43.Additionally, access to rich driving information can improve passengers’ understanding and predictability of system behavior, thereby enhancing perceived safety and alleviating anxiety⁴².

This view is supported by Brandt et al.⁴⁴, who found that when Robotaxi systems enable users to gradually perceive “what is happening” and “what will happen next,” their comprehension of the system’s behavior improves significantly, subsequently enhancing their perceived safety. Therefore, the following hypothesis is proposed:

H1

The system transparency of a Robotaxi positively influences users’ perceived ride safety.

Perceived ride control (PRC) refers to the extent to which users feel they are able to influence or intervene in the vehicle’s operation during the ride⁴⁵. Studies have demonstrated that a sense of control is vital to the formation of perceived safety, especially in the face of new technologies or uncertain situations^46,47. In the context of Robotaxis, although users cannot directly operate the vehicle, they may still expect to retain some level of control in special cases, such as temporarily changing the destination, adjusting the route, or triggering an emergency stop⁴⁸. If the Robotaxi allows users to perceive that they can quickly and effectively influence the vehicle’s behavior when necessary, it can reinforce the perception of behavioral controllability and thereby enhance perceived riding safety⁴⁹. Thus, the following hypothesis is proposed:

H2

The perceived ride control of a Robotaxi positively influences users’ perceived ride safety.

Furthermore, this study considers potential differences in the Robotaxi experience between daytime and nighttime settings. Prior studies have indicated that nighttime environments—due to limited lighting and visual constraints—significantly increase users’ perceptions of risk and uncertainty, which may influence their information-processing strategies. In such contexts, users may be more inclined to engage in effortful cognitive evaluation to assess safety^25,50. As a result, nighttime users may be more sensitive to central cues such as system transparency and perceived ride control, which in turn may exert a stronger influence on their perceived safety. Based on this, the following hypotheses are proposed:

H3a

The effect of system transparency on perceived riding safety differs significantly between daytime and nighttime contexts.

H3b

The effect of perceived ride control on perceived riding safety differs significantly between daytime and nighttime contexts.

Peripheral cues and perceived ride safety

Perceived Brand reputation (PBR), as a typical peripheral cue, refers to users’ safety judgments based on the accumulated market reputation, word-of-mouth, or perceived influence of the service provider²¹. In the context of Robotaxi services, users often rely on the brand image of the operator to quickly infer service reliability, using it as a key signal for evaluating safety. Prior research has shown that brand reputation can effectively reduce uncertainty toward unfamiliar technologies or services⁵¹, particularly in the early stages of technological adoption when user experience is limited. In such cases, peripheral information can quickly and effectively enhance users’ trust and sense of safety^39,52. For instance, during the initial commercialization of autonomous driving, users tend to choose well-known operators with strong brand credibility to mitigate perceived risk²¹. Accordingly, this study proposes the following hypothesis:

H4

The brand reputation of a Robotaxi positively influences users’ perceived ride safety.

Second, Perceived in-vehicle comfort (PIC), as a directly perceptible spatial experience, typically affects passengers’ psychological state and safety judgment through visual and physical sensations^53,54. Research has shown that the physical environment and ambient atmosphere significantly influence passengers’ emotional state and psychological comfort^55,56. For example, Yang et al.⁵⁷ confirmed a direct relationship between passengers’ subjective perceptions of the ride environment and their perceived safety, emphasizing the critical role of ambient conditions in shaping risk perception. In autonomous mobility settings, features such as soft lighting, a clean and quiet interior, and visible security equipment can serve as peripheral cues that require little cognitive effort to interpret, thereby quickly enhancing perceived ride safety. Based on this, we propose the following hypothesis:

H5

The in-vehicle comfort of a Robotaxi positively influences users’ perceived ride safety.

This study also examines how the effects of peripheral cues on perceived ride safety differ between daytime and nighttime contexts. Prior research suggests that nighttime environments are often associated with greater uncertainty and heightened perceived risk. Under such conditions, users’ attentional resources are more limited, making it more difficult to engage in effortful information processing²⁴. As a result, users are more likely to rely on heuristic-based judgments derived from peripheral cues such as brand reputation and in-vehicle comfort. Therefore, these peripheral cues may play a more prominent role at night than during the day. Accordingly, the following hypotheses are proposed:

H6a

The effect of brand reputation on perceived ride safety differs significantly between daytime and nighttime contexts.

H6b

The effect of in-vehicle comfort on perceived ride safety differs significantly between daytime and nighttime contexts.

The influence of perceived ride safety on trust

Trust(TR) is widely regarded as a core determinant in users’ adoption of emerging technologies and services, particularly in contexts involving high uncertainty and risk, such as autonomous vehicles^13,26,58. In Robotaxi ride scenarios, passengers must confront the complete absence of the traditional human driver—who typically serves as a safety agent—and are therefore more reliant on their own subjective perceptions of vehicle safety to establish trust^9,12. Accordingly, understanding how perceived ride safety (PRS) influences user trust in Robotaxi services holds significant theoretical and practical value.

Prior research suggests that perceived ride safety reflects a user’s subjective evaluation of the controllability of potential hazards in a risk-laden context, encompassing an overall judgment of the safety of the technology, environment, and service²¹. A higher level of perceived ride safety typically corresponds to lower perceived risk, making users more likely to develop stable and positive trust attitudes⁵⁹. For instance, Nordhoff and Hagenzieker¹⁴, based on in-depth interviews with actual users, found that perceived safety in autonomous driving scenarios directly facilitated users’ trust in such systems. In the case of Robotaxi experiences, when users perceive that the system operates transparently and offers sufficient controllability or when the in-vehicle environment provides reassuring peripheral cues, trust in the Robotaxi service is likely to increase accordingly.

Therefore, the following hypothesis is proposed:

H7

Users’ perceived ride safety positively influences their trust in Robotaxi services.

Furthermore, considering the possibility of contextual differences between daytime and nighttime ride experiences, this study posits that the effect of perceived ride safety on trust may vary depending on the environmental setting. Prior studies have noted that nighttime conditions are often associated with heightened uncertainty and perceived risk, which may lead users to rely more heavily on perceived ride safety as a key cognitive basis for trust formation^23,60.

However, from the perspective of trust theory, safety is conceptualized as a structural and cognition-based judgment concerning the system’s reliability and service competence^61,62. Such cognitive antecedents are generally considered contextually robust, meaning their effects are less susceptible to significant fluctuations across different environmental conditions⁶³. Therefore, although daytime and nighttime settings may alter users’ overall perceptions of risk, the predictive power of perceived ride safety on trust may remain stable across contexts.

Based on this reasoning, the following exploratory hypothesis is proposed:

H8

The effect of perceived ride safety on trust does not differ significantly between daytime and nighttime contexts.

Methodology

Measurement and ethics declarations

The first part of the survey collected demographic information necessary for subsequent analysis. The second part measured the study constructs using items adapted from established and validated scales grounded in prior literature. All items were assessed using a five-point Likert scale. Since the original instruments were developed in non-Chinese contexts, professional translators were engaged to ensure the semantic accuracy and cultural appropriateness of the translated items. Detailed measurement items are provided in the Supplementary Information (Table S1).

This study was conducted in accordance with the Declaration of Helsinki and relevant local regulations. All participants provided informed consent prior to participation. The study protocol was reviewed and approved by the Ethics Committee of the Qingdao University of Science and Technology, Qingdao, Shandong Province, China. The committee determined that the study design complies with the Chinese government’s Implementation Rules for Ethical Review of Life Sciences and Medical Research Related to Humans (https://www.gov.cn/zhengce/zhengceku/2023-02/28/content_5743658.htm). As this research is a non-clinical, low-risk behavioral study based on anonymized user survey data, it does not involve any medical intervention, sensitive personal information, or foreseeable risk to participants. Therefore, in accordance with the above regulations, this study was exempt from formal ethical review.

Sample and data collection

To ensure that the sample reflects authentic and representative Robotaxi riding experiences, this study adopted an “ecological usage maturity”–oriented strategy for selecting urban research sites. Specifically, the selection criteria included:(1) Public accessibility: Robotaxi services must be officially open to the public, operationalized as availability through app-based booking beyond closed pilots or internal tests. (2) Coverage of key zones: The operational coverage should include core residential zones, commercial areas, and major transportation hubs, defined as covering at least one major residential district, one central business district, and one transport hub within the city. (3) Stable operation and user base: The service must have a stable operational track record and a sizable user base, operationalized as continuous operation for more than 12 months with active public adoption. (4) Regulatory approval: The service provider must have a clearly identifiable operational background and a traceable development history, defined as holding government-issued permits for commercial operation.

Based on these criteria, four Chinese cities—Changsha, Guangzhou, Wuhan, and Shenzhen—were selected as data collection sites. These cities are among the earliest in China to publicly deploy Robotaxi services, and their operations have reached a mature, routine stage. By the time of our data collection, these services had also entered a stage of near-complete driverless operation, with the vast majority of commercial rides conducted without a safety operator in the driver’s seat (2025 China Autonomous Driving Taxi Industry Market Research Report). This provides a solid ecological foundation for investigating users’ perceived ride safety and trust during actual ride experiences. Figure 2 illustrates an example of a Robotaxi vehicle operating in Wuhan. An overview of Robotaxi development in each city is presented in Table 1.

Fig. 2.

Robotaxi in operation in Wuhan.

Table 1.

Overview of robotaxi deployment and operation in selected Chinese cities.

City	Major operators	Year of public launch	Development path	Current operational status (2025)
Changsha	Baidu Apollo	2020	Pilot → Public Launch → Normalized Operation	Normalized service, covers multiple districts, available via app
Guangzhou	Baidu Apollo WeRide, Pony.ai	2019	Corporate park pilots → Public roads in Huangpu → Normalized Operation	Normalized operation, stable user base, mature ride experience
Wuhan	Baidu Apollo	2021	Hanyang trials → Economic Zone expansion → Fully driverless enabled	Fully driverless supported in key transport zones and residential areas
Shenzhen	BaiduApollo Pony.ai	2022	Pingshan pilot → Fully driverless testing → Public launch	Fully driverless operation in Pingshan, active app-based ride services

Operational information of the four cities is sourced from the 2025 China Autonomous Driving Taxi Industry Market Research Report⁶⁹.

Survey data were collected using Wenjuanxing (www.wjx.cn), a leading Chinese online survey platform with a large user base and robust quality control mechanisms. The platform is widely used for academic research in China. The questionnaire was distributed anonymously online. Participants were required to respond to a screening question: “Have you ever taken a Robotaxi (autonomous taxi)?” Only those who answered “Yes” were permitted to proceed with the full questionnaire, ensuring that all respondents had real-world Robotaxi riding experience.

To control for potential biases introduced by different ride times, the questionnaire included a demographic item: ‘During which period did you take Robotaxis most frequently?’ with options: ‘Daytime,’ ‘Nighttime,’ and ‘Other/uncertain time.’ To ensure that responses reflected a consistent riding context, only participants who selected ‘daytime’ or ‘nighttime’ were retained in the final sample, while those who selected ‘other/uncertain’ were excluded as invalid cases. After making this choice, respondents were required to answer all subsequent questions based on their selected riding context (daytime or nighttime).” To ensure structural balance for subsequent multi-group analysis, the research team implemented quota sampling. Prior Partial Least Squares Structural Equation Modeling (PLS-SEM) literature highlights that highly unequal subgroup sizes can compromise statistical power and bias group comparisons^64–66. Based on this principle, we adopted a conservative threshold by ensuring that the proportion difference between the “daytime” and “nighttime” groups did not exceed 10% of the final sample size, thereby maintaining comparability between groups.

All participants were informed of the study’s purpose, anonymity, data confidentiality, and their right to withdraw at any time. Only those who provided informed consent were allowed to proceed. To enhance data quality and participant attention, respondents were clearly informed that they would receive a monetary reward of 20 RMB (approximately USD 3) for completing the survey conscientiously. Rewards were distributed immediately after completion via the Wenjuanxing platform in the form of digital red envelopes, effectively reducing inattentive or fraudulent responses and enhancing data reliability.

To determine an appropriate sample size, this study adopted Daniel Soper’s a priori sample size calculator⁶⁷, setting an expected effect size of 0.3⁶⁸, statistical power at 0.95, significance level at 0.05, and a model including 6 latent variables and 27 observed indicators. The required minimum structural sample size was 88, with at least 236 responses needed for sufficient statistical power. Considering the need for group comparison, the final minimum total sample size was determined to be 472.

Data collection was conducted over a three-week period (from November 4 to November 30, 2024), yielding a total of 650 responses. After data collection, rigorous quality screening based on response time was applied to eliminate careless or invalid responses. Drawing from prior research and cognitive processing theories, the following exclusion criteria were implemented: (1) any single-item response time under 3 s; (2) total questionnaire completion time under 2 min. After cleaning, 563 valid responses were retained (effective response rate: 86.6%), providing a robust foundation for subsequent data analysis. Detailed demographic information is presented in Table 2.

Table 2.

Sample description.

Demographic information	Category	Frequency	%
Gender	Male	307	54.5
	Female	256	45.4
Age	18–21	11	2.0
	22–25	37	6.6
	26–29	168	29.8
	30–34	175	31.1
	35–40 41–45 46–50  > 50	52 53 49 18	9.2 9.4 8.7 3.2
Education	< High School	94	16.7
	High school or technical school	151	26.8
	Bachelor or Diploma	292	51.9
	≥ Master	26	4.6
City	Changsha	149	26.5
	Guangzhou	139	24.7
	Wuhan	146	25.9
	Shenzhen	129	22.9
Time	Day	292	51.9
	Night	271	48.1

N = 563.

Common method bias

Common method bias (CMB) is not a major concern in this study. First, Harman’s single-factor test revealed that the variance explained by the first unrotated factor was 13.218%, which is well below the 50% threshold⁷⁰. Second, all inner and outer variance inflation factor (VIF) values were below the recommended threshold of 3.3 (see Table 3), indicating that common method bias is unlikely to be an issue in this study⁷¹.

Table 3.

Control variable results.

Paths	β	T	P	0.025	0.975	Significance
Age → PRS	0.053	1.794	0.073	− 0.008	0.106	Not significant
Age → TR	− 0.042	1.111	0.267	− 0.114	0.031	Not significant
City → PRS	− 0.012	0.422	0.673	− 0.070	0.044	Not significant
City → TR	− 0.035	0.952	0.341	− 0.118	0.025	Not significant
Education → PRS	− 0.004	0.149	0.881	− 0.064	0.054	Not significant
Education → TR	0.003	0.090	0.928	− 0.072	0.071	Not significant
Gender → PRS	− 0.023	0.694	0.488	− 0.082	0.044	Not significant
Gender → TR	0.010	0.259	0.796	− 0.064	0.082	Not significant

Control variables

To account for potential demographic and contextual heterogeneity, we incorporated gender, age, education, and city as control variables in the PLS-SEM, each modeled as a single-indicator construct with direct paths to Perceived Ride Safety (PRS) and Trust (TR). As shown in Table 3, none of these control variables had significant effects (all p > 0.05), and the hypothesized structural relationships remained robust.

In addition, because participants were sampled from four cities, we performed a multi-group analysis (MGA) to examine city-level robustness. Following the MICOM procedure, measurement invariance was established, enabling valid group comparisons. The MGA results (Table 4) show that no structural path differed significantly between cities (all p > 0.05). This indicates that the model is invariant across city contexts and that intra-city correlations did not materially affect the findings.

Table 4.

Multi-group path comparisons across cities.

Path	C vs G (P)	C vs S (P)	C vs W (P)	G vs S (P)	G vs W (P)	S vs W (P)
PBR → PRS	0.730 (n.s .)	0.968 (n.s .)	0.344 (n.s .)	0.696 (n.s .)	0.555 (n.s .)	0.306 (n.s .)
PIC → PRS	0.533 (n.s .)	0.832 (n.s .)	0.767 (n.s .)	0.445 (n.s .)	0.398 (n.s .)	0.930 (n.s .)
PRC → PRS	0.831 (n.s .)	0.414 (n.s .)	0.822 (n.s .)	0.507 (n.s .)	0.999 (n.s .)	0.523 (n.s .)
PST → PRS	0.340 (n.s .)	0.428 (n.s .)	0.372 (n.s .)	0.800 (n.s .)	0.861 (n.s .)	0.959 (n.s .)
PRS → TR	0.450 (n.s .)	0.661 (n.s .)	0.435 (n.s .)	0.238 (n.s .)	0.101 (n.s .)	0.719 (n.s .)

n.s.: not significant (p > 0.05). C: Changsha; G: Guangzhou; W: Wuhan; S: Shenzhen.

Results

Considering that the model involves multiple latent variables and complex paths, the PLS-SEM method offers better robustness and analytical efficiency in handling complex structures. In addition, the PLS-SEM method is suitable for research oriented toward theory exploration and development, which aligns with the exploratory nature of this study. Therefore, this study adopts the PLS-SEM approach to analyze the research hypotheses. SmartPLS version 3.29 was used for data analysis.

Measurement model

We first evaluated the reliability and validity of the measurement model. As shown in Table 5, all item loadings exceeded the recommended threshold of 0.7, indicating that each item effectively represents its corresponding latent construct⁷².

Table 5.

Scale items and convergent validity analysis.

Constructs	Codes	VIF(OM )	VIF(IM)	λ	α	CR	AVE
PST	PST1	1.799	1.362	0.830	0.835	0.890	0.669
	PST2	1.721		0.796
	PST3	1.885		0.817
	PST4	1.816		0.827
PRC	PRC1	1.903	1.330	0.820	0.851	0.899	0.691
	PRC2	1.833		0.832
	PRC3	1.853		0.813
	PRC4	2.184		0.858
PBR	PBR1	2.196	1.272	0.830	0.888	0.918	0.691
	PBR2	2.345		0.863
	PBR3	2.071		0.822
	PBR4	2.060		0.820
	PBR5	2.095		0.821
PIC	PIC1	2.021	1.469	0.816	0.882	0.914	0.679
	PIC2	2.108		0.825
	PIC3	2.061		0.825
	PIC4	2.105		0.836
	PIC5	2.083		0.817
PRS	PRS1	2.060	1.000	0.842	0.869	0.911	0.718
	PRS2	2.131		0.852
	PRS3	2.168		0.855
	PRS4	2.021		0.841
TR	TR1	2.421		0.855	0.897	0.924	0.708
	TR2	2.421		0.857
	TR3	2.083		0.818
	TR4	2.192		0.836
	TR5	2.237		0.841

PST: Perceived System Transparency; PRC: Perceived Ride Control; PBR: Perceived Brand Reputation; PIC: Perceived In-Vehicle Comfort; PRS: Perceived Ride Safety; TR: Trust.

Next, internal consistency reliability was assessed using Cronbach’s Alpha and Composite Reliability (CR). The results revealed that all constructs had values above the 0.7 benchmark, demonstrating satisfactory reliability⁷³.

Convergent validity was confirmed as all Average Variance Extracted (AVE) values exceeded the recommended cutoff of 0.5⁷⁴. Discriminant validity was then examined using both the Fornell-Larcker criterion and the Heterotrait-Monotrait (HTMT) ratio. As shown in Table 6, the square root of each construct’s AVE was greater than its correlations with other constructs, satisfying the Fornell-Larcker criterion⁷⁵. All HTMT values were below the conservative threshold of 0.85⁷⁶, further supporting the discriminant validity of the constructs (see Table 7).

Table 6.

Fornell-Larcker criterion.

	PST	PRC	PBR	PIC	PRS	TR
PST	0.818
PRC	0.395	0.831
PBR	0.352	0.304	0.831
PIC	0.434	0.432	0.409	0.824
PRS	0.501	0.496	0.480	0.526	0.848
TR	0.496	0.469	0.349	0.583	0.508	0.842

Table 7.

HTMT criterion.

	PST	PRC	PBR	PIC	PRS	TR
PST
PRC	0.466
PBR	0.409	0.351
PIC	0.504	0.498	0.461
PRS	0.584	0.574	0.543	0.599
TR	0.573	0.536	0.390	0.655	0.575

Model fit, explanatory power, and predictive power

To assess model fit, this study employed multiple indices including SRMR, squared Euclidean distance (d_ULS), geodesic distance (d_G), and NFI. As shown in Table 8, the SRMR value of the model was 0.039, which is below the recommended threshold of 0.08, indicating a good model fit⁷⁷. In addition, both d_ULS and d_G values were below the upper limits of the confidence intervals, and the NFI value exceeded 0.8, further confirming that the model demonstrates a satisfactory level of fit⁷⁴.

Table 8.

Model’s fit.

	Saturated model
SRMR	0.039
d_ULS	0.564
d_G	0.230
NFI	0.911

In terms of explanatory power (Table 9), the R² value for perceived ride safety was 0.464, and the R² value for trust was 0.258, indicating that the model has strong explanatory ability for the dependent variables⁷⁸. Moreover, the Q² values obtained through the blindfolding procedure were 0.328 for perceived ride safety and 0.181 for trust, both well above the threshold of 0, suggesting that the model possesses good predictive relevance⁷⁹.

Table 9.

Explanatory power, and predictive power.

	Q2	R2	Adjusted R2
PRS	0.328	0.464	0.460
TR	0.181	0.258	0.257

Structural model analysis

This study tested the direct effects of each path in the model using bootstrapping with 5,000 subsamples. The results are presented in Table 10. Perceived system transparency (PST, β = 0.225, T = 4.843, p < 0.001), perceived ride control (PRC, β = 0.236, T = 5.206, p < 0.001), brand reputation (PBR, β = 0.235, T = 4.892, p < 0.001), and perceived in-vehicle comfort (PIC, β = 0.230, T = 4.744, p < 0.001) all had significant positive effects on perceived ride safety (PRS), supporting hypotheses H1, H2, H4, and H5.

Table 10.

Hypotheses test results.

Hypotheses	Paths	β	T	P	0.025	0.975	Results
H1	PST → PRS	0.225	4.843	0.000	0.136	0.317	Supported
H2	PRC → PRS	0.236	5.206	0.000	0.147	0.324	Supported
H4	PBR → PRS	0.235	4.892	0.000	0.143	0.329	Supported
H5	PIC → PRS	0.230	4.744	0.000	0.134	0.327	Supported
H7	PRS → TR	0.508	13.059	0.000	0.433	0.586	Supported

In addition, the direct positive effect of perceived ride safety (PRS) on trust (TR) was also strongly supported (β = 0.508, T = 13.059, p < 0.001), confirming hypothesis H7. The final structural model is shown in Fig. 3.

Fig. 3.

Structural model research model.

Multi-group analysis

This study further conducted a measurement invariance test prior to the multi-group analysis (MGA) to ensure the comparability and robustness of the model across daytime and nighttime scenarios. Based on previous studies and professional guidelines^58,66, as shown in Table 11, both configural invariance and compositional invariance of all constructs were effectively verified (C values close to 1, and the confidence intervals include 1). In addition, the mean differences (Equal Mean) and variance differences (Equal Variances) of the constructs fell within the confidence intervals, indicating that the constructs did not show significant differences between the two groups. Therefore, full measurement invariance was established, providing a solid theoretical and methodological foundation for the subsequent multi-group path analysis.

Table 11.

Assessment of measurement invariance.

Group	Con	Conf. Inv	Comp. Inv		PMI Est	Equal mean		Equal Var		FMI Est
			C = 1	CI		Diff	CI	Diff	CI
Day	PST	Yes	0.999	[0.997; 1.000]	Yes	0.016	[− 0.139; 0.140]	0.140	[− 0.191; 0.179]	Yes
VS	PRC	Yes	0.998	[0.998; 1.000]	Yes	0.047	[− 0.129; 0.139]	0.055	[− 0.145; 0.156]	Yes
Night	PBR	Yes	0.998	[0.998; 1.000]	Yes	− 0.108	[− 0.146; 0.143]	0.169	[− 0.170; 0.178]	Yes
	PIC	Yes	1.000	[0.998; 1.000]	Yes	0.092	[− 0.127; 0.142]	0.040	[− 0.148; 0.138]	Yes
	PRS	Yes	0.999	[0.999; 1.000]	Yes	0.045	[− 0.139; 0.135]	0.020	[− 0.211; 0.236]	Yes
	TR	Yes	1.000	[0.999; 1.000]	Yes	0.112	[− 0.137; 0.135]	− 0.093	[− 0.165; 0.168]	Yes

Con.: Construct; Conf. Inv.: Configural Invariance; Comp. Inv.: Compositional Invariance; CI: Confidence Interval; PMI Est.: Partial Measurement Invariance Established; Equal Mean: Equal Mean Value; Diff.: Differences; Equal Var.: Equal Variances; FMI Est.: Full Measurement Invariance Established.

The results of the multi-group analysis (see Table 12) showed that the paths from perceived ride control (PRC → PRS, difference = − 0.199, p < 0.05) and brand reputation (PBR → PRS, difference = − 0.199, p < 0.05) differed significantly between daytime and nighttime scenarios, supporting hypotheses H3b and H6a. Specifically, the effects of perceived ride control and brand reputation on perceived ride safety were significantly stronger in the nighttime scenario (β = 0.354 and β = 0.343, respectively, p < 0.001) than in the daytime scenario (β = 0.155 and β = 0.143, respectively, p < 0.01). Moreover, the path effect of perceived in-vehicle comfort (PIC → PRS) was significantly stronger in the daytime scenario (β = 0.310, p < 0.001) than in the nighttime scenario (β = 0.120, p < 0.05), with a significant path difference between the two groups (difference = 0.191, p < 0.05), thereby supporting hypothesis H6b.The effect of perceived ride safety on trust did not reach a significant difference across groups (p > 0.05), indicating that this path remained stable in both contexts, thus supporting hypothesis H8.

Table 12.

Multigroup analysis results.

Group	Hypotheses	Relationship	β0	β1	Coefficient difference	Results
Day(0)	H3a	PST → PRS	0.242***	0.162*	0.080	Not supported
VS	H3b	PRC → PRS	0.155**	0.354***	− 0.199*	Supported
Night(1)	H6a	PBR → PRS	0.143*	0.343***	− 0.199*	Supported
	H6b	PIC → PRS	0.310***	0.120*	0.191*	Supported
	H8	PRS → TR	0.546***	0.473***	0.073	Supported

*P < 0.05; **P < 0.01; ***P < 0.001.

However, the paths from perceived system transparency to perceived ride safety (PST → PRS, path difference = 0.080, p > 0.05) and from perceived ride safety to trust (PRS → TR, path difference = 0.073, p > 0.05) did not exhibit significant differences across scenarios, and thus hypothesis H3a was not supported.

Discussion

Theoretical contributions

This study offers important theoretical contributions to the understanding of user trust mechanisms and sustainable consumption behavior in the field of autonomous driving services, particularly within the emerging smart mobility context of Robotaxi ride experiences. The theoretical innovations of this study are reflected in the following aspects:

First, this study breaks through the paradigm in previous autonomous driving research that primarily focused on users’ initial acceptance intentions and anticipated trust in technology, and it is the first to focus on the mechanism by which users develop perceived ride safety and trust during actual Robotaxi ride experiences. Previous studies were mostly based on hypothetical scenarios or virtual environments, exploring how potential users form trust and acceptance intentions based on expectations before real experiences^13,26. In contrast, this study clearly reveals how, in the context of losing the human driver as a traditional safety agent, users construct their perceived ride safety during actual rides, which in turn influences their trust in Robotaxi services. This research expands the application boundary of existing theories on autonomous vehicle acceptance, demonstrates stronger contextual validity, and provides important theoretical value for promoting sustainable consumption and broader social acceptance of Robotaxi services in smart cities.

Second, drawing upon the Elaboration Likelihood Model (ELM), this study is the first to systematically analyze how users form perceived ride safety through central cues (system transparency, perceived ride control) and peripheral cues (brand reputation, in-vehicle comfort) during real Robotaxi ride experiences. While existing studies have individually discussed factors such as technological transparency or brand reputation and their impacts on trust ^21,44, this study innovatively integrates these factors within a unified theoretical framework. From the perspective of information processing routes, it offers a comprehensive explanation of how users form safety judgments and trust attitudes toward emerging mobility services. This theoretical integration not only deepens the understanding of the user trust formation process but also enriches the theoretical foundation for attitude formation toward sustainable smart mobility services^24,80.

Third, this study explicitly explores and empirically verifies the differences between daytime and nighttime scenarios in the context of Robotaxi rides. The findings show significant variations across scenarios in the effects of perceived ride control, brand reputation, and in-vehicle comfort. Specifically, users were more sensitive to perceived ride control and brand reputation during nighttime, highlighting that in high-risk and uncertain environments, users are more likely to rely on specific peripheral cues to make quick safety judgments. Furthermore, in-vehicle comfort was found to be more influential in the daytime context, expanding the current theoretical understanding of environmental factors in autonomous driving service research. Notably, system transparency, as a central cue, did not show significant differences between the two scenarios. This indicates that regardless of environmental changes, system transparency consistently serves as a fundamental element in users’ safety perceptions, demonstrating strong contextual robustness⁴². This finding not only deepens the theoretical understanding of how contextual factors influence information processing and trust formation but also offers new theoretical insights for the contextualized design of autonomous driving services, emphasizing the role of environmental cues as key moderators in theories of sustainable consumption behavior.

Finally, this study finds that the core path from perceived ride safety to trust remains robust across daytime and nighttime scenarios. Although different contexts may alter the relative importance of specific information cues for users, perceived ride safety, as a cognitive antecedent in trust formation, demonstrates a stable path relationship. This finding further enriches the understanding of the coexistence of contextual robustness and dynamic sensitivity in trust theory⁸¹, especially by providing theoretical support in contexts involving the acceptance of emerging technological services.

In summary, this study has contributed to the theoretical development in the field of trust mechanisms and sustainable consumption behavior in autonomous driving services by: (1) focusing for the first time on the construction mechanism of trust and perceived ride safety in real ride experiences; (2) integrating and analyzing the effects of central and peripheral cues on users’ perceived ride safety based on the ELM theoretical framework; (3) clarifying the moderating role of daytime and nighttime contextual differences in the paths leading to perceived ride safety and trust; and (4) revealing the contextual robustness of the path from perceived ride safety to trust. These contributions provide a solid theoretical foundation for future research on Robotaxi services in smart cities.

Practical contributions

The findings of this study offer several valuable practical implications for Robotaxi service providers, designers, and relevant policymakers, contributing to the market adoption of Robotaxi and the promotion of sustainable consumption behavior.

First, given that perceived ride safety significantly influences users’ trust in Robotaxi, service providers should pay attention to the presentation of both central cues and peripheral cues during the ride experience. At the level of central cues, operators need to enhance the transparency of the autonomous driving system’s decision-making processes. For example, they should clearly communicate the planned navigation routes, upcoming driving behaviors such as lane changes or deceleration, and the underlying decision logic. This helps improve users’ understanding and predictability of vehicle behavior. In addition, operators should develop user-friendly and intuitive human–machine interfaces that enable users to perceive a clear sense of ride control in special situations such as emergency stops or route adjustments. This can effectively reduce psychological anxiety and safety uncertainty caused by the absence of a human driver, especially in nighttime or other high-risk contexts, thereby fostering long-term public trust in sustainable mobility services.

Second, this study finds that peripheral cues, such as brand reputation and in-vehicle comfort, also have a significant impact on users’ perceived ride safety. Therefore, autonomous driving service providers should actively engage in brand-building and reputation management. Through precise marketing strategies and effective public relations, they should establish a trustworthy, safe, and stable brand image that quickly reduces consumers’ perceived risk and uncertainty when facing emerging transportation services. In addition, the interior design of the vehicle should place greater emphasis on enhancing the spatial experience from both visual and physical perspectives. For example, maintaining cleanliness and hygiene inside the vehicle, creating a warm and comfortable lighting atmosphere, and ensuring the presence of visible safety monitoring devices and clear safety instructions can help users quickly perceive safety and comfort without requiring excessive cognitive processing. This kind of service design can not only directly enhance the passenger experience but also promote the social recognition and market adoption of Robotaxi services as a sustainable transportation option in smart cities.

Third, this study clearly identifies the differences in the formation paths of perceived ride safety under different contextual scenarios of day and night, highlighting the need for autonomous driving service providers to develop context-specific strategies. Especially at night, when risk perception and uncertainty are higher, service providers should enhance the presentation and optimization of perceived ride control and brand reputation. For example, they should improve the operability and visibility of nighttime interfaces to strengthen users’ sense of control, and reinforce the transmission of safety and reliability information about the brand in nighttime service scenarios to quickly alleviate user anxiety and stabilize and improve the market acceptance of nighttime autonomous driving services. In contrast, users in daytime scenarios tend to focus more on the details of the in-vehicle environment. Therefore, daytime service design should pay closer attention to users’ intuitive perceptions and comfort experience of the vehicle environment under sufficient lighting and lower perceived risk. By applying differentiated and refined design strategies, service providers can further strengthen users’ sense of safety and satisfaction^82,83. In addition, this study finds that system transparency does not exhibit significant differences across daytime and nighttime contexts, reflecting its universality and necessity as a key factor in service design. This means that service providers must ensure consistency and efficiency in information transparency design regardless of the external environment, as system transparency serves as the foundation for passengers’ sustained trust and safety perception, and it remains stable across different contextual conditions.

Finally, this study provides clear policy implications for policymakers and regulatory bodies. Given that perceived ride safety is critical to public trust in autonomous driving services, regulatory agencies may formulate and implement relevant standards and incentives that encourage system transparency and enhance users’ sense of control. These measures can guide enterprises to promote user-centered service design. At the same time, policymakers may highlight the importance of peripheral cues such as brand reputation and in-vehicle environment in public communication strategies, in order to more effectively reduce psychological resistance to new technologies and promote the rapid social acceptance and sustainable consumption of autonomous driving services.

In summary, by identifying the differentiated impacts of information cues on perceived ride safety during Robotaxi ride experiences, and clarifying the mechanisms of action under different daytime and nighttime scenarios, this study offers strong and practical guidance for the design optimization, policy formulation, and market promotion of Robotaxi services in the context of smart cities. These insights contribute to the sustainable development of urban transportation systems and the positive transformation of public consumption behavior.

Limitations and future research directions

Although this study offers certain innovations at both theoretical and practical levels, it still has some limitations that point to clear directions for future research.

First, this study employed cross-sectional data, which limits the ability to explore the dynamic evolution and potential mechanisms of change in users’ perceived ride safety and trust after long-term use of Robotaxi services. Future research could adopt longitudinal designs or field experiments to examine the dynamic changes in user trust as their experience accumulates, in order to gain a more comprehensive understanding of the continuity and mechanisms of trust formation.

Second, the participants in this study were limited to users within a specific region who had real-life experience with Robotaxi, which may affect the generalizability of the findings. Although driverless operation was already the mainstream model in these cities, a small number of pilot vehicles might still include onboard operators. While city- or operator-level subgroup analyses could be valuable, our sample sizes within each city were insufficient to support such comparisons. Future research with larger datasets should therefore explicitly differentiate between “fully driverless” and “with-operator” conditions, and where possible, conduct subgroup analyses across cities and operators.

In addition, this study did not include individual difference variables, such as technology readiness, risk preference, or prior experience, which may serve as potential moderators of perceived ride safety and trust. Considering that individual differences may significantly influence users’ acceptance of new technologies, future studies could incorporate such moderating variables to explore how personal characteristics specifically affect users’ selection of information processing routes and their trust formation mechanisms. This would provide more refined theoretical support for personalized service design.

Finally, although this study examined the differential effects of daytime and nighttime scenarios on users’ perceived ride safety, it did not address more complex or special traffic environments, such as severe weather conditions or peak-hour congestion. Future research could further investigate the impact of diverse environmental conditions on users’ safety perception pathways, to better understand the interaction between environmental factors and trust mechanisms.

In conclusion, these limitations provide clear and valuable directions for future research and can help further deepen theoretical understanding and practical application regarding user experience and trust formation in autonomous taxi services.

Conclusion

This study investigates how users perceive ride safety and build trust during actual Robotaxi rides in the absence of a traditional safety agent, namely the human driver, with a particular focus on the differences between daytime and nighttime ride contexts. Based on the Elaboration Likelihood Model (ELM) framework, the study examines the mechanisms of central cues (system transparency and perceived ride control) and peripheral cues (brand reputation and in-vehicle comfort).

The findings indicate that both system transparency and perceived ride control (central cues), as well as brand reputation and in-vehicle comfort (peripheral cues), significantly enhance users’ perceived ride safety, which in turn promotes trust in Robotaxi services. Moreover, the study clearly identifies that the effects of perceived ride control, brand reputation, and in-vehicle comfort on perceived ride safety are significantly moderated by the ride context (daytime vs. nighttime). In nighttime scenarios, users rely more heavily on perceived ride control and brand reputation to alleviate perceived risks and uncertainties, while in-vehicle comfort plays a more prominent role in daytime scenarios. In addition, system transparency consistently demonstrates a stable and significant positive influence across both contexts, highlighting its robustness as a core factor in service design. The study also confirms that the path from perceived ride safety to trust remains stable across different contexts.

From a theoretical perspective, this study moves beyond previous research that focused primarily on users’ initial acceptance intentions and anticipated trust. It innovatively applies the ELM framework to clarify the differentiated effects of central and peripheral cues on perceived ride safety during real experiences, while also exploring the moderating role of environmental factors (day vs. night). These findings enrich the theoretical understanding of safety perception and trust formation in the autonomous driving domain.

On a practical level, this study provides clear and differentiated guidance for the contextualized design of Robotaxi services. In nighttime scenarios, emphasis should be placed on enhancing perceived ride control and communicating brand safety information in order to quickly reduce users’ risk perception. In contrast, daytime services should focus more on improving in-vehicle comfort to meet users’ needs for intuitive and pleasant ride experiences. At the same time, operators must consistently ensure the quality of system transparency design to maintain user trust throughout the day. The findings also offer valuable guidance for policymakers, suggesting that promoting system transparency and enhancing users’ sense of control can foster broader acceptance and sustainable consumption of autonomous driving services in smart cities.

In conclusion, this study not only deepens the theoretical exploration of autonomous mobility services but also provides clear and actionable recommendations for the practical application of Robotaxi services within the context of smart cities and sustainable consumption. It contributes to the broader social acceptance and sustainable development of autonomous driving technologies.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Y.L. was responsible for conceptualization, methodology, formal analysis, original draft preparation, and visualization. H.L. contributed to software development, data curation, and visualization. Y.P. provided resources, supervision, project administration, and contributed to review and editing. Y.L. and Y.P. jointly revised the manuscript. All authors have read and approved the final version of the manuscript.

Funding

This work is supported by research projects of Qingdao University of Science and Technology (WST2021020) and Kookmin University.

Data availability

All data generated or analyzed during this study are included in this published article and its Supplementary Information files.

Declarations

Competing interests

The authors declare no competing interests.