Introduced around two decades ago, motion averaging provides a conceptually novel pathway to the classic problem of 3D reconstruction. It has since matured into a principled framework for camera motion estimation that hits the sweet spot of theoretical richness and empirical utility. In this talk I will delineate the motivation and development of motion averaging and its role in global SfM. I will illustrate the arguments by examining the paradigmatic method of rotation averaging that affords an interesting interplay of mathematical ideas (geometry, Lie groups, graph theory) and related considerations that arise in practice (optimization, robust statistics). I will also briefly describe the allied problem of translation averaging and outline some of its salient attributes and challenges.