On the use of autonomous unmanned vehicles in response to hazardous atmospheric release incidents

Recent events have induced a surge of interest in the methods of response to releases of hazardous materials or gases into the atmosphere. In the last decade there has been particular interest in mapping and quantifying emissions for regulatory purposes, emergency response, and environmental monitoring. Examples include: responding to events such as gas leaks, nuclear accidents or chemical, biological or radiological (CBR) accidents or attacks, and even exploring sources of methane emissions on the planet Mars. This thesis presents a review of the potential responses to hazardous releases, which includes source localisation, boundary tracking, mapping and source term estimation. Following the review, source term estimation was identified as a promising approach to develop upon during the remainder of the thesis, with mapping to follow. Current literature on source term estimation is focused on using an array of static sensors to infer the location of the source and its emission rate. Formulated as an inverse problem, optimisation or Bayesian inference algorithms are used to fuse point-wise concentration measurements of the hazard with meteorological information and a dispersion model. The inverse problem is highly non-linear, ill-posed and subject to input data that is typically sporadic, noisy and sparse. With the technological developments in sensing and robotics, sensor equipped unmanned vehicles are the modern approach to perform sensing tasks. In this thesis, the use of ground and aerial robots equipped with appropriate hazardous sensors are explored to estimate the source term of an atmospheric release. Previous work on the subject had been limited to simulations or tests using experimental datasets. One of the main aims of this thesis was to extend work on source estimation using mobile sensors from theory and simulations to real world experiments. This aim was achieved for the first time in the literature by the five main contributions of this thesis. A joint Bayesian estimation and planning algorithm was developed to plan the robots path, taking into account the gain in information provided by a new manoeuvre. An experimental set-up was devised to test source estimation algorithms in a controlled environment using a ground robot. Successful experiments were achieved by developing a novel likelihood function to account for the intermittent, noisy readings from short sensor measurement sampling times. An unmanned aerial system was developed for source estimation experiments in uncontrolled outdoor environments and the Bayesian estimation algorithm was extended to consider uncertainty in all the dispersion parameters. After successful experiments the methodology was extended to consider a non-continuously releasing source and mapping algorithms were assessed for particularly unstable atmospheric conditions where the performance of the source estimation algorithms were degraded.