关键词: 具身智能, 隐私政策, 友好度, 物元可拓模型

Abstract:

[Purpose/Significance] The rapid development of embodied intelligence has fundamentally transformed the landscape of privacy protection. Unlike traditional digital services, embodied intelligent devices deeply embed themselves into users' physical environments, continuously collecting multimodal data through sensors and autonomously executing actions. This physical embodiment, social interaction, and action autonomy fundamentally reshape privacy risk boundaries, rendering existing privacy policy evaluation frameworks inadequate. While privacy policy evaluation has been extensively studied for websites and mobile applications, research specifically addressing the unique challenges of embodied intelligence remains scarce. This study aims to fill this gap by developing a user-friendliness evaluation framework tailored for embodied intelligence privacy policies. The theoretical innovation lies in extending the friendliness concept from traditional digital spaces to physical interaction scenarios, systematically incorporating the technical characteristics of embodied intelligence into the evaluation system. The practical significance is to reveal the current state of privacy policy design in the embodied intelligence industry, identify compliance gaps and interaction design deficiencies, and provide actionable recommendations for policy optimization and user rights protection. [Method/Process] This study constructed a comprehensive evaluation system grounded in user experience theory and legal requirements. Theoretically, we introduced Peter Morville's user experience honeycomb model as the foundational framework, which comprises seven interrelated dimensions: useful, usable, desirable, findable, accessible, credible, and valuable. This model provides systematic coverage of the entire user journey from initial perception to trust establishment. Legally, the indicator system refers to the personal information protection law and information security technology - personal information security specification. Methodologically, we innovatively adopt the matter-element extension model as the core evaluation approach. Compared to traditional methods like analytic hierarchy process and fuzzy comprehensive evaluation, this model offers unique advantages in handling multi-indicator, multi-grade, and subjective perception data, providing precise association degrees between each indicator and evaluation grades. Empirically, we collected 336 valid questionnaires from users with experience using embodied intelligence products, consulting five experts (three in user information behavior and two in human-computer interaction) to establish evaluation grade standards. Six mainstream embodied intelligence applications across different product categories were selected as evaluation objects, including DJI (autonomous drones), Ecovacs (sweeping robots), and NIO (in-vehicle intelligent systems). [Results/Conclusions] The evaluation results reveal three key findings. First, the overall user-friendliness of current embodied intelligence privacy policies ranges from moderate to good, exhibiting a middle-clustered distribution with no applications achieving excellent ratings, indicating the industry is transitioning from basic compliance to experience optimization. Second, regarding core obligation indicators, basic compliance is generally achieved, yet significant deficiencies exist in deepening user rights (e.g., inclusive protection for special groups) and transparent risk communication (e.g., specific disclosure of physical security risks). Notably, Yushu Technology and Qianglang Smart received poor ratings for user consent respect, as their interfaces lack substantive refusal options. Third, interaction friendliness indicators show uneven performance: DJI and Ecovacs excel in built-in functionality (achieving excellent ratings by integrating policy reading and settings within the app), while most applications underperform in visual design, language simplicity, and navigation convenience. This study validates the applicability of the matter-element extension model in privacy policy evaluation and provides actionable recommendations: optimize consent mechanisms to ensure clear refusal options, enhance visual hierarchy and layout design to reduce cognitive load, and specifically address unique embodied intelligence risks such as physical security and environmental monitoring. Future research should expand sample sizes, incorporate multimodal sensing data, and explore dynamic real-time evaluation mechanisms.

Key words: embodied intelligence, privacy policy, friendliness, matter-element extension model