While pair learning (PL) is considered a potential method in STEM education including robotics education (RE), the question of how to pair students for superior learning effects remains. Current grouping strategies for PL in RE are constrained by fixed grouping (FG) and short of attention to dynamic grouping (DG). Therefore, this study aims to introduce DG into RE, as well as exploring its effects on PL for college and primary students. To this end, a two-stage study was conducted. A pre-post-test design was performed in the pilot study. Based on the pilot study, the authors refined the research design and conducted a comparison experiment (DG vs. FG) for primary school students. Results indicated that: (a) there was no significant difference in collaborative ability between DG and FG, but DG outperformed FG in enhancing students' contribution; (b) there was no significant difference between DG and FG in students' learning performance.