Point cloud data is becoming widely available, owing to commodity depth cameras and lidar systems. This creates many opportunities for creating 3D game assets, including environments and objects, from scanned real-world data. The realistic rendering of such data has been widely investigated (e.g. https://hal.inria.fr/hal-01017700/), and can be performed efficiently. However, physical interactions between point clouds, meshes, and other representations are also critical, for gaming and VR applications. These interactions, which include collision, deformation, occlusion, and other effects, may require alternative representations (e.g. https://arxiv.org/abs/1912.12033). This project will investigate geometric and machine learning approaches to these open problems, in typical game environments.