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

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.