Localizing Multiple Odor Sources With A Mobile Robot In Time-Varying Airflow Environments Using Dempster-Shafer Inference

This paper addresses the problem of multiple odor sources localization (MOSL) using a mobile robot in a time-varying airflow environment, and provides a localization method which applies the Dempster-Shafer (D-S) theory to map the possible locations of several odor sources and then uses the map to plan a searching route online for the robot. In the proposed method, the robot carries out the D-S inference and iteratively updates a grid map in which each cell has two states (i.e., occupied by an odor source, and not occupied by odor source), using the successive measurements from a gas sensor and an anemometer when the robot follows the dynamic-planned searching route in the given work area. The searching route is planned online to maximize the decline of the uncertainties of the grid cells around the robot. The mapping of the odor sources and the planning of the searching route work side-by-side. Simulations are carried out and the results in a time-varying airflow environment show that the locations of the two odor sources can be estimated online with the D-S inference, and the time cost can be greatly reduced by following the dynamic-planned searching route compared with by following a predefined path shaped like rectangular wave to cover the given searching area.