Progressive mesh decimation by successive edge collapses is a standard tool in geometry processing. A key element of such algorithms is the error metric, which prioritizes the edge collapses to greedily minimize the simplification error. Most previous works focus on preserving local shape properties. However, meshes describing complex systems often require significant decimation for fast transmission and visualization on low-end terminals, and preserving the arrangement of objects is required to maintain the overall system readability for applications such as on-site repair, inspection, training, serious games, etc. We present a novel approach for the joint decimation of multiple triangular meshes. We combine local edge error (e.g. Quadric Error Metric) with a proximity-aware penalty function, which increases the error of edge collapses modifying the geometry in proximity areas. We propose an automatic detection of proximity areas and we demonstrate the performances of our approach on several models generated from CAD scenes. 3 interacting with such data remains very challenging, as the data 4 complexity and scale stress both hardware and software com-5 ponents in real-life applications. This becomes even more criti-6 cal with mobile platforms and web-embedded 3D visualization, 7 which require fast 3D data transfer and rendering even on low 8 end terminals. Despite their limited performances, mobile plat-9 forms are very attractive as they bring access to rich information 10 on-site, and help users performing tasks in complex environ-11 ments (e.g. machinery maintenance). The relatively low cost 12 and ubiquity of mobile devices also make them very attractive 13 as support for teaching and training. 14 To mitigate the increasing complexity of 3D datasets for 15 low-performance hardware, a common solution is to reduce 16 the polygon count of large 3D meshes using decimation algo-17 rithms [1]. In order to keep the decimation algorithms tractable, 18 the preservation of the mesh properties (e.g. parameterization