The growing presence of autonomous robots in public spaces necessitates a more nuanced understanding of human-robot encounters, extending beyond direct interaction. Traditional human-machine interaction (HMI) research frequently overlooks the experiences of incidentally co-present persons (InCoPs), individuals who do not actively interact with robots but are nonetheless affected by their presence. This study employs an observational methodology to examine these indirect encounters in dynamic environments. A novel framework, the Observational Protocol for Automation (OPA), was developed and validated through three field studies using three robotic systems: Temi, MULI and SPOT. The study categorizes human behaviors in response to robotic presence, distinguishing between active users and passive observers. To ensure methodological rigor, an intercoder reliability (ICR) analysis was conducted, confirming the consistency of behavioral classifications. Findings reveal “interested” as the dominant human-robot encounter category, shaped by robot functionality, environmental context, and methodological constraints. However, challenges emerged in video-based observations, including privacy concerns, technical limitations, and data degradation due to anonymization. Moreover, classification inconsistencies highlighted the necessity for refined behavioral categorization, particularly for ambiguous responses like interest. The ICR problems further underscored the need for precise coding frameworks to minimize observer bias. The study recommends enhanced observer training, improved classification methodologies, and unobtrusive recording technologies to enhance data accuracy. Future research should expand observational approaches to capture more nuanced human behaviors in public human-robot encounters. These findings contribute to a robust framework for understanding human-robot coexistence, offering valuable insights for both academic research and practical robotic deployment strategies.