Simultaneous Robot Localization And Mapping Of Parameterized Spatio-temporal Fields Using Multi-scale Adaptive Sampling

This dissertation presents a Multi-scale Adaptive Sampling (AS) framework for combining measurements arriving from mobile robotic sensors of different scales, rates and accuracies, in order to reconstruct a parametric spatio-temporal field. The proposed sampling algorithm, "EKF-NN-GAS", is based on the Extended Kalman Filter (EKF), Radial Basis Function (RBF) Neural Networks and Greedy Search Heuristics. This novel AS algorithm responds to real-time measurements by continuously directing robots to locations most likely to yield maximum information about the sensed field. EKF is used to derive quantitative information measures for sampling locations. In addition, the localization uncertainty of the robots is minimized by combining the location states and field parameters in a Joint-EKF formulation. This feature is critical in GPS-denied environment such as inside buildings or underwater.
Secondary objectives such as sampling duration, computational cost and energy are minimized by adding several extensions called "Greedy Adaptive Sampling" (GAS) heuristics. The issue of thorough sampling in dense regions is addressed using Clustered Adaptive Sampling. Drawbacks of local searching approach used in GAS are overcome with Non-uniform Grid Size AS and Multi-step AS. The proposed sampling algorithms are compared with traditional raster-scanning through many examples. Results indicate that that the proposed parametric algorithm provides faster convergence with less number of samples. This dissertation also addresses issues of efficient partitioning of the sampling area, distribution of computations and communication for adaptive sampling with multiple robots. The performance of the algorithm was experimentally validated using indoor multi-robot testbed at ARRI's DIAL lab (Distributed Intelligence and Autonomy Lab). A real world scenario of mapping of forest fires is addressed in this thesis in conjunction with the proposed sampling algorithm. Our strategy combines measurements arriving at different times from sensors with different field of view (FOV) and resolution, such as ground, air-borne and space-borne observation platforms. In practice, such robots could be equipped with thermal imaging, topographic mapping and other sensors for measuring environmental conditions.