Searching and estimating source information such as the location and release rate, called a source term, have many applications across environmental, medical, and security domains. For autonomous source search and estimation in a turbulent environment, this thesis presents two information-theoretic search strategies. Firstly, Gaussian mixture model (GMM) based infotaxis, termed as GMM-Infotaxis, is presented. The GMM is used to determine the action candidates for the next best informative sampling position in a continuous domain by appropriately clustering possible source locations obtained from the particle filter, compared with Infotaxis using discrete action candidates. This facilitates the better trade-off between exploitation and exploration for search, resulting in more efficient search and better estimation performance. However, GMM-Infotaxis has limitations in complex environments with many obstacles such as urban area, as this approach only predicts one step ahead action and the obstacles prevent efficient search. To address this problem, Infotaxis combined with the Rapidly-exploring Random Trees (RRT) is proposed and termed as RRT-Infotaxis. By introducing new utility function which is designed to maximize entropy reduction and minimize searching path at the same time, RRT-Infotaxis has advantage of searching efficient path in obstacle-rich environments. With proposed utility function, this approach is designed not only to avoid obstacles but also to sample the next best sampling positions considering several steps ahead in a continuous domain. Numerical simulations for both strategies, GMM-Infotaxis and RRT-Infotaxis, are implemented to prove the enhanced performance compared to the conventional Infotaxis. Numerical simulations show that in an open space the performance of GMM-Infotaxis is better than the conventional Infotaxis and in various urban environments RRT-Infotaxis outperforms both original Infotaxis and GMM-Infotaxis. Besides, real outdoor flight experiments using a multirotor UAV in an open space for GMM-Infotaxis are conducted. It shows the superior performance of the GMM-Infotaxis compared with the original Infotaxis method.
Publisher
Ulsan National Institute of Science and Technology (UNIST)