Path efficient level set estimation for mobile sensors

The interest in using robotic sensors for monitoring spa- tial phenomena is steadily increasing. In the context of environmental analysis, operators typically focus their at- tention where measurements belong to a region of interest (e.g., when monitoring a body of water we might want to determine where the pH level is above a critical threshold). Most of the previous work in the literature represents the environmental phenomena with a Gaussian Process model, and then uses such a model to determine the best locations for measurements [3, 7]. In this paper we consider a specific scenario where a mobile platform with low computational power can continuously acquire measurements with a negli- gible cost. In this scenario, we seek to reduce the distance traveled by the mobile platform as it gathers information and to reduce the computation required by this path se- lection process. Starting from the LSE algorithm [7], we propose two novel approaches, PULSE and PULSE-batch, that exploit a new fast path selection procedure. We eval- uate the effectiveness of our approaches on two datasets: a dataset of the pH level of the water, acquired with a mobile watercraft, and a publicly available dataset that represents CO2 maps. Results show that our techniques can compute informative paths with a computation time that is an order of magnitude lower than other techniques.