Our work will produce an efficient map-representation that can be used by NASA's terrestrial experiments today and by missions in the future. We plan to release an open-source implementation for use by NASA. Evidence Meshes support Topic TA04 of NASA's RTA Systems Roadmap as follows: An Evidence Mesh is a natural framework for sensor fusion, which has relevance to NASA-relevant tasks such as grasping and manipulating objects. Manipulation tasks like these often face severe challenges from scale disparity. For example, an ISS map for robotic EVA must be large to support path-planning, yet must have sufficient resolution to model the fine geometry of small tools involved in the task. Planetary-surface rovers will also benefit from enhanced terrain mapping capability. Motion planning over very rough terrain (e.g., cliffs, lava tubes) requires large-scale yet high-fidelity terrain maps. Evidence Meshes are well suited for this becaus they address scale disparity and, because they use triangulated meshes, they support a variety of well-known collision detection algorithms. Their probabilistic framework makes Evidence Meshes a natural representation for planning algorithms that deal with uncertainty, such as probabilistic roadmaps. Evidence Meshes will also benefit existing visualization tools for mission planners and science teams by providing an efficient, compressible data format for transmitting maps via low-bandwidth communications channels.
The challenge of scale disparity abounds in industrial and commercial applications, so Evidence Meshes have clear potential outside NASA. For example, driverless cars model the road ahead with grids that can be on the order of 100 meters but must model potentially dangerous objects that are only tens of centimeters in size. Indoor applications that need fully 3D maps, such as robotic manipulation for assembly or for household tasks, suffer even more from the scale-disparity problem. We see the potential for Evidence Meshes to become an underlying technology for disparate products, including: Map-building software for robotic manipulation, especially in factory and logistics applications. Software that aligns LIDAR scans to produce as-built models. Evidence Meshes could improve the quality of alignment even with uncertainties in LIDAR pose. Perception and navigation software for autonomous unmanned vehicles. Infrastructure-free localization technologies such as SLAM.