Issue link: https://htpgraphics.uberflip.com/i/1385717
LEAD RESEARCHER: HORIA-MIHAI BUJANCA I am a researcher building 3D reconstruction and scene understanding solutions for mobile robots. In other words, I use cameras to enable robots to understand "what" is "where" as they move around. Recently, I have focused on evaluating and improving the robustness of robotic systems. RAIN presents an opportunity to understand some of the greatest challenges in building robots which operate safely in extreme environments, informing my work in failure prevention. SUMMARY // The aim of this project is to model the environment of a robot meaningfully, reliably, efficiently, and accurately, and understand the trade-offs between these goals. Given a robot which moves through its environment, the goal is to build a representation of the geometry and semantics (what the geometry represents) of its surroundings. For any task, the reconstruction needs to be done within some constraints (time, power budget, desired accuracy) and without failure in the presence of perturbations (lighting changes, wind, etc). To achieve this goal, the research carried out within the project targets two aspects: benchmarking and development. Benchmarking involves data collection and evaluation across each of the dimensions of interest and allows us to quickly understand what the limitations of the state-of-the-art and of our solutions are and helps us direct our efforts in overcoming such limitations. Development concerns creating a new 3D reconstruction system which is not only better than the state-of-the-art methods but can effortlessly adapt to a variety of use cases. SEMANTIC 3D RECONSTRUCTION AND BENCHMARKING 40 This shows the output of the 3D reconstruction system described. The colours represent semantic labels for each of the recognized objects.