Information-Based Search for an Atmospheric Release Using a Mobile Robot: Algorithm and Experiments

Michael Hutchinson,Cunjia Liu,Wen-Hua Chen

doi:10.1109/tcst.2018.2860548

Michael Hutchinson, Cunjia Liu + Show 1 more

Open Access

https://doi.org/10.1109/tcst.2018.2860548

Copy DOI

Abstract

Finding the location and strength of an unknown hazardous release is of paramount importance in emergency response and environmental monitoring; thus, it has been an active research area for several years known as source term estimation (STE). This paper presents a joint Bayesian estimation and planning algorithm to guide a mobile robot to collect informative measurements, allowing the source parameters to be estimated quickly and accurately. The estimation is performed recursively using Bayes’ theorem, where uncertainties in the meteorological and dispersion parameters are considered and the intermittent readings from a low-cost gas sensor are addressed by a novel likelihood function. The planning strategy is designed to maximize the expected utility function based on the estimated information gain of the source parameters. Subsequently, this paper presents the first experimental result of such a system in turbulent, diffusive conditions, in which a ground robot equipped with the low-cost gas sensor responds to the hazardous source simulated by incense sticks. The experimental results demonstrate the effectiveness of the proposed estimation and search algorithm for STE based on the mobile robot and the low-cost sensor.

Highlights

R ECENT events have induced a surge of interest in the methods of response to the releases of hazardous materials into the atmosphere [1], [2]
There are a number of reasons for the increased spread of the posterior for the larger release rate estimates: 1) modeled variance was increased with sensed value and the sensed value was larger with higher release rates; 2) a larger release rate leads to the possibility of a stronger source further away causing the increased spread of several posterior parameters; and 3) the final result was further from the prior distribution resulting in more spread
It is beneficial for these parameters to be entered into the algorithm accurately, and it has been shown that the algorithm is robust to quite uninformative prior information

Summary

Introduction

R ECENT events have induced a surge of interest in the methods of response to the releases of hazardous materials into the atmosphere [1], [2]. Examples include: responding to events, such as volcanic eruptions [3], nuclear power accidents [4] or chemical, biological, or radiological accidents or attacks, and even exploring methane emissions on the planet Mars [5]. In many cases, such as nuclear accidents or volcanic eruptions, the location of the source is Manuscript received February 13, 2018; accepted June 26, 2018.

Methods

Results

Conclusion