Gaurav Sukhatme, Ph.D.

University of Southern California, School of Engineering

Adaptive Sampling Methods for Robotic Observing Systems

Wednesday - January 21, 2009

Pacific Forum – 3:00 p.m.

We will describe an adaptive sampling algorithm for a robotic sensor network to estimate a scalar field (e.g., the spatial concentration of a particular nutrient in the water). The sensor network consists of static nodes and a mobile robot (a surface or underwater vehicle). The static nodes are able to make sensor measurements continuously in place, while the vehicle is able to move and make measurements at multiple locations. The measurements from the vehicle and the static nodes are used to reconstruct the underlying scalar field. The algorithm accepts the measurements made by the static nodes as inputs and produces an approximate solution to the following problem: With the constraint that the vehicle has limited energy, what path should it take so that the integrated mean square error of the reconstructed field is minimized? Our approach is to treat this problem in two parts. In the first part we determine the appropriate sampling bandwidth (i.e. how many additional samples should the vehicle take to reduce the error below a given threshold?). In the second part, we show how the order of sample collection influences the reconstruction and give a technique for optimal sample ordering. We will discuss the results of using this algorithm on 1. A system consisting of an autonomous surface vehicle and a set of static buoys operating in a lake over several km of traversed distance while reconstructing the temperature field of the lake surface; 2. A cable-driven mobile robot used to measure the fluorescence field in a vertical transect of the same lake. We will also discuss the applicability of the algorithm to a setting where cooperating multiple vehicles are available. We will conclude with an outline of ongoing work in our lab, which incorporates prior models, and examines the effects of communication uncertainty on the system.