Aligning point clouds (PCs) can be used for merging multiple datasets into a globally consistent reference system, mapping new measurements to a reference object to identify features or estimate pose. Goals: train and compare some State-of-the-Art (SoA) PC registration methods; test the best algorithm in a real-case scenario. Given 2 PCs of an object (from CAD and camera or roto-translated CAD), the task is computing an optimal rigid transformation to align them, ensuring the closest match. An efficient deep learning model for this on low-resource computing platforms can greatly improve industrial processes. Start with literature review of existing methods. Among SoA algorithms identified, select some for comparison (including ICP) to observe pros/cons of each.