Accurately evaluating the behavioral patterns of robots in complex dynamic environments can ensure the reliability of their operation. Among them, trajectory evaluation provides a data foundation for the safety and effectiveness of robots. However, existing single-dimensional evaluation methods fail to capture the full trajectory characteristics, particularly lacking in continuous assessment for trajectory under dynamic environments. To address this issue, this paper proposes a spatio-temporal fusion evaluation method based on signal temporal logic (STL), which aims to integrate the temporal characteristics and spatial distribution to evaluate sequential trajectory. This method initially identifies robots’ behavioral patterns and performance indicators through the analysis of trajectory signals and the extraction of sequential features. Subsequently, the behavior of robots is defined using formal semantic logic by constructing STL formulas in both temporal and spatial domains, and these formulas are optimized through a reinforcement learning algorithm. Finally, logical fusion in the spatio-temporal domain is applied to recognize the assessment levels corresponding to various trajectories. To validate this method, simulation data is utilized to evaluate robot performance across various behaviors, thereby confirming the effectiveness of this evaluation approach.