To Örebro University

In real-world scenarios, gas sensor responses to mixtures of different compositions can be costly to determine a-priori, posing difficulties in identifying the presence of target analytes. In this paper, we propose the use of graph convolutional networks (GCN) to handle gas mixtures with few labelled data. We transform sensor responses into a graph structure using manifold learning and clustering, and then apply GCN for semisupervised node classification. Our approach does not require extensive training data of gas mixtures like many competing approaches, but it outperforms classical semi-supervised learning methods and achieves classification accuracy exceeding 88.5% and over 0.85 Cohen's kappa score given only 5% labelled data for training. This result demonstrates the potential towards realistic gas identification when varied mixtures are present.

Long-term human motion prediction (LHMP) is essential for safely operating autonomous robots and vehicles in populated environments. It is fundamental for various applications, including motion planning, tracking, human-robot interaction and safety monitoring. However, accurate prediction of human trajectories is challenging due to complex factors, including, for example, social norms and environmental conditions. The influence of such factors can be captured through Maps of Dynamics (MoDs), which encode spatial motion patterns learned from (possibly scattered and partial) past observations of motion in the environment and which can be used for data-efficient, interpretable motion prediction (MoD-LHMP). To address the limitations of prior work, especially regarding accuracy and sensitivity to anomalies in long-term prediction, we propose the Laminar Component Enhanced LHMP approach (LaCE-LHMP). Our approach is inspired by data-driven airflow modelling, which estimates laminar and turbulent flow components and uses predominantly the laminar components to make flow predictions. Based on the hypothesis that human trajectory patterns also manifest laminar flow (that represents predictable motion) and turbulent flow components (that reflect more unpredictable and arbitrary motion), LaCE-LHMP extracts the laminar patterns in human dynamics and uses them for human motion prediction. We demonstrate the superior prediction performance of LaCE-LHMP through benchmark comparisons with state-of-the-art LHMP methods, offering an unconventional perspective and a more intuitive understanding of human movement patterns.

Detecting chemical compounds using electronic noses is important in many gas sensing related applications. A gas detection system is supposed to indicate a significant event, such as the presence of new chemical compounds or a noteworthy change of concentration levels. Existing gas detection methods typically rely on prior knowledge of target analytes to prepare a dedicated, supervised learning model. However, in some scenarios, such as emergency response, not all the analytes of concern are a priori known and their presence are unlikely to be controlled. In this paper, we take a step towards addressing this issue by proposing an ensemble learning-based approach (ELBA) that integrates several one-class classifiers and learns online. The proposed approach is initialized by training several one-class models using clean air only. During the sampling process, the initialized system detects the presence of chemicals, allowing to learn another one-class model and update existing models with self-labelled data. We validated the proposed approach with real-world experiments, in which a mobile robot equipped with an e-nose was remotely controlled to interact with different chemical analytes in an uncontrolled environment. We demonstrated that the ELBA algorithm not only can detect gas exposures but also recognize baseline responses under a suspect short-term sensor drift condition. Depending on the problem setups in practical applications, the present work can be easily hybridized to integrate other supervised learning models when the prior knowledge of target analytes is partially available.

To better understand the dynamics in hazardous environments, gas distribution mapping aims to map the gas concentration levels of a specified area precisely. Sampling is typically carried out in a spatially sparse manner, either with a mobile robot or a sensor network and concentration values between known data points have to be interpolated. In this paper, we investigate sequential deep learning models that are able to map the gas distribution based on a multiple time step input from a sensor network. We propose a novel hybrid convolutional LSTM - transpose convolutional structure that we train with synthetic gas distribution data. Our results show that learning the spatial and temporal correlation of gas plume patterns outperforms a non-sequential neural network model.

Emergency response personnel can be exposed to various extreme hazards during the response to natural and human-made disasters. In many of the scenarios, one of the risk factors is the presence of hazardous airborne chemicals. Addressing this risk factor requires typical tiring, taxing and toxic operations that are suitable to be aided by Mobile Robot Olfaction (MRO) techniques. MRO is the research domain combining intelligent mobile robots with an artificial sense of smell. It presents the prospect of practical applications for emergency response as it allows to convey useful information on-site and online without risking the safety of human responders. However, standard gas sampling procedures for laboratory use are not directly applicable to MRO due to the complexity of uncontrolled environments and the need for fast deployment and analysis. Besides, state-of-the-art gas sensing approaches have difficulties handling A Priori Unknown Gases (APUG). In APUG situations, the number or/and identities of the present chemicals are unknown, posing challenges in recognizing the underlying risks with conventional solutions such as supervised learning-based electronic noses or dedicated gas sensors targeting known analytes.

This dissertation focuses on contributions toward real-world applications of robot-aided gas sensing with an APUG problem setup. The dissertation starts with a requirement analysis of Gas Sensing for Emergency Response (GSER) to identify the key tasks in ad hoc applications. Considering that not all analytes of interest in a field application may be known in advance, a pipeline incorporating non-supervised detection and discrimination of multiple chemicals and consequent distribution modelling is found to be important for GSER. The remainder of the thesis fills this pipeline with three steps: 1) An ensemble learning-based gas detection approach is proposed to recognize significant changes from sensor signals as well as model the baseline response pattern. 2) A clustering analysis-based gas discrimination approach is developed to perform online analysis that automatically learns the number of different chemical compounds from the acquired measurements and provides a probabilistic representation of their class labels. 3) The integration of the proposed non-supervised gas detection and gas discrimination approaches with gas distribution modelling allows prototyping of a GSER system, which can enhance emergency responders’ situational awareness in the target environment. This GSER system demonstrates the concept of discriminating and mapping multiple unknown chemical compounds in uncontrolled environments with validation and evaluation using real-world data sets.

During the research on the GSER system, gas dispersal simulation is also investigated to facilitate MRO algorithm development and validation in general. In-field experiments of MRO algorithms are often time-consuming, expensive, cumber some, and lack repeatability, while most of the available simulation tools are limited to insitu gas sensors and simple environments. These issues were addressed by improving a simulation framework to replicate geometrical representations of actual real-world environments and support a variety of gas sensor models. The potential applicability of the resulting work is demonstrated by simulating a gas emission monitoring task and facilitating the development process of a state-of-the-art time-dependent gas distribution modelling algorithm.

Gas identification using an electronic nose (e-nose) typically relies on a multi-class classifier trained with extensive data of a limited set of target analytes. Usually, classification performance degrades in the presence of mixtures that include interferents not represented in the training data. This issue limits the applicability of e-noses in real-world scenarios where interferents are a priori unknown. This paper investigates the feasibility of tackling this particular gas identification problem using one-class learning. We propose several training strategies for a one-class support vector machine to deal with gas mixtures composed of a target analyte and an interferent at different concentration levels. Our evaluation indicates that accurate identification of the presence of a target analyte is achievable if it is dominant in a mixture. For interferent-dominant mixtures, extensive training is required, which implies that an improvement in the generalization ability of the one-class model is needed.

In this work we present a novel multi-sensor unit, a.k.a. FireNose, to detect and discriminate both known and unknown gases in uncontrolled conditions to aid firefighters under harsh conditions. The unit includes three metal oxide (MOX) gas sensors with CMOS micro heaters, a plasmonic enhanced non-dispersive infrared (NDIR) sensor optimized for the detection of CO2, a commercial temperature humidity sensor, and a flow sensor. We developed custom film coatings for the MOX sensors (SnO2, WO3 and NiO) which greatly improved the gas sensitivity, response time and lifetime of the miniature devices. Our proposed system exhibits promising performance for gas sensing in harsh environments, in terms of power consumption (∼ 35 mW at 350°C per MOX sensor), response time (<10 s), robustness and physical size. The sensing unit was evaluated with plumes of gases in both, a laboratory setup on a gas testing rig and on-board a mobile robot operating indoors. These high sensitivity, high-bandwidth sensors, together with online unsupervised gas discrimination algorithms, are able to detect and generate their spatial distribution maps accordingly. In the robotic experiments, the resulting gas distribution maps corresponded well to the actual location of the sources. Therefore, we verified its ability to differentiate gases and generate gas maps in real-world experiments.

Detecting chemical compounds using electronic noses is important in many gas sensing related applications. Existing gas detection methods typically use prior knowledge of the target analytes. However, in some scenarios, the analytes to be detected are not fully known in advance, and preparing a dedicated model is not possible. To address this issue, we propose a gas detection approach using an ensemble of one-class classifiers. The proposed approach is initialized by learning a Mahalanobis-based and a Gaussian based model using clean air only. During the sampling process, the presence of chemicals is detected by the initialized system, which allows to learn a one-class nearest neighbourhood model without supervision. From then on the gas detection considers the predictions of the three one-class models. The proposed approach is validated with real-world experiments, in which a mobile robot equipped with an e-nose was remotely controlled to interact with different chemical analytes in an open environment.

Emergency personnel, such as firefighters, bomb technicians, and urban search and rescue specialists, can be exposed to a variety of extreme hazards during the response to natural and human-made disasters. In many of these scenarios, a risk factor is the presence of hazardous airborne chemicals. The recent and rapid advances in robotics and sensor technologies allow emergency responders to deal with such hazards from relatively safe distances. Mobile robots with gas-sensing capabilities allow to convey useful information such as the possible source positions of different chemicals in the emergency area. However, common gas sampling procedures for laboratory use are not applicable due to the complexity of the environment and the need for fast deployment and analysis. In addition, conventional gas identification approaches, based on supervised learning, cannot handle situations when the number and identities of the present chemicals are unknown. For the purpose of emergency response, all the information concluded from the gas detection events during the robot exploration should be delivered in real time. To address these challenges, we developed an online gas-sensing system using an electronic nose. Our system can automatically perform unsupervised learning and update the discrimination model as the robot is exploring a given environment. The online gas discrimination results are further integrated with geometrical information to derive a multi-compound gas spatial distribution map. The proposed system is deployed on a robot built to operate in harsh environments for supporting fire brigades, and is validated in several different real-world experiments of discriminating and mapping multiple chemical compounds in an indoor open environment. Our results show that the proposed system achieves high accuracy in gas discrimination in an online, unsupervised, and computationally efficient manner. The subsequently created gas distribution maps accurately indicate the presence of different chemicals in the environment, which is of practical significance for emergency response.

Gas discrimination in open and uncontrolled environments based on smart low-cost electro-chemical sensor arrays (e-noses) is of great interest in several applications, such as exploration of hazardous areas, environmental monitoring, and industrial surveillance. Gas discrimination for e-noses is usually based on supervised pattern recognition techniques. However, the difficulty and high cost of obtaining extensive and representative labeled training data limits the applicability of supervised learning. Thus, to deal with the lack of information regarding target substances and unknown interferents, unsupervised gas discrimination is an advantageous solution. In this work, we present a cluster-based approach that can infer the number of different chemical compounds, and provide a probabilistic representation of the class labels for the acquired measurements in a given environment. Our approach is validated with the samples collected in indoor and outdoor environments using a mobile robot equipped with an array of commercial metal oxide sensors. Additional validation is carried out using a multi-compound data set collected with stationary sensor arrays inside a wind tunnel under various airflow conditions. The results show that accurate class separation can be achieved with a low sensitivity to the selection of the only free parameter, namely the neighborhood size, which is used for density estimation in the clustering process.