Informative path planner with exploration-exploitation trade-off for radiological surveys in non-convex scenarios

The risk toward human lives in situations involving chemical, biological, radiological, and nuclear (CBRN) threats can be mitigated or even neutralized by deploying carrying a suite of suitable sensors. Furthermore, mobile robots open up the possibility for automated radiological field surveys and monitoring operations, which have important applications in scenarios with CBRN threats. A path planner is one of the essential tools required for these robots to perform their tasks autonomously. Moreover, sophisticated path planners can greatly increase the efficiency of monitoring tasks by maximizing the information gathered in the minimum amount of time. This work proposes an informative path planner as an instrument to efficiently estimate maps of scalar quantities (e.g., radiation intensity, chemical concentration), motivated by applications in radiological inspection. The proposed path planner models the path with B-splines, enabling planning in continuous space. A Gaussian Process with a squared exponential kernel is used to model the underlying field. A modified form of mutual information, estimated from the Gaussian Process, is maximized to determine the most informative path, additionally rewarding observations made in regions where the field magnitude is large (e.g., near a radioactive source). A maximum likelihood estimator for source parameters is used to demonstrate that the proposed solution increases the accuracy of the estimated source positions. Simulation results show that the informative path planner adapts to non-convex environments and increases the number of observations made close to radioactive sources while avoiding obstacles. (C) 2020 Elsevier B.V. All rights reserved.