A Taxonomy of Environmental Cues for Quadruped Robot Navigation in Construction Environments Informed by Human Operator Behavior

A Taxonomy of Environmental Cues for Quadruped Robot Navigation in Construction Environments Informed by Human Operator Behavior | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article A Taxonomy of Environmental Cues for Quadruped Robot Navigation in Construction Environments Informed by Human Operator Behavior Liqun Xu, Geonho Kim, Zhenhua Zhu This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8254973/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Autonomous robot navigation in construction environments remains a significant challenge due to their complex spatial configurations, dynamic obstacles, and socially interactive nature. Existing approaches rely on pre-programmed rules and costs to interpret environment cues (e.g., maintaining a minimum distance of 0.3 meters from obstacles, assigning higher costs to uneven terrain or proximity to obstacles). However, there is no empirical evidence identifying which specific cues are essential for safe and effective navigation in these settings. Research in autonomous driving has demonstrated that analyzing human drivers’ behavior through egocentric video can reveal the environmental cues critical to safe navigation. Inspired by this approach, we investigate theenvironmental cues that inform human navigation behavior during manual control of a quadruped robot on an active construction site. We deployed a manually controlled quadruped on an active site and collected egocentric video using naturalistic field methods. After that, we analyzed the videos using thematic analysis to identify the environmental features that triggered operators’ actions, such as stopping and turning. Our analysis resulted in a taxonomy of three cue categories: topography cues, which pertain to ground surface texture and terrain geometry; obstacle and hazard cues, which involve both static and dynamic objects; and communicative cues, which include human gestures and embedded visual indicators. These findings provide an empirically grounded understanding of the environmental cues that are essential for safe robot navigation in active construction environments and, in turn, inform the design of perception modules for context-aware, socially competent autonomy. Autonomous navigation construction robots environmental cues robots in the wild video analysis Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 1 Introduction The use of mobile robots in construction management is a rapidly growing area of interest [ 1 ]. Quadruped robots, in particular, offer significant mobility advantages over wheeled or tracked platforms in the unstructured and complex environments typical of construction site [ 2 ]. These robots have the potential to automate tasks such as site monitoring, surveying, and logistics [ 1 ]. However, realizing this potential depends critically on enabling safe autonomous navigation in construction environments that are not only physically complex but also socially interactive and in constant flux [ 1 ]. Autonomous navigation in construction sites presents unique challenges. These environments are characterized by evolving layouts, temporary and often ambiguous obstacles, and frequent interactions with human workers. Most current navigation systems rely on geometry-based methods, such as LiDAR-based SLAM, to create occupancy maps and plan collision-free paths [ 3 ], [ 4 ]. While effective in controlled settings, these systems often fail in construction environments. This is because elements like caution tape and temporary barriers may appear benign to geometry-based navigation systems but hold significant operational meaning for safe navigation [ 5 ]. To mitigate these limitations, recent efforts have focused on augmenting geometry-based methods with semantic perception, using deep learning models for object detection and segmentation [ 6 ], [ 7 ]. These hybrid systems have improved the robot’s ability to recognize contextual categories such as equipment, terrain, and people, enabling more informed navigation strategies [ 8 ], [ 9 ]. Despite these advances, the decision-making logic in existing systems still relies on pre-programmed rules to interpret environmental cues (e.g., avoiding entry into the falling zone of materials lifted by a tower crane) [ 4 ], [ 10 ]. Yet, there is no empirical evidence identifying which cues are essential for safe navigation in construction environments. As a result, navigation systems that appear effective in one scenario often fail to generalize across sites with different layouts, activities, or social dynamics. This lack of an empirically grounded understanding of environmental cues represents a fundamental barrier to the development of robust and transferable navigation frameworks for construction robotics. Research in autonomous driving provides a practical approach to addressing this challenge. In that field, egocentric video analysis has been used to link driver behaviors, such as slowing down, turning, or stopping, to specific environmental cues including pedestrians, signage, and road conditions [ 11 ], [ 12 ]. By capturing and annotating the relationship between human decision-making and environmental cues, these studies have identified the environmental cues critical for safe autonomous navigation. Building on this insight, we argue that a similar approach can be applied to construction robotics. By analyzing the navigation operations of human operators during manual control and linking their operations to environmental cues, it becomes possible to uncover the cues that autonomous navigation systems must recognize to perform reliably in construction settings. In this study, we deployed a manually controlled quadruped on an active site and collected egocentric video using naturalistic field methods. From these recordings, we traced operator actions to the visual and contextual information that occasioned them. Through thematic analysis of the visual and contextual information, we distill these observations into a taxonomy of environmental cues that humans use for navigation. We identify three categories of cues: topography cues, which pertain to ground surface texture and terrain geometry; obstacle and hazard cues, which involve physical impediments or safety risks; and communicative cues, which include human gestures and embedded visual indicators. The findings provide empirical evidence for which cues are essential to safe navigation on construction sites and, in turn, inform the design of perception modules for autonomous navigation in these environments. 2 Related works 2.1 Quadruped robots in construction management Quadruped robots are increasingly being adopted in construction for their unique mobility and ability to navigate complex, unstructured environments [ 1 ]. Unlike wheeled robots, quadrupeds can traverse stairs, debris, and uneven terrain, making them well-suited for use in construction sites [ 1 ]. Researchers have applied these robots to tasks such as inspection, progress monitoring, and model reconstruction using onboard sensors like Light Detection and Ranging (LiDAR) and 360° cameras [ 13 ]. Inspection is one of the key applications of quadruped robots in construction. Their ability to walk over debris, climb stairs, and move through narrow or uneven spaces enables ground-level inspection tasks in challenging environments. Equipped with visual sensors such as Red-Green-Blue (RGB), thermal, and 360° cameras, quadrupeds can autonomously collect data for code compliance and safety verification. For instance, Aziz et al. deployed a Spot robot with a PTZ and multiple fisheye cameras to perform autonomous fire safety equipment inspections in buildings [ 14 ]. Similarly, Halder et al. used quadrupeds to support human inspectors by enabling remote image capture [ 13 ]. In some cases, tactile and auditory inspection methods have been explored. Jang et al. equipped a quadruped with a robotic arm to detect pipe damage through sound [ 15 ]. Kolvenbach et al. utilized foot-mounted sensors for concrete condition assessment in sewers [ 16 ]. Progress monitoring is another key application where quadruped robots enhance construction management. By autonomously capturing site imagery, these robots can generate accurate records of component installation over time. Zhao et al. demonstrated the use of vision-equipped quadrupeds in warehouses and residential buildings to document the installation of doors, lighting fixtures, and safety systems [ 17 ]. These data streams were integrated with Building Information Modeling (BIM) and sensor inputs to produce real-time progress reports using automated detection algorithms. Halder et al. extended this approach by combining 360° imaging with augmented reality (AR), enabling remote stakeholders to assess progress against BIM models [ 18 ]. This method supports early identification of delays or discrepancies and improves coordination between off-site and on-site teams. Model reconstruction is another application of quadruped robots, which can serve as mobile platforms for collecting spatial information both indoors and outdoors [ 19 ]. In exterior environments, Kim et al. used LiDAR-equipped quadrupeds to scan scaffolding and perform semantic segmentation for automatic model generation [ 20 ]. Indoors, Hu et al. applied Simultaneous Localization and Mapping (SLAM) techniques with LiDAR to reconstruct detailed models of walls, furniture, and other elements [ 21 ]. 2.2 Robot navigation algorithms in construction environment Autonomous navigation in construction sites is highly challenging due to their cluttered layouts, evolving obstacles, and frequent human activity. To address these challenges, researchers have developed a range of navigation algorithms that can broadly be divided into geometry-based methods and integrative semantic methods. Geometry-based approaches rely on sensor data to build occupancy maps and generate collision-free paths, whereas integrative semantic approaches augment these geometric foundations with contextual information such as object categories and building models. 2.2.1 Geometry-based methods Geometry-based methods rely primarily on spatial and structural information captured by LiDAR, cameras, or depth sensors. These approaches focus on obstacle avoidance and path efficiency by constructing occupancy maps and minimizing traversal costs. For example, Kim et al. introduced GRoMI, a SLAM-driven platform that detects both static and dynamic obstacles and avoids them by dynamically updating a 3D occupancy map [ 3 ]. Similarly, Ren and Jebelli combined SLAM with reinforcement learning to handle obstacles like truss structures in attics, where avoidance strategies were learned through trial-and-error cost minimization [ 4 ]. Geometry-based methods mainly treat obstacles as geometric constraints, using rules (e.g., follow walls, bypass boundaries) or cost-based optimization (e.g., shortest safe path). While robust in static or well-mapped settings, they cannot interpret contextually meaningful but geometrically subtle cues, such as caution tape or temporary barriers. 2.2.2 Integrative semantic methods Integrative semantic methods extend geometry-based navigation by incorporating contextual meaning from perception systems or external sources like Building Information Models (BIM). These approaches consider categories of environmental factors such as people, equipment, terrain types, or restricted zones, and adjust avoidance strategies accordingly. For instance, semantic perception systems use object detection or segmentation to recognize categories like workers, vehicles, or hazardous zones, with avoidance strategies typically governed by predefined rules (e.g., stop when detecting humans, detour around machinery). Guan et al. developed the Terrain Navigability System (TNS), which fuses semantic segmentation of RGB images with geometric analysis of slope and step height to generate continuous traversability scores [ 10 ]. Terrain features such as flat ground, rock piles, water puddles, and bumpy surfaces are classified, and each grid cell is assigned a traversability value between 0 (impassable) and 1 (fully traversable). This enables cost-based avoidance, where planners favor safer, smoother terrain rather than relying solely on binary rules. BIM-based methods similarly integrate semantic knowledge into navigation. For example, Karimi et al. proposed the Building Information Robotic System (BIRS), which uses BIM to assign semantic meaning to map regions, allowing robots to select optimal paths (e.g., safer or smoother) rather than only shortest ones [ 22 ]. Beyond static semantics, Wu et al. introduced a human-centered navigation framework that integrates global (A*) and local (DWA) planners with an RL-based fine-tuning layer [ 23 ]. In their system, the RL agent takes as input the robot’s state, nearby obstacles, and worker positions, and outputs refined velocity commands. By optimizing a reward function that penalizes collisions, intrusion into workers’ comfort zones, and inefficient detours, the system enables dynamic, socially aware navigation that adapts robot behavior in the presence of humans. Integrative semantic methods enrich navigation by combining geometric mapping with contextual layers such as terrain type, functional areas, and human presence. They extend avoidance strategies beyond geometry alone, employing rules, cost-based traversability scores, or reinforcement learning policies to enable safer and more adaptive navigation in complex sites. However, these methods still face limitations: rule-based and cost-based traversability strategies depend heavily on designer assumptions and predefined heuristics for weighting hazards [ 10 ], [ 24 ]. In practice, the environmental factors considered in the rule or cost are limited and it is impractical to manually encode all the dynamic, contextual, and semantic elements present from construction environments. Moreover, these methods oversimplify the scene by grouping diverse elements into broad categories. This abstraction neglects the semantic meaning and task relevance of different objects. RL-based fine-tuning requires extensive training data and may not generalize well across diverse construction environments [ 25 ]. Moreover, it is difficult to simulate the complexity of real-world construction sites. Critically, none of these approaches are grounded in empirical evidence of the environmental cues that human operators rely on, leaving a key knowledge gap that motivates the present study. 2.3 Leveraging human-identified cues to improve navigation Studies in autonomous driving have shown that human-identified environmental cues, extracted from video, can significantly improve navigation performance. In these studies, human annotators label both the operators’ behaviors (e.g., turning, slowing, stopping) and the specific cues that triggered those behaviors (e.g., pedestrians crossing, traffic lights, congestion) [ 11 ], [ 12 ]. This annotation process links motion patterns with contextual causes, creating datasets that go beyond raw perception or vehicle dynamics. Using these labeled datasets, models can be trained not only to predict what action a vehicle should take but also to understand why the action is appropriate. Xu et al. [ 12 ] demonstrated that incorporating object-triggered action annotations improved action prediction accuracy, especially for complex maneuvers such as left or right turns. Similarly, Ramanishka et al. showed that models trained on causally annotated driver behaviors achieved higher accuracy than models using vision or dynamics alone [ 11 ]. These improvements arise precisely because navigation decisions are grounded in environmental cues that humans judged to be salient. 3 Methodology 3.1 Study design This study employed a naturalistic field design to investigate the environmental cues used by human operators when navigating quadruped robots in construction environments. Naturalistic designs are particularly appropriate for HRI research when the goal is to capture authentic behaviors that emerge in dynamic, real-world contexts, rather than in scripted or laboratory-controlled settings [ 26 ], [ 27 ]. By situating the study on a live construction site, we ensured ecological validity and captured the complexity of social and environmental interactions that autonomous systems must eventually contend with. As shown in Fig. 1 , The study site was an active construction project on the University of Wisconsin–Madison campus, which provided the evolving spatial layouts, temporary obstacles, and worker activities characteristic of real-world deployment scenarios. Conducting HRI studies in such naturalistic, high-stakes domains has been shown to yield insights that are not observable in laboratory contexts, particularly regarding the interaction between human activity and robotic behavior [ 28 ]. Importantly, worker movements were not scripted, and no prior coordination occurred with site staff, ensuring that all interactions unfolded organically during routine operations. The robotic task was designed to emulate autonomous construction progress monitoring, a common application of quadruped robots. During each session, the operator manually navigated the robot between predefined waypoints while recording egocentric RGB video of the environment. This task framing aligns with established practices in HRI and field robotics research, where task realism is emphasized to maximize the transferability of study findings to autonomous system design [ 29 ]. The egocentric video served as the primary data source for linking operator decisions with environmental cues, following methods successfully applied in autonomous driving research [ 11 ], [ 12 ]. The data collection procedure was designed to maximize diversity and capture the evolving nature of construction sites. Five sessions were conducted across both indoor and outdoor environments on May 27, June 1, June 4, June 19, and June 20, 2025. These sessions were spaced to reflect different stages of construction progress and to capture varied spatial configurations, material placements, and levels of human and machinery activity. This temporal sampling strategy aligns with HRI field study practices that emphasize longitudinal and contextually diverse data to improve generalizability of findings [ 30 ]. Each session was conducted under the supervision of a licensed site engineer, and all safety procedures adhered strictly to site protocols. The data analysis process was structured to ensure reliability and reduce bias. Thematic analysis method was used to identify the environmental cues essential for safe navigation on construction site. Two coders were first trained in the study objectives and coding protocol, after which they independently coded the video data to identify instances where operator navigation decisions were linked to environmental cues. A triangulation approach was applied, whereby results were compared and discrepancies discussed until consensus was reached. 3.2 Robotic platform The robotic platform for this study was a commercial quadruped robot (Unitree B2 [ 31 ]). The robot supports a maximum walking speed of approximately 6 m/s. Its battery life of up to 4–5 hours per charge allowed for extended recording sessions without frequent interruptions. The robot was equipped with a 360-degree panoramic camera mounted on its body. The camera provided continuous omnidirectional RGB video, capturing the surrounding environmental context. The robot was manually controlled using a handheld controller, which allowed the operator to issue commands for moving forward, backward, turning left and right. 3.3 Data collection procedure The data collection was structured to capture navigation behavior under diverse and evolving construction scenarios while maintaining ecological validity. Five field sessions were conducted at the University of Wisconsin–Madison construction site on May 27, June 1, June 4, June 19, and June 20, 2025. These sessions were strategically spaced over several weeks to reflect different stages of construction progress, enabling the study to capture variations in spatial layout, material placement, and levels of human and machinery activity. Longitudinal sampling of this kind is widely recommended in HRI field studies to improve the robustness and generalizability of findings [ 32 ]. Figure 3 shows the data collection on May 27, 2025. The robot was manually navigated between waypoints that were randomly selected but distributed across the entire site. This approach ensured that the robot’s trajectories provided broad spatial coverage while also introducing unpredictability into the navigation task. Randomized waypoint selection exposed the robot to a wide range of scenarios, including varied terrain, equipment zones, and worker activity, while guaranteeing that all major site areas were represented. During navigation, a 360-degree RGB camera mounted on the robot continuously recorded egocentric video of the environment. To capture the operator’s perspective and maintain situational awareness, the operator walked behind and followed the robot during each session. This arrangement ensured that the recorded video reflected the visual context available to the human operator when making navigation decisions. The sessions were conducted under live construction conditions without scripting or intervention. No communication occurred with site workers in advance, and their movements and activities were left entirely unscripted. This choice ensured that all human–robot encounters unfolded organically during routine operations, preserving the authenticity of interactions. Such unscripted field studies are critical in HRI because they reveal contextual patterns and environmental cues that cannot be fully replicated in laboratory-based experiments [ 33 ]. The data collection process also prioritized safety and ethical compliance. Each session was supervised by a licensed site engineer, and all activities adhered strictly to construction site safety protocols. The presence of a safety supervisor ensured that both workers and research staff were protected during robot deployment, aligning with best practices for conducting HRI research in high-risk field environments. 3.4 Data processing procedure The data processing procedure was structured to identify and classify the environmental cues that influenced human navigation decisions. The first step involved video segmentation, which was based on the completion of each navigation trial. Specifically, each segment began when the robot departed from a designated starting point (e.g., point A) and ended upon arrival at a corresponding endpoint (e.g., point B). Table 1 summarize the manual navigation video segments. Table 1 Summary of manual navigation video segments Test date Cumulative video duration No. of video segments May 27, 2025 7 min 28 sec 6 June 4, 2025 37 min 16 sec 16 June 6, 2025 4 min 8 sec 4 June 19, 2025 7 min 9 sec 4 June 20, 2025 20 min 28 sec 11 To analyze the environmental cues underpinning these navigation episodes, we employed thematic analysis, a widely used qualitative method for identifying, analyzing, and reporting patterns (themes) within data. This approach offers both flexibility and methodological rigor and is particularly suited to exploratory studies in dynamic, real-world environments such as construction sites. We followed the six-phase framework proposed by Braun and Clarke [ 34 ], using an inductive, data-driven approach in which themes were grounded in the observed video data rather than shaped by predefined categories. The six phases proceeded as follows: Familiarization with the data – Coders repeatedly reviewed the egocentric video recordings while making notes on observable operator actions and the surrounding environmental context. This process established a foundational understanding of how navigation behavior emerged in situation. Generating initial codes – Two coders independently created time-stamped annotations for segments where meaningful navigation actions occurred. Each instance was labeled with the corresponding action, classified into one of four categories: left turn, right turn, stop, or resume walking, and an open-coded description of the environmental cue that appeared to influence that action. Constructing initial themes – The coders collaboratively examined the set of environmental cue codes, grouping similar codes together based on their shared characteristics. These groups formed provisional themes representing broader categories of navigation-relevant cues. Reviewing themes – Each provisional theme was evaluated for coherence and distinctiveness. The team revisited all relevant video segments to confirm that coded instances were appropriately assigned, and to resolve any ambiguities or overlaps among themes. Themes that lacked sufficient support or clarity were modified, merged, or discarded. Defining and naming themes – Finalized themes were clearly defined and labeled to capture the essence of the underlying cue category. Definitions emphasized how each cue type contributed to human navigation behavior in the construction context. Producing the report – The finalized thematic structure served as the foundation for our taxonomy of environmental cues. This taxonomy directly informed the presentation of findings and the development of implications for future autonomous navigation systems. To enhance analytic rigor, we adopted an investigator triangulation strategy. Two coders with prior experience in construction robotics completed the annotation and theme development independently before engaging in consensus-building discussions. 4 Results Through thematic analysis of egocentric video recordings captured during manual quadruped robot navigation trials, we identified three categories of environmental cues that influenced operator decision-making: topography cues, obstacle and hazard cues, and communicative cues as shown in Fig. 4 . 4.1 Topography cues Topography cues refer to environmental features that inform whether a given path is physically navigable by the robot. These cues affect stability, traction, and gait planning, and encompass both surface conditions and terrain features. Operators relied on visual assessment of the ground’s composition and topography to preemptively adjust heading and avoid risky or difficult to traverse areas. Two subtypes of Topography cues were observed: The first one is surface texture included puddles, sand, gravel, paved areas and so on. These cues influenced decisions to avoid slippage or prevent damage to the robot’s hardware. The second is terrain features encompass the macro-structural characteristics of the ground, such as inclines, declines, steps, slopes, and irregular or rugged topography. These features impact the robot's balance and gait and may pose a risk of stumbling or instability, particularly in unstructured environments like construction sites. For example, in one trial, the robot approached a section of the site covered by standing water. Recognizing the risk of slipping and potential harm to onboard electronics, the operator steered the robot away from the puddle and selected a drier, more stable path (Fig. 5 ). In another case (Fig. 6 ), the operator identified a zone of uneven terrain ahead. Rather than proceeding directly, they modified the robot’s heading to avoid the unstable area and guide it toward a smoother section of the site. 4.2 Obstacle and hazard cues Obstacle and hazard cues include physical entities that obstruct the robot’s path or present collision and safety risks. These cues can be static or dynamic and require the operator to engage in real-time avoidance, stopping, or rerouting behaviors. This category consists of: Static objects, such as tools, and construction materials that remain stationary during the navigation sequence. Dynamic objects, including moving vehicles, machinery, suspended loads, and human workers. In a representative case, the robot was headed toward a group of pipe covers placed flush with the ground. Although not elevated, the operator identified these as sensitive equipment and took action to steer the robot around them to avoid potential damage (Fig. 7 ). During one session, the robot’s path was blocked by two simultaneously moving hazards: suspended construction material being lifted overhead and a concrete mixer truck crossing laterally. The operator promptly stopped the robot and waited until both hazards had cleared the area before resuming movement (Fig. 8 ). 4.3 Communicative cues Communicative cues are explicit or implicit signals used to coordinate navigation. These cues serve to convey intent, structure turn-taking, and guide socially appropriate behavior. They may be generated by humans or embedded in the environment as conventional indicators. This category includes human-generated cues, such as hand gestures, body posture, and verbal instructions. Environmentally embedded cues, such as warning signs, cones, barricade tape, and flashing lights. In one example, the robot approached a path intersected by a telescopic forklift. The operator halted the robot to avoid conflict. After a short delay, the forklift driver signaled with a hand gesture that it was safe to proceed. The robot operator then resumed navigation (Fig. 9 ). In another example, after observing a red waning line, the operator controlled the robot to turn right to avoid the area surrounded by the waning line (Fig. 10 ). 5 Discussion This study provides an empirically grounded understanding of the environmental cues that are essential for safe robot navigation in active construction environments. Through naturalistic field observations and thematic analysis of egocentric video data, we developed a taxonomy of three cue categories: topography cues, which pertain to ground surface texture and terrain geometry; obstacle and hazard cues, which involve both static and dynamic objects; and communicative cues, which include human gestures and embedded visual indicators. Together, these findings address a major gap in construction robotics research: the lack of empirical evidence identifying which environmental features are meaningful to human operators during navigation. By documenting these human-identified cues, this study contributes to a richer understanding of how contextual awareness can enhance autonomous navigation in complex, dynamic, and socially populated environments. 5.1 Implications for navigation system design The taxonomy derived from this study offers practical guidance for developing context-aware perception and navigation systems. First, the identification of topography cues emphasizes the importance of fine-grained surface assessment beyond simple geometric mapping. While traditional SLAM-based systems can detect obstacles, they often fail to distinguish between traversable and unstable terrain. Our findings suggest that surface texture and terrain geometry, such as puddles, loose gravel, or inclines, are salient features that humans continuously evaluate to ensure stability and safety. Incorporating these topographic considerations into perception modules may enhance a robot’s ability to assess risk and adapt its gait and trajectory accordingly. Second, obstacle and hazard cues underscore the need for temporal reasoning in navigation algorithms. Human operators exhibited not only obstacle avoidance but also anticipatory behaviors, such as pausing when observing moving machinery or suspended loads. This indicates that effective autonomous navigation in construction contexts requires systems capable of predicting object motion and integrating short-term forecasting into their path planning. Third, communicative cues highlight the role of social intelligence in robot navigation. Operators relied on both explicit gestures from workers and implicit environmental signals, such as barricade tape or warning signs, to guide their actions. Current navigation systems largely ignore these social and semiotic dimensions, yet they are essential for safe and cooperative operation in shared workspaces. Future design should therefore incorporate multimodal perception that can interpret human gestures, body posture, and visual indicators, aligning robotic behavior with human expectations and safety norms. 5.2 Limitations and future work Although this study provides novel insights, several limitations must be acknowledged. The data were collected from a single construction site and one operator, which constrains the diversity of environmental and behavioral conditions captured. Future research should replicate this study across multiple sites and operators to examine the consistency and variability of cue use. Additionally, our analysis focused on visual data; integrating other sensory modalities such as audio or tactile feedback may yield a more complete picture of how operators assess environmental risk and intent. Finally, while this study identifies which cues humans rely on, future work should translate these findings into algorithmic implementations. 5.3 Broader impacts and generalizability Beyond construction robotics, the broader significance of this work lies in its methodological and conceptual contributions to HRI. Methodologically, it demonstrates the value of naturalistic field studies in capturing authentic interactions between humans, robots, and their environments—interactions that are often obscured in simulation-based or lab-controlled research. Conceptually, it highlights the importance of environmental cues as an organizing principle for designing socially and contextually intelligent robotic behavior. As human-robot collaboration increasingly moves into complex real-world domains, systems that understand and act on nuanced environmental cues will be essential for safe, effective, and socially acceptable autonomy. 6 Conclusion Autonomous navigation in construction environments continues to present significant challenges due to their complex spatial layouts, changing obstacles, and the need for interaction with human workers. Existing approaches often rely on manually programmed rules and costs to interpret environmental cues (e.g., maintaining a minimum distance of 0.3 meters from obstacles, assigning higher costs to uneven terrain or proximity to obstacles), but lack empirical insight into which specific cues are critical for safe navigation. To address this gap, we conducted a naturalistic field study informed by methods used in autonomous driving research. We deployed a manually controlled quadruped on an active site and collected egocentric video using naturalistic field methods. After that, we analyzed the videos using thematic analysis to identify the environmental features that influenced human navigation decisions. Our findings resulted in a taxonomy of three categories of environmental cues essential for safe navigation on construction site. Topography cues include features of the terrain and ground conditions that affect robot stability and gait. Obstacle and hazard cues consist of both stationary and moving elements that may require avoidance or path redirection. Communicative cues include signals from human workers or environmental markers, such as gestures, signs, or caution tape. These findings provide an empirically grounded understanding of the environmental cues that are essential for safe robot navigation in active construction environments. However, this study has several limitations. The dataset derives from one site and a single operator, which supports analytic (mechanism-level) generalization rather than prevalence across projects, robots, or teams. Moreover, our focus on visual evidence omits potentially relevant auditory and haptic cues. Declarations ACKNOWLEDGMENTS This paper is based in part upon work supported by the Wisconsin Alumni Research Foundation (WARF) under Project No. AAM3225. We also gratefully acknowledge JP Cullen for facilitating site access and providing logistical support during on-site data collection and analysis. Any opinions, findings, and conclusions or recommendations expressed in this paper are those of the author(s) and do not necessarily reflect the views of WARF or JP Cullen. DATA AVAILABILITY STATEMENT Available with restricted access. The field data that support the findings of this study (including egocentric 360° video recordings and associated robot log data collected on an active construction site) are subject to confidentiality agreements with the industry partner and contain information that could compromise site privacy. For these reasons, the data are not publicly available. An anonymized subset of the data may, however, be made available from the corresponding author upon reasonable request and with the permission of the collaborating construction company and the authors’ institution. Ethics approval This study involved non-interventional observational data collection on an active construction site. Video data were collected with the consent of JP Cullen, and a site engineer supervised the data collection process to prevent safety incidents. All video used for analysis was de-identified prior to analysis by blurring workers’ faces and masking identifying marks and logos, and no demographic or personally identifiable information about workers was collected. Conflict of interest The authors have no competing interests to declare that are relevant to the content of this article. References H. Yue, Y. Sun, N. Zeng, S. Chen, Y. Tan, and Q. Wang, “Legged Robots for Construction Management: Applications and Challenges,” Journal of Construction Engineering and Management , vol. 151, no. 8, p. 03125006, Aug. 2025, doi: 10.1061/JCEMD4.COENG-16185. P. Biswal and P. K. Mohanty, “Development of quadruped walking robots: A review,” Ain Shams Engineering Journal , vol. 12, no. 2, pp. 2017–2031, June 2021, doi: 10.1016/j.asej.2020.11.005. P. Kim, J. Chen, J. Kim, and Y. K. Cho, “SLAM-Driven Intelligent Autonomous Mobile Robot Navigation for Construction Applications,” in Advanced Computing Strategies for Engineering , I. F. C. Smith and B. Domer, Eds., Cham: Springer International Publishing, 2018, pp. 254–269. doi: 10.1007/978-3-319-91635-4_14. T. Ren and H. Jebelli, “Efficient 3D robotic mapping and navigation method in complex construction environments,” Computer-Aided Civil and Infrastructure Engineering , vol. 40, no. 12, pp. 1580–1605, 2025, doi: 10.1111/mice.13353. M. Wigness, J. G. Rogers, C.-E. Tsai, C. Mertz, L. Navarro-Serment, and J. Oh, “Using Perception Cues for Context-Aware Navigation in Dynamic Outdoor Environments,” Field Robotics , vol. 1, pp. 1–33, Oct. 2021, doi: 10.55417/fr.2021001. Z. Alenzi, E. Alenzi, M. Alqasir, M. Alruwaili, T. Alhmiedat, and O. M. Alia, “A Semantic Classification Approach for Indoor Robot Navigation,” Electronics , vol. 11, no. 13, p. 2063, Jan. 2022, doi: 10.3390/electronics11132063. J. Crespo, J. C. Castillo, O. M. Mozos, and R. Barber, “Semantic Information for Robot Navigation: A Survey,” Applied Sciences , vol. 10, no. 2, p. 497, Jan. 2020, doi: 10.3390/app10020497. Z. Jiang, “A Review of Recent Developments in Quadruped Robots,” Theoretical and Natural Science , vol. 79, pp. 28–32, Jan. 2025, doi: 10.54254/2753-8818/2025.19933. Y. Xu, B. Li, S. Yang, Y. Zhi, R. Li, and T. Chen, “MSOT-Q: Multi-Sensor Fusion SLAM with Object Detection Tracking for Quadruped Robots,” in 2024 36th Chinese Control and Decision Conference (CCDC) , May 2024, pp. 4485–4492. doi: 10.1109/CCDC62350.2024.10587825. T. Guan, Z. He, R. Song, D. Manocha, and L. Zhang, “TNS: Terrain Traversability Mapping and Navigation System for Autonomous Excavators,” May 05, 2022, arXiv : arXiv:2109.06250. doi: 10.48550/arXiv.2109.06250. V. Ramanishka, Y.-T. Chen, T. Misu, and K. Saenko, “Toward Driving Scene Understanding: A Dataset for Learning Driver Behavior and Causal Reasoning,” presented at the Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018, pp. 7699–7707. Accessed: Sept. 05, 2025. [Online]. Available: https://openaccess.thecvf.com/content_cvpr_2018/html/Ramanishka_Toward_Driving_Scene_CVPR_2018_paper.html Y. Xu et al. , “Explainable Object-Induced Action Decision for Autonomous Vehicles,” presented at the Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020, pp. 9523–9532. Accessed: Sept. 05, 2025. [Online]. Available: https://openaccess.thecvf.com/content_CVPR_2020/html/Xu_Explainable_Object-Induced_Action_Decision_for_Autonomous_Vehicles_CVPR_2020_paper.html S. Halder, K. Afsari, E. K. Chiou, R. N. C. Patrick, and K. Hamed, “Construction inspection & monitoring with quadruped robots in future human-robot teaming: A preliminary study,” Journal of Building Engineering , 2023, doi: 10.1016/j.jobe.2022.105814. A. Aziz, P. Herbers, H. Bayer, and M. König, “Fully Autonomous Fire Safety Equipment Inspection Missions on a Legged Robot,” pp. 804–812, Jan. 2024, doi: 10.1061/9780784485224.097. Y. Jang, W. Seol, K. Lee, K.-S. Kim, and S. Kim, “Development of quadruped robot for inspection of underground pipelines in nuclear power plants,” Electronics Letters , vol. 58, no. 6, pp. 234–236, 2022, doi: 10.1049/ell2.12414. H. Kolvenbach et al. , “Towards autonomous inspection of concrete deterioration in sewers with legged robots,” Journal of Field Robotics , vol. 37, no. 8, pp. 1314–1327, 2020, doi: 10.1002/rob.21964. X. Zhao, Y. Jin, N. M. Selvaraj, M. Ilyas, and C. C. Cheah, “Platform-independent visual installation progress monitoring for construction automation,” Automation in Construction , vol. 154, p. 104996, Oct. 2023, doi: 10.1016/j.autcon.2023.104996. S. Halder, K. Afsari, J. Serdakowski, S. DeVito, M. Ensafi, and W. Thabet, “Real-Time and Remote Construction Progress Monitoring with a Quadruped Robot Using Augmented Reality,” Buildings , vol. 12, no. 11, p. 2027, Nov. 2022, doi: 10.3390/buildings12112027. Y. Tao, M. Popović, Y. Wang, S. T. Digumarti, N. Chebrolu, and M. Fallon, “3D Lidar Reconstruction with Probabilistic Depth Completion for Robotic Navigation,” in 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , Oct. 2022, pp. 5339–5346. doi: 10.1109/IROS47612.2022.9981531. D. Jiang et al. , “An Integrated Autonomous Dynamic Navigation Approach toward a Composite Air–Ground Risk Construction Scenario,” Sensors , vol. 24, no. 1, Art. no. 1, Jan. 2024, doi: 10.3390/s24010221. D. Hu, V. J. L. Gan, and C. Yin, “Robot-assisted mobile scanning for automated 3D reconstruction and point cloud semantic segmentation of building interiors,” Automation in Construction , vol. 152, p. 104949, Aug. 2023, doi: 10.1016/j.autcon.2023.104949. S. Karimi, R. G. Braga, I. Iordanova, and D. St-Onge, “Semantic Optimal Robot Navigation Using Building Information on Construction Sites,” ISARC Proceedings , vol. 2021 Proceedings of the 38th ISARC, Dubai, UAE, pp. 57–64, Nov. 2021, doi: 10.22260/ISARC2021/0010. Y. Wu, J. Wei, J. Oh, and D. C. Llach, “Towards Human-Centered Construction Robotics: A Reinforcement Learning-Driven Companion Robot for Contextually Assisting Carpentry Workers,” in 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , Oct. 2024, pp. 90–97. doi: 10.1109/IROS58592.2024.10802698. R. G. Braga, M. O. Tahir, S. Karimi, U. Dah-Achinanon, I. Iordanova, and D. St-Onge, “Intuitive BIM-aided robotic navigation and assets localization with semantic user interfaces,” Front Robot AI , vol. 12, p. 1548684, Mar. 2025, doi: 10.3389/frobt.2025.1548684. Y. Wu, E. Erdolu, J. Wei, J. Oh, and D. Cardoso Llach, “Robot in the loop: a human-centered approach to contextualizing AI and robotics in construction,” Constr Robot , vol. 9, no. 1, p. 1, Dec. 2024, doi: 10.1007/s41693-024-00144-8. E. Eriksson, T. R. Hansen, and A. Lykke-Olesen, “Reclaiming public space: designing for public interaction with private devices,” in Proceedings of the 1st international conference on Tangible and embedded interaction , in TEI ’07. New York, NY, USA: Association for Computing Machinery, Feb. 2007, pp. 31–38. doi: 10.1145/1226969.1226976. M. Stenmark, M. Haage, and E. A. Topp, “Simplified Programming of Re-usable Skills on a Safe Industrial Robot: Prototype and Evaluation,” in Proceedings of the 2017 ACM/IEEE International Conference on Human-Robot Interaction , in HRI ’17. New York, NY, USA: Association for Computing Machinery, Mar. 2017, pp. 463–472. doi: 10.1145/2909824.3020227. A. Sauppé and B. Mutlu, “The Social Impact of a Robot Co-Worker in Industrial Settings,” in Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems , in CHI ’15. New York, NY, USA: Association for Computing Machinery, Apr. 2015, pp. 3613–3622. doi: 10.1145/2702123.2702181. N. Hudson, C. Alcock, and P. K. Chilana, “Understanding Newcomers to 3D Printing: Motivations, Workflows, and Barriers of Casual Makers,” in Proceedings of the 2016 CHI Conference on Human Factors in Computing Systems , in CHI ’16. New York, NY, USA: Association for Computing Machinery, May 2016, pp. 384–396. doi: 10.1145/2858036.2858266. O. Alvarado and A. Waern, “Towards Algorithmic Experience: Initial Efforts for Social Media Contexts,” in Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems , in CHI ’18. New York, NY, USA: Association for Computing Machinery, Apr. 2018, pp. 1–12. doi: 10.1145/3173574.3173860. “Robot Dog B2_Inspection Robot_B2 Robot Dog Advantage Usage | Unitree Robotics.” Accessed: Sept. 26, 2025. [Online]. Available: https://www.unitree.com/b2 O. Alvarado and A. Waern, “Towards Algorithmic Experience: Initial Efforts for Social Media Contexts,” in Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems , in CHI ’18. New York, NY, USA: Association for Computing Machinery, Apr. 2018, pp. 1–12. doi: 10.1145/3173574.3173860. A. Sauppé and B. Mutlu, “The Social Impact of a Robot Co-Worker in Industrial Settings,” in Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems , in CHI ’15. New York, NY, USA: Association for Computing Machinery, Apr. 2015, pp. 3613–3622. doi: 10.1145/2702123.2702181. V. Braun and V. Clarke, “Using thematic analysis in psychology,” Qualitative Research in Psychology , vol. 3, no. 2, pp. 77–101, Jan. 2006, doi: 10.1191/1478088706qp063oa. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8254973","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":573611586,"identity":"ad2c9795-6868-4ebb-bf52-3c871ba8cac3","order_by":0,"name":"Liqun Xu","email":"","orcid":"","institution":"University of Wisconsin–Madison","correspondingAuthor":false,"prefix":"","firstName":"Liqun","middleName":"","lastName":"Xu","suffix":""},{"id":573611587,"identity":"6e4a48af-472c-4a8e-8419-b68f8123a5cf","order_by":1,"name":"Geonho Kim","email":"","orcid":"","institution":"University of Wisconsin–Madison","correspondingAuthor":false,"prefix":"","firstName":"Geonho","middleName":"","lastName":"Kim","suffix":""},{"id":573611588,"identity":"309546e0-3875-4fce-b1a0-0ada84594e4d","order_by":2,"name":"Zhenhua Zhu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAx0lEQVRIiWNgGAWjYFACHgaGDxU2cgzMYB4zcVoYZ5xJMyZNCzNv26HEBgZitcj3nz0mwXPmQPqG4+wPPzBUWMP04gaMDefSJCQq7uRuOMxjLMFwJp2wFmbGHjMJgzPPcrcd5mFjYGw7TFgLGzOPmURi2+F0s8PszxgY/xGhhYcNqOVg2+EEs8MMZkBnEqFFgofH2LLhTJrhfpBfEo6lGxPUIt9/xvD2nwobecn+4w8/fKixliWoBRUkkKZ8FIyCUTAKRgEuAACESDvHgmlnXAAAAABJRU5ErkJggg==","orcid":"","institution":"University of Wisconsin–Madison","correspondingAuthor":true,"prefix":"","firstName":"Zhenhua","middleName":"","lastName":"Zhu","suffix":""}],"badges":[],"createdAt":"2025-12-02 02:23:14","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8254973/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8254973/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":100228577,"identity":"c114db9c-975a-4cc9-a0f5-5534977f28db","added_by":"auto","created_at":"2026-01-14 10:59:37","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":3518582,"visible":true,"origin":"","legend":"","description":"","filename":"Anoymisedmanuscript.docx","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/de0622dcebbb533701def362.docx"},{"id":100228561,"identity":"24a0aa21-c149-42f0-9887-9d80a1762f37","added_by":"auto","created_at":"2026-01-14 10:59:37","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":6559,"visible":true,"origin":"","legend":"","description":"","filename":"818b15a9f65544a89f9e142c1f1c1e58.json","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/6495b58c61ec6996ff6a39ad.json"},{"id":100228570,"identity":"f9101807-3ad7-4a73-b9dd-511e8a2f5754","added_by":"auto","created_at":"2026-01-14 10:59:37","extension":"xml","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":99847,"visible":true,"origin":"","legend":"","description":"","filename":"818b15a9f65544a89f9e142c1f1c1e581enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/8729b4963b97a288e074c055.xml"},{"id":100370658,"identity":"b7b06886-001a-4819-9566-338f35fcf267","added_by":"auto","created_at":"2026-01-16 08:07:13","extension":"png","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1180198,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/203435d1e6742ec21802243c.png"},{"id":100228563,"identity":"4360aa63-4f73-434f-b8cf-07f09e155dc6","added_by":"auto","created_at":"2026-01-14 10:59:37","extension":"jpeg","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":483248,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage10.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/1a54a95bcfc96babe039d74b.jpeg"},{"id":100371014,"identity":"ad8327b8-27e1-433b-b225-0a541cb63853","added_by":"auto","created_at":"2026-01-16 08:09:12","extension":"png","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":242728,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/d9647ad50444e23521ef46fc.png"},{"id":100370196,"identity":"cf30202a-0d85-4a06-8a2b-dbf4ae0b90d3","added_by":"auto","created_at":"2026-01-16 08:00:26","extension":"jpeg","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":225131,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/a4485b53504fb52c9e719b64.jpeg"},{"id":100370588,"identity":"05bd3f52-8685-4bb9-87ff-c253497024e5","added_by":"auto","created_at":"2026-01-16 08:06:39","extension":"png","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":673382,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/ff62fad02f6d26db7d6d14e9.png"},{"id":100371162,"identity":"b22041f1-6e2a-4724-8fee-fba058997fd2","added_by":"auto","created_at":"2026-01-16 08:09:34","extension":"jpeg","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":474023,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage5.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/d0439b8d51d2c76fe4c3aec8.jpeg"},{"id":100370888,"identity":"85218361-6fa8-4a0e-8cf3-2b01749cbaf3","added_by":"auto","created_at":"2026-01-16 08:08:55","extension":"jpeg","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":507352,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage6.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/fb3b942a755f507f7f1d6dd7.jpeg"},{"id":100228579,"identity":"a7a6ea1f-008d-4242-841c-92589a845c3f","added_by":"auto","created_at":"2026-01-14 10:59:37","extension":"jpeg","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":374604,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage7.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/a6915ee07da12def3006466d.jpeg"},{"id":100370641,"identity":"52315153-6a1d-4ddf-893e-31feee0e039b","added_by":"auto","created_at":"2026-01-16 08:07:06","extension":"jpeg","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":426476,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage8.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/40ca0b7d894eee2fe3e24458.jpeg"},{"id":100370883,"identity":"9cc297eb-f17d-40d8-a3e1-3ead990f6ff0","added_by":"auto","created_at":"2026-01-16 08:08:52","extension":"jpeg","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":444619,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage9.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/f1bec9ebc652f9848ab531b5.jpeg"},{"id":100228589,"identity":"caa8556f-53d7-471a-8f08-7ae554946fce","added_by":"auto","created_at":"2026-01-14 10:59:38","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":171141,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/0119f31fd03fddb95ea949fd.png"},{"id":100228571,"identity":"bc4e4e5d-3b66-44f8-9c72-0f2855c9e125","added_by":"auto","created_at":"2026-01-14 10:59:37","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":196891,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage10.png","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/a4daf49351d88a2d3b71f333.png"},{"id":100228590,"identity":"45e366dd-efec-4c68-86a1-a0abc5f78512","added_by":"auto","created_at":"2026-01-14 10:59:38","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":36766,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/a11d76c948dd51f9025ac75a.png"},{"id":100228583,"identity":"f5d6f01a-477c-4655-9e3f-b4f0173113f9","added_by":"auto","created_at":"2026-01-14 10:59:38","extension":"png","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":140697,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/e151f1da66749124e5690eb9.png"},{"id":100228587,"identity":"248f4b82-8e2a-4b3a-9fee-c82935804e68","added_by":"auto","created_at":"2026-01-14 10:59:38","extension":"png","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":102749,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/5a84db01f4845fff90901d69.png"},{"id":100228584,"identity":"8615791e-c47f-4e4e-8886-77dc5e19fc1c","added_by":"auto","created_at":"2026-01-14 10:59:38","extension":"png","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":183303,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/2455431f776aa93f700e1beb.png"},{"id":100228580,"identity":"ace5798d-9819-4fe6-ba68-43fe5172925f","added_by":"auto","created_at":"2026-01-14 10:59:38","extension":"png","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":209723,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/cef4e3b956d38aee0100b93f.png"},{"id":100370921,"identity":"eabca023-4b1f-4163-902c-a0b0862c1a76","added_by":"auto","created_at":"2026-01-16 08:09:00","extension":"png","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":131101,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/82018097b49ea4dd8c02caad.png"},{"id":100370243,"identity":"55eba16e-d103-4c58-a82c-192378ffd242","added_by":"auto","created_at":"2026-01-16 08:01:19","extension":"png","order_by":21,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":143496,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/263001615546b526ba5e75ed.png"},{"id":100228586,"identity":"bebcdf6c-3ca8-4441-a7c9-49b1039a7f93","added_by":"auto","created_at":"2026-01-14 10:59:38","extension":"png","order_by":22,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":148251,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage9.png","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/9d55f8494ee031290f7eeaf8.png"},{"id":100370947,"identity":"a884189c-076f-4da7-8714-d5cedef5bd9a","added_by":"auto","created_at":"2026-01-16 08:09:02","extension":"xml","order_by":23,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":98028,"visible":true,"origin":"","legend":"","description":"","filename":"818b15a9f65544a89f9e142c1f1c1e581structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/7e3c91d9310cef426578943c.xml"},{"id":100228591,"identity":"68dcb502-cd20-4ffc-a54b-039f668d33f7","added_by":"auto","created_at":"2026-01-14 10:59:38","extension":"html","order_by":24,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":109619,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/6121f041deb1e523116105a0.html"},{"id":100228559,"identity":"313048bb-ec5d-4ed0-9183-0d2626d50e06","added_by":"auto","created_at":"2026-01-14 10:59:37","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":1027627,"visible":true,"origin":"","legend":"\u003cp\u003eLayout of the construction site, photographed on May 21, 2025.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/f93b90a5eb49c5f6892f4476.png"},{"id":100371094,"identity":"10b4adf9-41ad-4792-92ba-98fa5d145990","added_by":"auto","created_at":"2026-01-16 08:09:25","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":116162,"visible":true,"origin":"","legend":"\u003cp\u003eRobot setup\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/b4ad8344a9d7d09e2e1468d8.png"},{"id":100371831,"identity":"1d06af0f-fa5e-45e0-a9f4-cbefbc5ea8a6","added_by":"auto","created_at":"2026-01-16 08:11:02","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":448797,"visible":true,"origin":"","legend":"\u003cp\u003eField data collection on May 27, 2025\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/7ced80ea58662a2f9bd11ba1.png"},{"id":100370920,"identity":"eaf8fbba-680a-4957-88c3-db8239baabb5","added_by":"auto","created_at":"2026-01-16 08:09:00","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":375337,"visible":true,"origin":"","legend":"\u003cp\u003eTaxonomy of environmental cues essential for safe navigation\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/147f5217c7d6d503d84ef5dc.png"},{"id":100370800,"identity":"be23c787-37f2-4d92-b66b-c0682e08c375","added_by":"auto","created_at":"2026-01-16 08:08:20","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":363283,"visible":true,"origin":"","legend":"\u003cp\u003eRobot turns right to avoid puddle\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/6d3c14dbe4c546a31d8f6391.png"},{"id":100228565,"identity":"a4150613-5b7b-428c-be57-de4eb6718f03","added_by":"auto","created_at":"2026-01-14 10:59:37","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":387370,"visible":true,"origin":"","legend":"\u003cp\u003eRobot turns left to avoid rough terrain\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/f1926822c826ce8168ffb561.png"},{"id":100371853,"identity":"38db345e-383d-4272-9e3b-a08fd8ac7df1","added_by":"auto","created_at":"2026-01-16 08:11:07","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":306758,"visible":true,"origin":"","legend":"\u003cp\u003eRobot turns left to avoid pipe cover\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/982ac6b1b8e368ee2a20e310.png"},{"id":100228567,"identity":"790e4e01-37b0-4c47-bf35-da38fd375524","added_by":"auto","created_at":"2026-01-14 10:59:37","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":324209,"visible":true,"origin":"","legend":"\u003cp\u003eRobot stopsto yield to an overhead payload and concrete mixer truck\u003c/p\u003e","description":"","filename":"8.png","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/18d2b051650c58aabb93ccdb.png"},{"id":100228582,"identity":"afe05923-9a94-4d08-8c15-d66aff3bee6f","added_by":"auto","created_at":"2026-01-14 10:59:38","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":332295,"visible":true,"origin":"","legend":"\u003cp\u003eRobot moves forward after receiving a hand gesture from the telescopic forklift operator\u003c/p\u003e","description":"","filename":"9.png","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/e1130912bd8745a7c0970b3f.png"},{"id":100228578,"identity":"b2e50747-eed0-444b-bdc0-856824b5342f","added_by":"auto","created_at":"2026-01-14 10:59:37","extension":"png","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":376527,"visible":true,"origin":"","legend":"\u003cp\u003eRobot turns right to avoid crossing warning line\u003c/p\u003e","description":"","filename":"10.png","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/d5c782c734227205cf4b5c7c.png"},{"id":104807328,"identity":"86329d45-223b-4bd0-85c8-a193493b5f04","added_by":"auto","created_at":"2026-03-17 11:57:10","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":5488042,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8254973/v1/0836b43d-baf1-4f9e-8db3-872b97d67103.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"A Taxonomy of Environmental Cues for Quadruped Robot Navigation in Construction Environments Informed by Human Operator Behavior","fulltext":[{"header":"1 Introduction","content":"\u003cp\u003eThe use of mobile robots in construction management is a rapidly growing area of interest [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Quadruped robots, in particular, offer significant mobility advantages over wheeled or tracked platforms in the unstructured and complex environments typical of construction site [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. These robots have the potential to automate tasks such as site monitoring, surveying, and logistics [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. However, realizing this potential depends critically on enabling safe autonomous navigation in construction environments that are not only physically complex but also socially interactive and in constant flux [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eAutonomous navigation in construction sites presents unique challenges. These environments are characterized by evolving layouts, temporary and often ambiguous obstacles, and frequent interactions with human workers. Most current navigation systems rely on geometry-based methods, such as LiDAR-based SLAM, to create occupancy maps and plan collision-free paths [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e], [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. While effective in controlled settings, these systems often fail in construction environments. This is because elements like caution tape and temporary barriers may appear benign to geometry-based navigation systems but hold significant operational meaning for safe navigation [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. To mitigate these limitations, recent efforts have focused on augmenting geometry-based methods with semantic perception, using deep learning models for object detection and segmentation [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e], [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. These hybrid systems have improved the robot\u0026rsquo;s ability to recognize contextual categories such as equipment, terrain, and people, enabling more informed navigation strategies [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e], [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eDespite these advances, the decision-making logic in existing systems still relies on pre-programmed rules to interpret environmental cues (e.g., avoiding entry into the falling zone of materials lifted by a tower crane) [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e], [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. Yet, there is no empirical evidence identifying which cues are essential for safe navigation in construction environments. As a result, navigation systems that appear effective in one scenario often fail to generalize across sites with different layouts, activities, or social dynamics. This lack of an empirically grounded understanding of environmental cues represents a fundamental barrier to the development of robust and transferable navigation frameworks for construction robotics.\u003c/p\u003e \u003cp\u003eResearch in autonomous driving provides a practical approach to addressing this challenge. In that field, egocentric video analysis has been used to link driver behaviors, such as slowing down, turning, or stopping, to specific environmental cues including pedestrians, signage, and road conditions [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e], [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. By capturing and annotating the relationship between human decision-making and environmental cues, these studies have identified the environmental cues critical for safe autonomous navigation. Building on this insight, we argue that a similar approach can be applied to construction robotics. By analyzing the navigation operations of human operators during manual control and linking their operations to environmental cues, it becomes possible to uncover the cues that autonomous navigation systems must recognize to perform reliably in construction settings.\u003c/p\u003e \u003cp\u003eIn this study, we deployed a manually controlled quadruped on an active site and collected egocentric video using naturalistic field methods. From these recordings, we traced operator actions to the visual and contextual information that occasioned them. Through thematic analysis of the visual and contextual information, we distill these observations into a taxonomy of environmental cues that humans use for navigation. We identify three categories of cues: topography cues, which pertain to ground surface texture and terrain geometry; obstacle and hazard cues, which involve physical impediments or safety risks; and communicative cues, which include human gestures and embedded visual indicators. The findings provide empirical evidence for which cues are essential to safe navigation on construction sites and, in turn, inform the design of perception modules for autonomous navigation in these environments.\u003c/p\u003e"},{"header":"2 Related works","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Quadruped robots in construction management\u003c/h2\u003e \u003cp\u003eQuadruped robots are increasingly being adopted in construction for their unique mobility and ability to navigate complex, unstructured environments [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Unlike wheeled robots, quadrupeds can traverse stairs, debris, and uneven terrain, making them well-suited for use in construction sites [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Researchers have applied these robots to tasks such as inspection, progress monitoring, and model reconstruction using onboard sensors like Light Detection and Ranging (LiDAR) and 360\u0026deg; cameras [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eInspection is one of the key applications of quadruped robots in construction. Their ability to walk over debris, climb stairs, and move through narrow or uneven spaces enables ground-level inspection tasks in challenging environments. Equipped with visual sensors such as Red-Green-Blue (RGB), thermal, and 360\u0026deg; cameras, quadrupeds can autonomously collect data for code compliance and safety verification. For instance, Aziz et al. deployed a Spot robot with a PTZ and multiple fisheye cameras to perform autonomous fire safety equipment inspections in buildings [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. Similarly, Halder et al. used quadrupeds to support human inspectors by enabling remote image capture [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. In some cases, tactile and auditory inspection methods have been explored. Jang et al. equipped a quadruped with a robotic arm to detect pipe damage through sound [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. Kolvenbach et al. utilized foot-mounted sensors for concrete condition assessment in sewers [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eProgress monitoring is another key application where quadruped robots enhance construction management. By autonomously capturing site imagery, these robots can generate accurate records of component installation over time. Zhao et al. demonstrated the use of vision-equipped quadrupeds in warehouses and residential buildings to document the installation of doors, lighting fixtures, and safety systems [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. These data streams were integrated with Building Information Modeling (BIM) and sensor inputs to produce real-time progress reports using automated detection algorithms. Halder et al. extended this approach by combining 360\u0026deg; imaging with augmented reality (AR), enabling remote stakeholders to assess progress against BIM models [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. This method supports early identification of delays or discrepancies and improves coordination between off-site and on-site teams.\u003c/p\u003e \u003cp\u003eModel reconstruction is another application of quadruped robots, which can serve as mobile platforms for collecting spatial information both indoors and outdoors [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. In exterior environments, Kim et al. used LiDAR-equipped quadrupeds to scan scaffolding and perform semantic segmentation for automatic model generation [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. Indoors, Hu et al. applied Simultaneous Localization and Mapping (SLAM) techniques with LiDAR to reconstruct detailed models of walls, furniture, and other elements [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Robot navigation algorithms in construction environment\u003c/h2\u003e \u003cp\u003eAutonomous navigation in construction sites is highly challenging due to their cluttered layouts, evolving obstacles, and frequent human activity. To address these challenges, researchers have developed a range of navigation algorithms that can broadly be divided into geometry-based methods and integrative semantic methods. Geometry-based approaches rely on sensor data to build occupancy maps and generate collision-free paths, whereas integrative semantic approaches augment these geometric foundations with contextual information such as object categories and building models.\u003c/p\u003e \u003cdiv id=\"Sec5\" class=\"Section3\"\u003e \u003ch2\u003e2.2.1 Geometry-based methods\u003c/h2\u003e \u003cp\u003eGeometry-based methods rely primarily on spatial and structural information captured by LiDAR, cameras, or depth sensors. These approaches focus on obstacle avoidance and path efficiency by constructing occupancy maps and minimizing traversal costs. For example, Kim et al. introduced GRoMI, a SLAM-driven platform that detects both static and dynamic obstacles and avoids them by dynamically updating a 3D occupancy map [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Similarly, Ren and Jebelli combined SLAM with reinforcement learning to handle obstacles like truss structures in attics, where avoidance strategies were learned through trial-and-error cost minimization [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eGeometry-based methods mainly treat obstacles as geometric constraints, using rules (e.g., follow walls, bypass boundaries) or cost-based optimization (e.g., shortest safe path). While robust in static or well-mapped settings, they cannot interpret contextually meaningful but geometrically subtle cues, such as caution tape or temporary barriers.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section3\"\u003e \u003ch2\u003e2.2.2 Integrative semantic methods\u003c/h2\u003e \u003cp\u003eIntegrative semantic methods extend geometry-based navigation by incorporating contextual meaning from perception systems or external sources like Building Information Models (BIM). These approaches consider categories of environmental factors such as people, equipment, terrain types, or restricted zones, and adjust avoidance strategies accordingly. For instance, semantic perception systems use object detection or segmentation to recognize categories like workers, vehicles, or hazardous zones, with avoidance strategies typically governed by predefined rules (e.g., stop when detecting humans, detour around machinery). Guan et al. developed the Terrain Navigability System (TNS), which fuses semantic segmentation of RGB images with geometric analysis of slope and step height to generate continuous traversability scores [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. Terrain features such as flat ground, rock piles, water puddles, and bumpy surfaces are classified, and each grid cell is assigned a traversability value between 0 (impassable) and 1 (fully traversable). This enables cost-based avoidance, where planners favor safer, smoother terrain rather than relying solely on binary rules. BIM-based methods similarly integrate semantic knowledge into navigation. For example, Karimi et al. proposed the Building Information Robotic System (BIRS), which uses BIM to assign semantic meaning to map regions, allowing robots to select optimal paths (e.g., safer or smoother) rather than only shortest ones [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Beyond static semantics, Wu et al. introduced a human-centered navigation framework that integrates global (A*) and local (DWA) planners with an RL-based fine-tuning layer [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. In their system, the RL agent takes as input the robot\u0026rsquo;s state, nearby obstacles, and worker positions, and outputs refined velocity commands. By optimizing a reward function that penalizes collisions, intrusion into workers\u0026rsquo; comfort zones, and inefficient detours, the system enables dynamic, socially aware navigation that adapts robot behavior in the presence of humans.\u003c/p\u003e \u003cp\u003eIntegrative semantic methods enrich navigation by combining geometric mapping with contextual layers such as terrain type, functional areas, and human presence. They extend avoidance strategies beyond geometry alone, employing rules, cost-based traversability scores, or reinforcement learning policies to enable safer and more adaptive navigation in complex sites. However, these methods still face limitations: rule-based and cost-based traversability strategies depend heavily on designer assumptions and predefined heuristics for weighting hazards [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e], [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. In practice, the environmental factors considered in the rule or cost are limited and it is impractical to manually encode all the dynamic, contextual, and semantic elements present from construction environments. Moreover, these methods oversimplify the scene by grouping diverse elements into broad categories. This abstraction neglects the semantic meaning and task relevance of different objects. RL-based fine-tuning requires extensive training data and may not generalize well across diverse construction environments [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. Moreover, it is difficult to simulate the complexity of real-world construction sites. Critically, none of these approaches are grounded in empirical evidence of the environmental cues that human operators rely on, leaving a key knowledge gap that motivates the present study.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e2.3 Leveraging human-identified cues to improve navigation\u003c/h2\u003e \u003cp\u003eStudies in autonomous driving have shown that human-identified environmental cues, extracted from video, can significantly improve navigation performance. In these studies, human annotators label both the operators\u0026rsquo; behaviors (e.g., turning, slowing, stopping) and the specific cues that triggered those behaviors (e.g., pedestrians crossing, traffic lights, congestion) [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e], [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. This annotation process links motion patterns with contextual causes, creating datasets that go beyond raw perception or vehicle dynamics.\u003c/p\u003e \u003cp\u003eUsing these labeled datasets, models can be trained not only to predict what action a vehicle should take but also to understand why the action is appropriate. Xu et al. [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e] demonstrated that incorporating object-triggered action annotations improved action prediction accuracy, especially for complex maneuvers such as left or right turns. Similarly, Ramanishka et al. showed that models trained on causally annotated driver behaviors achieved higher accuracy than models using vision or dynamics alone [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. These improvements arise precisely because navigation decisions are grounded in environmental cues that humans judged to be salient.\u003c/p\u003e \u003c/div\u003e"},{"header":"3 Methodology","content":"\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Study design\u003c/h2\u003e \u003cp\u003eThis study employed a naturalistic field design to investigate the environmental cues used by human operators when navigating quadruped robots in construction environments. Naturalistic designs are particularly appropriate for HRI research when the goal is to capture authentic behaviors that emerge in dynamic, real-world contexts, rather than in scripted or laboratory-controlled settings [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e], [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. By situating the study on a live construction site, we ensured ecological validity and captured the complexity of social and environmental interactions that autonomous systems must eventually contend with.\u003c/p\u003e \u003cp\u003eAs shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, The study site was an active construction project on the University of Wisconsin\u0026ndash;Madison campus, which provided the evolving spatial layouts, temporary obstacles, and worker activities characteristic of real-world deployment scenarios. Conducting HRI studies in such naturalistic, high-stakes domains has been shown to yield insights that are not observable in laboratory contexts, particularly regarding the interaction between human activity and robotic behavior [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. Importantly, worker movements were not scripted, and no prior coordination occurred with site staff, ensuring that all interactions unfolded organically during routine operations.\u003c/p\u003e \u003cp\u003eThe robotic task was designed to emulate autonomous construction progress monitoring, a common application of quadruped robots. During each session, the operator manually navigated the robot between predefined waypoints while recording egocentric RGB video of the environment. This task framing aligns with established practices in HRI and field robotics research, where task realism is emphasized to maximize the transferability of study findings to autonomous system design [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. The egocentric video served as the primary data source for linking operator decisions with environmental cues, following methods successfully applied in autonomous driving research [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e], [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe data collection procedure was designed to maximize diversity and capture the evolving nature of construction sites. Five sessions were conducted across both indoor and outdoor environments on May 27, June 1, June 4, June 19, and June 20, 2025. These sessions were spaced to reflect different stages of construction progress and to capture varied spatial configurations, material placements, and levels of human and machinery activity. This temporal sampling strategy aligns with HRI field study practices that emphasize longitudinal and contextually diverse data to improve generalizability of findings [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. Each session was conducted under the supervision of a licensed site engineer, and all safety procedures adhered strictly to site protocols.\u003c/p\u003e \u003cp\u003eThe data analysis process was structured to ensure reliability and reduce bias. Thematic analysis method was used to identify the environmental cues essential for safe navigation on construction site. Two coders were first trained in the study objectives and coding protocol, after which they independently coded the video data to identify instances where operator navigation decisions were linked to environmental cues. A triangulation approach was applied, whereby results were compared and discrepancies discussed until consensus was reached.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Robotic platform\u003c/h2\u003e \u003cp\u003eThe robotic platform for this study was a commercial quadruped robot (Unitree B2 [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]). The robot supports a maximum walking speed of approximately 6 m/s. Its battery life of up to 4\u0026ndash;5 hours per charge allowed for extended recording sessions without frequent interruptions. The robot was equipped with a 360-degree panoramic camera mounted on its body. The camera provided continuous omnidirectional RGB video, capturing the surrounding environmental context. The robot was manually controlled using a handheld controller, which allowed the operator to issue commands for moving forward, backward, turning left and right.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Data collection procedure\u003c/h2\u003e \u003cp\u003eThe data collection was structured to capture navigation behavior under diverse and evolving construction scenarios while maintaining ecological validity. Five field sessions were conducted at the University of Wisconsin\u0026ndash;Madison construction site on May 27, June 1, June 4, June 19, and June 20, 2025. These sessions were strategically spaced over several weeks to reflect different stages of construction progress, enabling the study to capture variations in spatial layout, material placement, and levels of human and machinery activity. Longitudinal sampling of this kind is widely recommended in HRI field studies to improve the robustness and generalizability of findings [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]. Figure\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e shows the data collection on May 27, 2025.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe robot was manually navigated between waypoints that were randomly selected but distributed across the entire site. This approach ensured that the robot\u0026rsquo;s trajectories provided broad spatial coverage while also introducing unpredictability into the navigation task. Randomized waypoint selection exposed the robot to a wide range of scenarios, including varied terrain, equipment zones, and worker activity, while guaranteeing that all major site areas were represented. During navigation, a 360-degree RGB camera mounted on the robot continuously recorded egocentric video of the environment. To capture the operator\u0026rsquo;s perspective and maintain situational awareness, the operator walked behind and followed the robot during each session. This arrangement ensured that the recorded video reflected the visual context available to the human operator when making navigation decisions.\u003c/p\u003e \u003cp\u003eThe sessions were conducted under live construction conditions without scripting or intervention. No communication occurred with site workers in advance, and their movements and activities were left entirely unscripted. This choice ensured that all human\u0026ndash;robot encounters unfolded organically during routine operations, preserving the authenticity of interactions. Such unscripted field studies are critical in HRI because they reveal contextual patterns and environmental cues that cannot be fully replicated in laboratory-based experiments [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe data collection process also prioritized safety and ethical compliance. Each session was supervised by a licensed site engineer, and all activities adhered strictly to construction site safety protocols. The presence of a safety supervisor ensured that both workers and research staff were protected during robot deployment, aligning with best practices for conducting HRI research in high-risk field environments.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e3.4 Data processing procedure\u003c/h2\u003e \u003cp\u003eThe data processing procedure was structured to identify and classify the environmental cues that influenced human navigation decisions. The first step involved video segmentation, which was based on the completion of each navigation trial. Specifically, each segment began when the robot departed from a designated starting point (e.g., point A) and ended upon arrival at a corresponding endpoint (e.g., point B). Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e summarize the manual navigation video segments.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eSummary of manual navigation video segments\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTest date\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCumulative video duration\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNo. of video segments\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMay 27, 2025\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e7 min 28 sec\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eJune 4, 2025\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e37 min 16 sec\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e16\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eJune 6, 2025\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e4 min 8 sec\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eJune 19, 2025\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e7 min 9 sec\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eJune 20, 2025\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e20 min 28 sec\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e11\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eTo analyze the environmental cues underpinning these navigation episodes, we employed thematic analysis, a widely used qualitative method for identifying, analyzing, and reporting patterns (themes) within data. This approach offers both flexibility and methodological rigor and is particularly suited to exploratory studies in dynamic, real-world environments such as construction sites. We followed the six-phase framework proposed by Braun and Clarke [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e], using an inductive, data-driven approach in which themes were grounded in the observed video data rather than shaped by predefined categories. The six phases proceeded as follows:\u003c/p\u003e \u003cp\u003e \u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eFamiliarization with the data \u0026ndash; Coders repeatedly reviewed the egocentric video recordings while making notes on observable operator actions and the surrounding environmental context. This process established a foundational understanding of how navigation behavior emerged in situation.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eGenerating initial codes \u0026ndash; Two coders independently created time-stamped annotations for segments where meaningful navigation actions occurred. Each instance was labeled with the corresponding action, classified into one of four categories: left turn, right turn, stop, or resume walking, and an open-coded description of the environmental cue that appeared to influence that action.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eConstructing initial themes \u0026ndash; The coders collaboratively examined the set of environmental cue codes, grouping similar codes together based on their shared characteristics. These groups formed provisional themes representing broader categories of navigation-relevant cues.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eReviewing themes \u0026ndash; Each provisional theme was evaluated for coherence and distinctiveness. The team revisited all relevant video segments to confirm that coded instances were appropriately assigned, and to resolve any ambiguities or overlaps among themes. Themes that lacked sufficient support or clarity were modified, merged, or discarded.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eDefining and naming themes \u0026ndash; Finalized themes were clearly defined and labeled to capture the essence of the underlying cue category. Definitions emphasized how each cue type contributed to human navigation behavior in the construction context.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eProducing the report \u0026ndash; The finalized thematic structure served as the foundation for our taxonomy of environmental cues. This taxonomy directly informed the presentation of findings and the development of implications for future autonomous navigation systems.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003c/p\u003e \u003cp\u003eTo enhance analytic rigor, we adopted an investigator triangulation strategy. Two coders with prior experience in construction robotics completed the annotation and theme development independently before engaging in consensus-building discussions.\u003c/p\u003e \u003c/div\u003e"},{"header":"4 Results","content":"\u003cp\u003eThrough thematic analysis of egocentric video recordings captured during manual quadruped robot navigation trials, we identified three categories of environmental cues that influenced operator decision-making: topography cues, obstacle and hazard cues, and communicative cues as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e4.1 Topography cues\u003c/h2\u003e \u003cp\u003eTopography cues refer to environmental features that inform whether a given path is physically navigable by the robot. These cues affect stability, traction, and gait planning, and encompass both surface conditions and terrain features. Operators relied on visual assessment of the ground\u0026rsquo;s composition and topography to preemptively adjust heading and avoid risky or difficult to traverse areas. Two subtypes of Topography cues were observed: The first one is surface texture included puddles, sand, gravel, paved areas and so on. These cues influenced decisions to avoid slippage or prevent damage to the robot\u0026rsquo;s hardware. The second is terrain features encompass the macro-structural characteristics of the ground, such as inclines, declines, steps, slopes, and irregular or rugged topography. These features impact the robot's balance and gait and may pose a risk of stumbling or instability, particularly in unstructured environments like construction sites.\u003c/p\u003e \u003cp\u003eFor example, in one trial, the robot approached a section of the site covered by standing water. Recognizing the risk of slipping and potential harm to onboard electronics, the operator steered the robot away from the puddle and selected a drier, more stable path (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e). In another case (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e), the operator identified a zone of uneven terrain ahead. Rather than proceeding directly, they modified the robot\u0026rsquo;s heading to avoid the unstable area and guide it toward a smoother section of the site.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003e4.2 Obstacle and hazard cues\u003c/h2\u003e \u003cp\u003eObstacle and hazard cues include physical entities that obstruct the robot\u0026rsquo;s path or present collision and safety risks. These cues can be static or dynamic and require the operator to engage in real-time avoidance, stopping, or rerouting behaviors. This category consists of: Static objects, such as tools, and construction materials that remain stationary during the navigation sequence. Dynamic objects, including moving vehicles, machinery, suspended loads, and human workers.\u003c/p\u003e \u003cp\u003eIn a representative case, the robot was headed toward a group of pipe covers placed flush with the ground. Although not elevated, the operator identified these as sensitive equipment and took action to steer the robot around them to avoid potential damage (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e). During one session, the robot\u0026rsquo;s path was blocked by two simultaneously moving hazards: suspended construction material being lifted overhead and a concrete mixer truck crossing laterally. The operator promptly stopped the robot and waited until both hazards had cleared the area before resuming movement (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003e4.3 Communicative cues\u003c/h2\u003e \u003cp\u003eCommunicative cues are explicit or implicit signals used to coordinate navigation. These cues serve to convey intent, structure turn-taking, and guide socially appropriate behavior. They may be generated by humans or embedded in the environment as conventional indicators. This category includes human-generated cues, such as hand gestures, body posture, and verbal instructions. Environmentally embedded cues, such as warning signs, cones, barricade tape, and flashing lights.\u003c/p\u003e \u003cp\u003eIn one example, the robot approached a path intersected by a telescopic forklift. The operator halted the robot to avoid conflict. After a short delay, the forklift driver signaled with a hand gesture that it was safe to proceed. The robot operator then resumed navigation (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e). In another example, after observing a red waning line, the operator controlled the robot to turn right to avoid the area surrounded by the waning line (Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"5 Discussion","content":"\u003cp\u003eThis study provides an empirically grounded understanding of the environmental cues that are essential for safe robot navigation in active construction environments. Through naturalistic field observations and thematic analysis of egocentric video data, we developed a taxonomy of three cue categories: topography cues, which pertain to ground surface texture and terrain geometry; obstacle and hazard cues, which involve both static and dynamic objects; and communicative cues, which include human gestures and embedded visual indicators. Together, these findings address a major gap in construction robotics research: the lack of empirical evidence identifying which environmental features are meaningful to human operators during navigation. By documenting these human-identified cues, this study contributes to a richer understanding of how contextual awareness can enhance autonomous navigation in complex, dynamic, and socially populated environments.\u003c/p\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003e5.1 Implications for navigation system design\u003c/h2\u003e \u003cp\u003eThe taxonomy derived from this study offers practical guidance for developing context-aware perception and navigation systems. First, the identification of topography cues emphasizes the importance of fine-grained surface assessment beyond simple geometric mapping. While traditional SLAM-based systems can detect obstacles, they often fail to distinguish between traversable and unstable terrain. Our findings suggest that surface texture and terrain geometry, such as puddles, loose gravel, or inclines, are salient features that humans continuously evaluate to ensure stability and safety. Incorporating these topographic considerations into perception modules may enhance a robot\u0026rsquo;s ability to assess risk and adapt its gait and trajectory accordingly.\u003c/p\u003e \u003cp\u003eSecond, obstacle and hazard cues underscore the need for temporal reasoning in navigation algorithms. Human operators exhibited not only obstacle avoidance but also anticipatory behaviors, such as pausing when observing moving machinery or suspended loads. This indicates that effective autonomous navigation in construction contexts requires systems capable of predicting object motion and integrating short-term forecasting into their path planning.\u003c/p\u003e \u003cp\u003eThird, communicative cues highlight the role of social intelligence in robot navigation. Operators relied on both explicit gestures from workers and implicit environmental signals, such as barricade tape or warning signs, to guide their actions. Current navigation systems largely ignore these social and semiotic dimensions, yet they are essential for safe and cooperative operation in shared workspaces. Future design should therefore incorporate multimodal perception that can interpret human gestures, body posture, and visual indicators, aligning robotic behavior with human expectations and safety norms.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003e5.2 Limitations and future work\u003c/h2\u003e \u003cp\u003eAlthough this study provides novel insights, several limitations must be acknowledged. The data were collected from a single construction site and one operator, which constrains the diversity of environmental and behavioral conditions captured. Future research should replicate this study across multiple sites and operators to examine the consistency and variability of cue use. Additionally, our analysis focused on visual data; integrating other sensory modalities such as audio or tactile feedback may yield a more complete picture of how operators assess environmental risk and intent. Finally, while this study identifies which cues humans rely on, future work should translate these findings into algorithmic implementations.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003e5.3 Broader impacts and generalizability\u003c/h2\u003e \u003cp\u003eBeyond construction robotics, the broader significance of this work lies in its methodological and conceptual contributions to HRI. Methodologically, it demonstrates the value of naturalistic field studies in capturing authentic interactions between humans, robots, and their environments\u0026mdash;interactions that are often obscured in simulation-based or lab-controlled research. Conceptually, it highlights the importance of environmental cues as an organizing principle for designing socially and contextually intelligent robotic behavior. As human-robot collaboration increasingly moves into complex real-world domains, systems that understand and act on nuanced environmental cues will be essential for safe, effective, and socially acceptable autonomy.\u003c/p\u003e \u003c/div\u003e"},{"header":"6 Conclusion","content":"\u003cp\u003eAutonomous navigation in construction environments continues to present significant challenges due to their complex spatial layouts, changing obstacles, and the need for interaction with human workers. Existing approaches often rely on manually programmed rules and costs to interpret environmental cues (e.g., maintaining a minimum distance of 0.3 meters from obstacles, assigning higher costs to uneven terrain or proximity to obstacles), but lack empirical insight into which specific cues are critical for safe navigation. To address this gap, we conducted a naturalistic field study informed by methods used in autonomous driving research. We deployed a manually controlled quadruped on an active site and collected egocentric video using naturalistic field methods. After that, we analyzed the videos using thematic analysis to identify the environmental features that influenced human navigation decisions. Our findings resulted in a taxonomy of three categories of environmental cues essential for safe navigation on construction site. Topography cues include features of the terrain and ground conditions that affect robot stability and gait. Obstacle and hazard cues consist of both stationary and moving elements that may require avoidance or path redirection. Communicative cues include signals from human workers or environmental markers, such as gestures, signs, or caution tape. These findings provide an empirically grounded understanding of the environmental cues that are essential for safe robot navigation in active construction environments.\u003c/p\u003e \u003cp\u003eHowever, this study has several limitations. The dataset derives from one site and a single operator, which supports analytic (mechanism-level) generalization rather than prevalence across projects, robots, or teams. Moreover, our focus on visual evidence omits potentially relevant auditory and haptic cues.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003eACKNOWLEDGMENTS\u003c/p\u003e\n\u003cp\u003eThis paper is based in part upon work supported by the Wisconsin Alumni Research Foundation (WARF) under Project No. AAM3225. We also gratefully acknowledge JP Cullen for facilitating site access and providing logistical support during on-site data collection and analysis. Any opinions, findings, and conclusions or recommendations expressed in this paper are those of the author(s) and do not necessarily reflect the views of WARF or JP Cullen.\u003c/p\u003e\n\u003cp\u003eDATA AVAILABILITY STATEMENT\u003c/p\u003e\n\u003cp\u003eAvailable with restricted access. The field data that support the findings of this study (including egocentric 360\u0026deg; video recordings and associated robot log data collected on an active construction site) are subject to confidentiality agreements with the industry partner and contain information that could compromise site privacy. For these reasons, the data are not publicly available. An anonymized subset of the data may, however, be made available from the corresponding author upon reasonable request and with the permission of the collaborating construction company and the authors\u0026rsquo; institution.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval\u003c/strong\u003e This study involved non-interventional observational data collection on an active construction site. Video data were collected with the consent of JP Cullen, and a site engineer supervised the data collection process to prevent safety incidents. All video used for analysis was de-identified prior to analysis by blurring workers\u0026rsquo; faces and masking identifying marks and logos, and no demographic or personally identifiable information about workers was collected.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflict of interest\u003c/strong\u003e The authors have no competing interests to declare that are relevant to the content of this article.\u0026nbsp;\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eH. Yue, Y. Sun, N. Zeng, S. Chen, Y. Tan, and Q. Wang, \u0026ldquo;Legged Robots for Construction Management: Applications and Challenges,\u0026rdquo; \u003cem\u003eJournal of Construction Engineering and Management\u003c/em\u003e, vol. 151, no. 8, p. 03125006, Aug. 2025, doi: 10.1061/JCEMD4.COENG-16185.\u003c/li\u003e\n\u003cli\u003eP. Biswal and P. K. Mohanty, \u0026ldquo;Development of quadruped walking robots: A review,\u0026rdquo; \u003cem\u003eAin Shams Engineering Journal\u003c/em\u003e, vol. 12, no. 2, pp. 2017\u0026ndash;2031, June 2021, doi: 10.1016/j.asej.2020.11.005.\u003c/li\u003e\n\u003cli\u003eP. Kim, J. Chen, J. Kim, and Y. K. Cho, \u0026ldquo;SLAM-Driven Intelligent Autonomous Mobile Robot Navigation for Construction Applications,\u0026rdquo; in \u003cem\u003eAdvanced Computing Strategies for Engineering\u003c/em\u003e, I. F. C. Smith and B. Domer, Eds., Cham: Springer International Publishing, 2018, pp. 254\u0026ndash;269. doi: 10.1007/978-3-319-91635-4_14.\u003c/li\u003e\n\u003cli\u003eT. Ren and H. Jebelli, \u0026ldquo;Efficient 3D robotic mapping and navigation method in complex construction environments,\u0026rdquo; \u003cem\u003eComputer-Aided Civil and Infrastructure Engineering\u003c/em\u003e, vol. 40, no. 12, pp. 1580\u0026ndash;1605, 2025, doi: 10.1111/mice.13353.\u003c/li\u003e\n\u003cli\u003eM. Wigness, J. G. Rogers, C.-E. Tsai, C. Mertz, L. Navarro-Serment, and J. Oh, \u0026ldquo;Using Perception Cues for Context-Aware Navigation in Dynamic Outdoor Environments,\u0026rdquo; \u003cem\u003eField Robotics\u003c/em\u003e, vol. 1, pp. 1\u0026ndash;33, Oct. 2021, doi: 10.55417/fr.2021001.\u003c/li\u003e\n\u003cli\u003eZ. Alenzi, E. Alenzi, M. Alqasir, M. Alruwaili, T. Alhmiedat, and O. M. Alia, \u0026ldquo;A Semantic Classification Approach for Indoor Robot Navigation,\u0026rdquo; \u003cem\u003eElectronics\u003c/em\u003e, vol. 11, no. 13, p. 2063, Jan. 2022, doi: 10.3390/electronics11132063.\u003c/li\u003e\n\u003cli\u003eJ. Crespo, J. C. Castillo, O. M. Mozos, and R. Barber, \u0026ldquo;Semantic Information for Robot Navigation: A Survey,\u0026rdquo; \u003cem\u003eApplied Sciences\u003c/em\u003e, vol. 10, no. 2, p. 497, Jan. 2020, doi: 10.3390/app10020497.\u003c/li\u003e\n\u003cli\u003eZ. Jiang, \u0026ldquo;A Review of Recent Developments in Quadruped Robots,\u0026rdquo; \u003cem\u003eTheoretical and Natural Science\u003c/em\u003e, vol. 79, pp. 28\u0026ndash;32, Jan. 2025, doi: 10.54254/2753-8818/2025.19933.\u003c/li\u003e\n\u003cli\u003eY. Xu, B. Li, S. Yang, Y. Zhi, R. Li, and T. Chen, \u0026ldquo;MSOT-Q: Multi-Sensor Fusion SLAM with Object Detection Tracking for Quadruped Robots,\u0026rdquo; in \u003cem\u003e2024 36th Chinese Control and Decision Conference (CCDC)\u003c/em\u003e, May 2024, pp. 4485\u0026ndash;4492. doi: 10.1109/CCDC62350.2024.10587825.\u003c/li\u003e\n\u003cli\u003eT. Guan, Z. He, R. Song, D. Manocha, and L. Zhang, \u0026ldquo;TNS: Terrain Traversability Mapping and Navigation System for Autonomous Excavators,\u0026rdquo; May 05, 2022, \u003cem\u003earXiv\u003c/em\u003e: arXiv:2109.06250. doi: 10.48550/arXiv.2109.06250.\u003c/li\u003e\n\u003cli\u003eV. Ramanishka, Y.-T. Chen, T. Misu, and K. Saenko, \u0026ldquo;Toward Driving Scene Understanding: A Dataset for Learning Driver Behavior and Causal Reasoning,\u0026rdquo; presented at the Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018, pp. 7699\u0026ndash;7707. Accessed: Sept. 05, 2025. [Online]. Available: https://openaccess.thecvf.com/content_cvpr_2018/html/Ramanishka_Toward_Driving_Scene_CVPR_2018_paper.html\u003c/li\u003e\n\u003cli\u003eY. Xu \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;Explainable Object-Induced Action Decision for Autonomous Vehicles,\u0026rdquo; presented at the Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020, pp. 9523\u0026ndash;9532. Accessed: Sept. 05, 2025. [Online]. Available: https://openaccess.thecvf.com/content_CVPR_2020/html/Xu_Explainable_Object-Induced_Action_Decision_for_Autonomous_Vehicles_CVPR_2020_paper.html\u003c/li\u003e\n\u003cli\u003eS. Halder, K. Afsari, E. K. Chiou, R. N. C. Patrick, and K. Hamed, \u0026ldquo;Construction inspection \u0026amp; monitoring with quadruped robots in future human-robot teaming: A preliminary study,\u0026rdquo; \u003cem\u003eJournal of Building Engineering\u003c/em\u003e, 2023, doi: 10.1016/j.jobe.2022.105814.\u003c/li\u003e\n\u003cli\u003eA. Aziz, P. Herbers, H. Bayer, and M. K\u0026ouml;nig, \u0026ldquo;Fully Autonomous Fire Safety Equipment Inspection Missions on a Legged Robot,\u0026rdquo; pp. 804\u0026ndash;812, Jan. 2024, doi: 10.1061/9780784485224.097.\u003c/li\u003e\n\u003cli\u003eY. Jang, W. Seol, K. Lee, K.-S. Kim, and S. Kim, \u0026ldquo;Development of quadruped robot for inspection of underground pipelines in nuclear power plants,\u0026rdquo; \u003cem\u003eElectronics Letters\u003c/em\u003e, vol. 58, no. 6, pp. 234\u0026ndash;236, 2022, doi: 10.1049/ell2.12414.\u003c/li\u003e\n\u003cli\u003eH. Kolvenbach \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;Towards autonomous inspection of concrete deterioration in sewers with legged robots,\u0026rdquo; \u003cem\u003eJournal of Field Robotics\u003c/em\u003e, vol. 37, no. 8, pp. 1314\u0026ndash;1327, 2020, doi: 10.1002/rob.21964.\u003c/li\u003e\n\u003cli\u003eX. Zhao, Y. Jin, N. M. Selvaraj, M. Ilyas, and C. C. Cheah, \u0026ldquo;Platform-independent visual installation progress monitoring for construction automation,\u0026rdquo; \u003cem\u003eAutomation in Construction\u003c/em\u003e, vol. 154, p. 104996, Oct. 2023, doi: 10.1016/j.autcon.2023.104996.\u003c/li\u003e\n\u003cli\u003eS. Halder, K. Afsari, J. Serdakowski, S. DeVito, M. Ensafi, and W. Thabet, \u0026ldquo;Real-Time and Remote Construction Progress Monitoring with a Quadruped Robot Using Augmented Reality,\u0026rdquo; \u003cem\u003eBuildings\u003c/em\u003e, vol. 12, no. 11, p. 2027, Nov. 2022, doi: 10.3390/buildings12112027.\u003c/li\u003e\n\u003cli\u003eY. Tao, M. Popović, Y. Wang, S. T. Digumarti, N. Chebrolu, and M. Fallon, \u0026ldquo;3D Lidar Reconstruction with Probabilistic Depth Completion for Robotic Navigation,\u0026rdquo; in \u003cem\u003e2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)\u003c/em\u003e, Oct. 2022, pp. 5339\u0026ndash;5346. doi: 10.1109/IROS47612.2022.9981531.\u003c/li\u003e\n\u003cli\u003eD. Jiang \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;An Integrated Autonomous Dynamic Navigation Approach toward a Composite Air\u0026ndash;Ground Risk Construction Scenario,\u0026rdquo; \u003cem\u003eSensors\u003c/em\u003e, vol. 24, no. 1, Art. no. 1, Jan. 2024, doi: 10.3390/s24010221.\u003c/li\u003e\n\u003cli\u003eD. Hu, V. J. L. Gan, and C. Yin, \u0026ldquo;Robot-assisted mobile scanning for automated 3D reconstruction and point cloud semantic segmentation of building interiors,\u0026rdquo; \u003cem\u003eAutomation in Construction\u003c/em\u003e, vol. 152, p. 104949, Aug. 2023, doi: 10.1016/j.autcon.2023.104949.\u003c/li\u003e\n\u003cli\u003eS. Karimi, R. G. Braga, I. Iordanova, and D. St-Onge, \u0026ldquo;Semantic Optimal Robot Navigation Using Building Information on Construction Sites,\u0026rdquo; \u003cem\u003eISARC Proceedings\u003c/em\u003e, vol. 2021 Proceedings of the 38th ISARC, Dubai, UAE, pp. 57\u0026ndash;64, Nov. 2021, doi: 10.22260/ISARC2021/0010.\u003c/li\u003e\n\u003cli\u003eY. Wu, J. Wei, J. Oh, and D. C. Llach, \u0026ldquo;Towards Human-Centered Construction Robotics: A Reinforcement Learning-Driven Companion Robot for Contextually Assisting Carpentry Workers,\u0026rdquo; in \u003cem\u003e2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)\u003c/em\u003e, Oct. 2024, pp. 90\u0026ndash;97. doi: 10.1109/IROS58592.2024.10802698.\u003c/li\u003e\n\u003cli\u003eR. G. Braga, M. O. Tahir, S. Karimi, U. Dah-Achinanon, I. Iordanova, and D. St-Onge, \u0026ldquo;Intuitive BIM-aided robotic navigation and assets localization with semantic user interfaces,\u0026rdquo; \u003cem\u003eFront Robot AI\u003c/em\u003e, vol. 12, p. 1548684, Mar. 2025, doi: 10.3389/frobt.2025.1548684.\u003c/li\u003e\n\u003cli\u003eY. Wu, E. Erdolu, J. Wei, J. Oh, and D. Cardoso Llach, \u0026ldquo;Robot in the loop: a human-centered approach to contextualizing AI and robotics in construction,\u0026rdquo; \u003cem\u003eConstr Robot\u003c/em\u003e, vol. 9, no. 1, p. 1, Dec. 2024, doi: 10.1007/s41693-024-00144-8.\u003c/li\u003e\n\u003cli\u003eE. Eriksson, T. R. Hansen, and A. Lykke-Olesen, \u0026ldquo;Reclaiming public space: designing for public interaction with private devices,\u0026rdquo; in \u003cem\u003eProceedings of the 1st international conference on Tangible and embedded interaction\u003c/em\u003e, in TEI \u0026rsquo;07. New York, NY, USA: Association for Computing Machinery, Feb. 2007, pp. 31\u0026ndash;38. doi: 10.1145/1226969.1226976.\u003c/li\u003e\n\u003cli\u003eM. Stenmark, M. Haage, and E. A. Topp, \u0026ldquo;Simplified Programming of Re-usable Skills on a Safe Industrial Robot: Prototype and Evaluation,\u0026rdquo; in \u003cem\u003eProceedings of the 2017 ACM/IEEE International Conference on Human-Robot Interaction\u003c/em\u003e, in HRI \u0026rsquo;17. New York, NY, USA: Association for Computing Machinery, Mar. 2017, pp. 463\u0026ndash;472. doi: 10.1145/2909824.3020227.\u003c/li\u003e\n\u003cli\u003eA. Saupp\u0026eacute; and B. Mutlu, \u0026ldquo;The Social Impact of a Robot Co-Worker in Industrial Settings,\u0026rdquo; in \u003cem\u003eProceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems\u003c/em\u003e, in CHI \u0026rsquo;15. New York, NY, USA: Association for Computing Machinery, Apr. 2015, pp. 3613\u0026ndash;3622. doi: 10.1145/2702123.2702181.\u003c/li\u003e\n\u003cli\u003eN. Hudson, C. Alcock, and P. K. Chilana, \u0026ldquo;Understanding Newcomers to 3D Printing: Motivations, Workflows, and Barriers of Casual Makers,\u0026rdquo; in \u003cem\u003eProceedings of the 2016 CHI Conference on Human Factors in Computing Systems\u003c/em\u003e, in CHI \u0026rsquo;16. New York, NY, USA: Association for Computing Machinery, May 2016, pp. 384\u0026ndash;396. doi: 10.1145/2858036.2858266.\u003c/li\u003e\n\u003cli\u003eO. Alvarado and A. Waern, \u0026ldquo;Towards Algorithmic Experience: Initial Efforts for Social Media Contexts,\u0026rdquo; in \u003cem\u003eProceedings of the 2018 CHI Conference on Human Factors in Computing Systems\u003c/em\u003e, in CHI \u0026rsquo;18. New York, NY, USA: Association for Computing Machinery, Apr. 2018, pp. 1\u0026ndash;12. doi: 10.1145/3173574.3173860.\u003c/li\u003e\n\u003cli\u003e\u0026ldquo;Robot Dog B2_Inspection Robot_B2 Robot Dog Advantage Usage | Unitree Robotics.\u0026rdquo; Accessed: Sept. 26, 2025. [Online]. Available: https://www.unitree.com/b2\u003c/li\u003e\n\u003cli\u003eO. Alvarado and A. Waern, \u0026ldquo;Towards Algorithmic Experience: Initial Efforts for Social Media Contexts,\u0026rdquo; in \u003cem\u003eProceedings of the 2018 CHI Conference on Human Factors in Computing Systems\u003c/em\u003e, in CHI \u0026rsquo;18. New York, NY, USA: Association for Computing Machinery, Apr. 2018, pp. 1\u0026ndash;12. doi: 10.1145/3173574.3173860.\u003c/li\u003e\n\u003cli\u003eA. Saupp\u0026eacute; and B. Mutlu, \u0026ldquo;The Social Impact of a Robot Co-Worker in Industrial Settings,\u0026rdquo; in \u003cem\u003eProceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems\u003c/em\u003e, in CHI \u0026rsquo;15. New York, NY, USA: Association for Computing Machinery, Apr. 2015, pp. 3613\u0026ndash;3622. doi: 10.1145/2702123.2702181.\u003c/li\u003e\n\u003cli\u003eV. Braun and V. Clarke, \u0026ldquo;Using thematic analysis in psychology,\u0026rdquo; \u003cem\u003eQualitative Research in Psychology\u003c/em\u003e, vol. 3, no. 2, pp. 77\u0026ndash;101, Jan. 2006, doi: 10.1191/1478088706qp063oa.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Autonomous navigation, construction robots, environmental cues, robots in the wild, video analysis","lastPublishedDoi":"10.21203/rs.3.rs-8254973/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8254973/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eAutonomous robot navigation in construction environments remains a significant challenge due to their complex spatial configurations, dynamic obstacles, and socially interactive nature. Existing approaches rely on pre-programmed rules and costs to interpret environment cues (e.g., maintaining a minimum distance of 0.3 meters from obstacles, assigning higher costs to uneven terrain or proximity to obstacles). However, there is no empirical evidence identifying which specific cues are essential for safe and effective navigation in these settings. Research in autonomous driving has demonstrated that analyzing human drivers’ behavior through egocentric video can reveal the environmental cues critical to safe navigation. Inspired by this approach, we investigate theenvironmental cues that inform human navigation behavior during manual control of a quadruped robot on an active construction site. We deployed a manually controlled quadruped on an active site and collected egocentric video using naturalistic field methods. After that, we analyzed the videos using thematic analysis to identify the environmental features that triggered operators’ actions, such as stopping and turning. Our analysis resulted in a taxonomy of three cue categories: topography cues, which pertain to ground surface texture and terrain geometry; obstacle and hazard cues, which involve both static and dynamic objects; and communicative cues, which include human gestures and embedded visual indicators. These findings provide an empirically grounded understanding of the environmental cues that are essential for safe robot navigation in active construction environments and, in turn, inform the design of perception modules for context-aware, socially competent autonomy.\u003c/p\u003e","manuscriptTitle":"A Taxonomy of Environmental Cues for Quadruped Robot Navigation in Construction Environments Informed by Human Operator Behavior","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-01-14 10:59:32","doi":"10.21203/rs.3.rs-8254973/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"d5327e28-495d-4f3a-8e92-f49b2e8beb1e","owner":[],"postedDate":"January 14th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-03-17T11:55:46+00:00","versionOfRecord":[],"versionCreatedAt":"2026-01-14 10:59:32","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8254973","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8254973","identity":"rs-8254973","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}