Methods and Technologies
A digital twin is a representation of a real object by a computer model that respects the geometry, the appearance, or other properties of the original object. A digital twin can be a visible replica, created by scanning the object with a 3D scanner (e.g., LiDAR), or non-visible. Digital twins are linked to the real objects in the world, and update in near real time to reflect the original object.
Digital twins are important for several reasons, most notably for the sake of analysis, storage, and preservation of objects in the real world to which one might not have access to. Another important and currently relevant use for digital twins is in Augmented, Virtual, and Mixed Reality (referred to collectively as eXtended reality or XR), where the XR system projects to the user digital twins of objects of interest into a virtual world (in the spirit of the Metaverse). One condition for the successful deployment of these twins is accurate localization of the viewing agent, such that the objects are projected at the correct perspective inside the virtual world and in sync with their real twin counterparts. The localization problem becomes significantly more difficult when the scale of the virtual world grows, sometimes to the size of neighborhoods or cities. Another important research problem under digital twinning is the construction of the digital object using 3D reconstruction techniques. The challenges here include the accuracy achievable in the twinning, the time it takes to complete a reconstruction, the detection of changes, and the updates of the twins in near real-time.
My team and I have been working on Simultaneous Localization and Mapping (SLAM) for over 15 years, be it from the localization perspective or that of mapping. We have proposed two unique approaches for SLAM which I believe make our systems stand out: the first one is human-assisted SLAM, which allows a human to mitigate errors in any SLAM system (which is inevitable) and correct the mapping and localization on-the-fly. The second is the hybrid SLAM system we released this year, which uses both direct and indirect SLAM in a complementary manner, yielding a faster and more accurate SLAM. I believe our expertise in SLAM has made our solution attractive to teams looking to apply localization and mapping to their problems.
