Gas emissions play a crucial role in many environmental and industrial processes, driving a growing effort to understand their dispersion in air. Nonetheless, gas distribution mapping is inherently challenging due to the complex interplay between gas diffusion and wind flows. Mobile robots provide a compelling alternative to static sensor networks for gas sensing, having greater mobility and minimizing the need to permanently deploy assets in the environment. However, robotic platforms typically collect only sparse measurements due to constraints, such as limited battery life, and state-of-the-art methods often fail to accurately interpolate between scattered data. To address this limitation, we introduce ADApprox, a novel gas mapping algorithm. By leveraging the underlying physics which governs gas dispersion, ADAapprox offers superior interpolation capabilities. Our method locally approximates advection-diffusion equation for an entire grid of points and learns the model parameters from gas measurements. The learned parameters are subsequently used to predict gas concentrations across the entire environment. Extensive simulations and physical experiments are conducted using a nano aerial vehicle. The mapping results demonstrate that ADApprox consistently outperforms the state-of-the-art algorithm Kernel DM+V/W while having a comparable computational cost. In addition, we evaluate the effectiveness in localizing a gas source based on the predicted gas maps. Our findings indicate that ADApprox effectively localizes the gas source, achieving a median error of 18cm on an area of 12m2in physical experiments.

Localizing gas sources is a challenging task due to the complex nature of gas dispersion. Informative Path Planning (IPP) plays a crucial role in guiding robots to sample at high-information positions, thereby accelerating the estimation process. Existing probabilistic gas source localization methods often require robots to halt at sampling positions, averaging gas measurements over time. Consequently, when selecting the next sampling position, information gains are usually computed precisely through computationally heavy procedures, limiting evaluations to a small set of potential positions. In our previous work, we introduced a sense-in-motion strategy that eliminates the need for prolonged stops at sampling points, therefore allowing the incorporation of measurements taken during robot movement. Building upon this advancement, we propose to extend information gain evaluation in a more continuous manner, from a point evaluation to a path evaluation. However, existing IPP methods are too computationally expensive when transitioning from goal-based to region-based evaluations. To address this challenge, we first assess three lightweight information extraction metrics. Based on the selected metrics, we propose a novel IPP algorithm that computes cumulative information gain along the robot’s path and dynamically prioritizes exploration or exploitation based on the uncertainty of the source estimation. The proposed method is extensively evaluated through both high-fidelity simulations and physical experiments. Results show that our proposed method consistently outperforms a benchmark state-of-the-art method, achieving a 40% increase in source localization success rate and halving the experimental time in challenging environments.

Given the patchy nature of gas plumes and the slow response of conventional gas sensors, the use of mobile robots for Gas Source Localization (GSL) tasks presents significant challenges. These aspects increase the difficulties in obtaining gas measurements, encompassing both qualitative and quantitative aspects. Most existing model-based GSL algorithms rely on lengthy stops at each sampling point to ensure accurate gas measurements. However, this approach not only prolongs the time required for a single measurement but also hinders sampling during robot motion, thus exacerbating the scarcity of available gas measurements. In this work, our goal is to push the boundaries in terms of continuity in sampling to enhance system efficiency. Firstly, we decouple and comprehensively evaluate the impact of both plume dynamics and gas sensor properties on the GSL performance. Secondly, we demonstrate that adopting a continuous sampling strategy, which has been generally overlooked in prior research, markedly enhances the system efficiency by obviating the prolonged measurement pauses and leveraging all the data gathered during the robot motion. Thirdly, we further expand the capabilities of the continuous sampling by introducing a novel informative path-planning strategy, which takes into account all the information gathered along the robot's movement. The proposed method is evaluated in both simulation and reality under different scenarios emulating indoor environmental conditions.

The deployment of robots for Gas Source Localization (GSL) tasks in hazardous scenarios significantly reduces the risk to humans and animals. Gas sensing using mobile robots focuses primarily on simplified scenarios, due to the complexity of gas dispersion, with a current trend towards tackling more complex environments. However, most state-of-art GSL algorithms for environments with obstacles only depend on local information, leading to low efficiency in large and more structured spaces. The efficiency of GSL can be improved dramatically by coupling it with a global knowledge of gas distribution in the environment. However, since gas dispersion in a built environment is difficult to model analytically, most previous work incorporating a gas dispersion model was tested under simplified assumptions, which do not take into consideration the impact of the presence of obstacles to the airflow and gas plume. In this paper, we propose a probabilistic algorithm that enables a robot to efficiently localize gas sources in built environments, by combining a state-of-the-art probabilistic GSL algorithm, Source Term Estimation (STE) with a learned plume model. The pipeline of generating gas dispersion datasets from realistic simulations, the training and validation of the model, as well as the integration of the learned model with the STE framework are presented. The performance of the algorithm is validated both in high-fidelity simulations and real experiments, with promising results obtained under various obstacle configurations.

International audience