Enabled by technological innovations and evolving theories of cognition, embodied learning designs have proliferated over the last few decades. Collaborative tasks in particular offer rich learning opportunities as learners overtly coordinate and negotiate their work. However, less attention has been paid to the ways in which social relationships among participants--be they strangers, friends, or family members--can shape and constrain their movement and learning opportunities. I propose a five-part framework to characterize participants' physical proximity, a marker of familiarity. I demonstrate the efficacy of this framework by using it to analyze video recordings of 41 dyads of families, friends, and strangers as they work on a body-scale geometry and spatial reasoning exhibit at a science museum. Findings suggest that all dyads established successful collaborative movement patterns, though strangers shared space and established physical touch markedly less than family members and friends. Considering these patterns in the design of other collaborative embodied learning activities could create a more comfortable and supportive environment for learners to move and learn together. The analytic framework could also inform the design and evaluation of other movement-based collaborative educational activities.