Probabilistic Boundary Coverage for Unknown Target Fields with Large Perception Uncertainty and Limited Sensing Range

We introduce a new type of probabilistic boundary coverage problem where a robot has to enclose unknown target fields (UTFs) with large perception uncertainty and limited sensing range. When the robot gets closer to UTF and accumulates sufficient sensory readings, it employs Gaussian processes (GPs) as a local belief function to approximate field boundary distribution in an ellipse-shaped local region. The local belief function allows us to predict UTF boundary trends and establish an adjacent ellipse for further exploration. The process is governed by a depth-first search process until UTF is approximately enclosed by connected ellipses when the boundary coverage process ends. We formally prove that our boundary coverage process guarantees the enclosure above a given coverage ratio with a preset probability threshold. We have implemented our algorithm and tested it under different field types in simulation.