Revealing cluster formation over huge volatile robotic data

In this paper, we propose a driven by the robotics field method for revealing global clusters over a fast, huge and volatile stream of robotic data. The stream comes from a mobile robot which autonomously navigates in an unknown environment perceiving it through its sensors. The sensor data arrives fast, is huge and evolves quickly over time as the robot explores the environment and observes new objects or new parts of already observed objects. To deal with the nature of data, we propose a grid-based algorithm that updates the grid structure and adjusts the so far built clusters online. Our method is capable of detecting object formations over time based on the partial observations of the robot at each time point. Experiments on real data verify the usefulness and efficiency of our method.