Robotic Exploration for Learning Human Motion Patterns

Understanding how people are likely to move is key efficient and safe robot navigation in human environments. However, mobile robots can only observe a fraction of the environment at time, while activity patterns may also change different times. This article introduces new methodology for exploration maximize knowledge by deciding where when collect observations. We introduce an policy driven entropy levels spatio-temporal map pedestrian flows, compare multiple strategies including both informed uninformed approaches. The evaluation performed simulating using real sensory data from three long-term datasets. results show that certain scenarios models built with proposed system better predict flow than strategies, allowing more socially compliant way, ratio factor it comes model prediction accuracy.