Autonomous source search for UAVs using Gaussian mixture model-based Infotaxis: algorithm and flight experiments

Abstract

Estimating and searching source information such as the location and release rate, called a source term, have many applications across environmental, medical, and security domains. The Bayesian inference framework is used to estimate the source term. Since the gas dispersion and its measurements in a turbulent environment are highly non-Gaussian and nonlinear, the particle filter, one of sequential Monte Carlo methods, is adopted in this article. For autonomous source term search in a turbulent environment, this article presents an information-theoretic search approach that combines a widely used Infotaxis approach with the Gaussian mixture model (GMM), termed as GMM-Infotaxis. The GMM is used to determine action candidates for the next best informative sampling position in a continuous domain by appropriately clustering possible source locations obtained from the particle filter. This facilitates a better trade-off between exploitation and exploration for autonomous source search. Through the outdoor experiments, on average, the estimation errors for localization and release rate of the source are reduced 74 and 73%, respectively, while the search time step of the mobile agent is decreased around 18%. Real outdoor flight experiments using a multirotor UAV in various experimental setups show the superior performance of the proposed approach compared with the original Infotaxis method.