Human-Grounded Representations for Neuro-Symbolic Monitoring of Multi-Agent Coordination

Embodied multi-agent systems require representations that connect observable actions to their meaning for coordination. Yet the symbolic concepts for representing how humans understand such coordination remain underexplored. to investigate human observation as a source for symbolic concepts, we collected 1,746 time-anchored feedback items from 216 participants who observed 108 multi-agent trajectories in a simulated factory. Our mixed-method analysis identified five dimensions of human-observed coordination: Unit of Observation, Observed Behavior, Monitoring Strategy, Interpreting Behavior, and Feedback Type. Together, these dimensions capture not only which agents, actions, and resources observers attend to, but also the relational and interpretive meanings they attribute to observed behavior and the evaluations or alternatives they express. Observers also differed in which moments they selected and how they interpreted overlapping activity, indicating that perceptual evidence does not necessarily map to a single coordination interpretation. We propose these empirically grounded dimensions as a starting point for symbolic representations that connect embodied behavioral traces with human-interpretable coordination semantics.