Robotic applications like self-driving vehicles rely on Simultaneous Localization and Mapping (SLAM) for their position, which accumulates error over time. It is also requires initialization and periodically fails. This makes place recognition functionality vitally important, allowing recovery in these situations. LIDAR sensors have seen intense research for their immunity to lighting and generation of geometric data, while many recent description approaches have been based on graphs of semantically-relevant objects. This discretization into landmarks is less affected by viewpoint and occlusion, which can disrupt the distribution of single points and thus the effectiveness of global scan descriptors. One such method, Gosmatch, makes use of inter-object Euclidean distance in a series of histogram descriptors. While this approach works, it may be advantageous to incorporate more information into these descriptors. Affine Grassmannian distance, an approach combining relative position and orientation into a distance metric, is a promising approach to accomplish this. In this work we evaluate their suitability as a drop-in replacement for conventional Euclidean distances in the initial matching stage of Gosmatch's approach. As virtually all of Gosmatch's internal descriptors rely on distance histograms in some form, we believe this can provide an indication of the potential overall benefit affine Grassmannian distances offer.