An improved mobile robot based gas source localization with temperature and humidity compensation via slam and gas distribution mapping

Abstract

This research is concerned with the problem of localizing gas source in indoor environment using a mobile robot. The problem could be seen as similar to the event of hazardous gas leak in a building. Since the environment is often unknown to the robot, the Simultaneous Localization and Mapping (SLAM) operation is required. Two open source SLAM techniques (i.e. Gmapping and Hector SLAM) were implemented to provide this crucial information. Extensive experiments and analysis on both SLAM techniques yielded that the Hector SLAM is more suitable for gas distribution mapping (GDM) application due to the improved robot pose estimation, less computational requirement and only performs map correction locally. Therefore, the Hector SLAM is combined with Kernel DM+V algorithm to achieve real-time SLAM-GDM for predicting gas source location. Rigorous real-time experiments were conducted to verify the performance of the proposed SLAM-GDM method in an uncontrolled office building with the presence of ethanol emission. The experimental results showed that the prediction of gas source location is often accurate to 0.5 to 2.0m. Furthermore, an Epanechnikov based Kernel DM+V algorithm was also introduced to limit extrapolation range in GDM computations. The observed advantages were lower computational requirement and slightly more accurate prediction on gas source location. More importantly, it was found that the maps produced were able to indicate the areas of unexplored gas distribution and therefore could be used for the robot‘s path planning. The final and the main part of the thesis deals with the effect of ambient temperature and humidity on metal oxide gas sensor (i.e. TGS 2600) response; which could affect the GDM results. Linear regression processes were conducted to create a model to correct the temperature and humidity drift of the gas sensor response. The model (i.e. function) was tested in various configurations and was found to minimize the effects of the two environmental factors on the gas sensor response in different gas concentrations. Finally, two versions of Kernel DM+V/T/H algorithms were proposed and coupled with the drift model to compensate for temperature and humidity variation during the GDM task. The experimental results showed that the Kernel DM+V/T/H algorithms were able to produce more stable gas distribution maps and improve the accuracy of gas source localization prediction by 34%.