Geometric real-time modeling and scanning using distributed depth sensors

Abstract: Technological breakthroughs have in recent years heavily increased the availability of different types of sensors and the amount of computational power, allowing for much more advanced robotic applications. By mounting multiple depth sensors on the robot and designing a fast and robust system with state-of-the-art algorithms we have made the robot aware of its surroundings. First, the surrounding is scanned into a static geometrical model. Then that model is used to detect and extract deviations in new scannings, for further analysis, to make suitable decisions on how the robot should act. We have demonstrated the system in a safety-critical setup for industrial robots, slowing down and stopping the robot when a human is too close. The system is designed for modularity, allowing for many other completely different applications, such as complex object tracking and motion planning.