Volcanoes emit large amounts of CO2, directly influencing human lives. Mapping volcanic gas emissions helps to forecast eruptions and understand the impact of volcanoes on climate and the environment. Drone-based gas sensing significantly reduces risks in volcanic monitoring but faces technical limitations when measuring gas, as rotor downwash disperses the gas plume before detection. Gas Tomography using remote gas sensing addresses this challenge. At the Salinelle dei Cappuccini mud volcanoes, we demonstrate that while drone-mounted in-situ sensors failed to detect CO2 emissions due to aerodynamic disturbance, open-path sensing successfully enabled remote gas distribution mapping. We present a novel model-based gas tomographic reconstruction approach that incorporates a Lagrangian model to compensate for wind-induced advection. The resulting gas distribution maps align with manually collected in-situ measurements, confirming that model-based gas tomography effectively overcomes downwash limitations and enables accurate mapping of volcanic emissions.

Monitoring airborne pollutant emissions and discovering their sources is an important task that mobile robots and Unmanned Aerial Vehicles (UAVs) are well suited for. However, it requires suitable gas sensors compatible with these platforms, alongside sampling strategies that direct them to optimal sampling locations. This article presents a dataset comprising numerous wind tunnel experiments where a synthetic gas plume was systematically scanned using various gas and environmental sensors in a dense 3D grid pattern, complemented by additional sampling trajectories. The dataset allows direct comparison between low-cost metal oxide semiconductor gas sensors (MiCS-5524, MiCS-6814) and advanced photoionization detectors (PID-AH2), examining both their static and dynamic responses to the plume. Additionally, a scale model industrial facility enables the evaluation of gas dispersion models in complex settings under controlled wind conditions. The data support the comparison and evaluation of novel sampling strategies for mobile robotic sensor systems for gas distribution mapping and source localization, providing more realistic experimental data compared to computational fluid dynamics simulations. Meeting the need for objective comparability of approaches in robotic gas sensing, this experimental dataset can serve as a standard corpus and benchmark.

Passive seismic ambient noise interferometry (ANI) has shown potential for lunar seismic exploration, offering the capability to detect near-surface subsurface structures critical for future lunar mission, such as near-surface ice deposits and lava tubes, without the need for active seismic sources. Performing ANI on the Moon can be realized with a multi-agent system, in which a network of individual rovers either carry or deploy seismic receivers. However, these systems have inherent uncertainties in localization and timing. Additionally, methods used to extract dispersion curves from cross-correlations are fundamentally limited in achievable time-frequency resolution, which we demonstrate for the continuous wavelet transform (CWT). Quantifying how these factors propagate into Rayleigh wave velocity estimates is essential for accurate detection of lunar subsurface features. In this study, analytical error formulas are derived and validated through Monte Carlo simulations using passive seismic data from the Apollo 17 lunar seismic profiling experiment. Results indicate that velocity uncertainties due to localization errors remain around an acceptable for realistic positional standard deviations of at the receiver distance of as in the Apollo 17 Lunar Seismic Profiling Experiment. Timing errors induced by clock instabilities are negligible. However, uncertainties in seismic travel-time estimations are significantly dominated by the resolution limits imposed by the CWT. The developed analytical uncertainty model thus provides a critical foundation for designing autonomous lunar seismic networks for future lunar missions.

Full waveform inversion (FWI) provides high-resolution subsurface models but typically lacks measures of uncertainty, which are crucial for assessing imaging reliability. While state-of-the-art probabilistic approaches such as Markov chain Monte Carlo provide rigorous uncertainty estimates, they remain computationally prohibitive for high-dimensional inverse problems. We propose a Hessian-based approach to uncertainty quantification in FWI using a low-rank Gauss-Newton approximation combined with shot-encoding. In numerical experiments, our method achieves covariance estimates close to the original low-rank Gauss-Newton Hessian while reducing computation time by 72%. These results highlight the potential of shot-encoded Gauss-Newton methods to make uncertainty-aware FWI more computationally feasible for large-scale seismic imaging.

Search and rescue missions are often critical following sudden natural disasters or in high-risk environmental situations. The most challenging search and rescue missions involve difficult-to-access terrains, such as dense forests with high occlusion. Deploying uncrewed aerial vehicles for exploration can significantly enhance search effectiveness, facilitate access to challenging environments, and reduce search time. However, in dense forests, the effectiveness of uncrewed aerial vehicles depends on their ability to capture clear views of the ground, necessitating a robust search strategy to optimize camera positioning and perspective. This work presents an optimized planning strategy and an efficient algorithm for the next best view problem in occluded environments. Two novel optimization heuristics, a geometry heuristic, and a visibility heuristic, are proposed to enhance search performance by selecting optimal camera viewpoints. Comparative evaluations in both simulated and real-world settings reveal that the visibility heuristic achieves greater performance, identifying over 90% of hidden objects in simulated forests and offering 10% better detection rates than the geometry heuristic. In addition, real-world experiments demonstrate that the visibility heuristic provides better coverage under the canopy, highlighting its potential for improving search and rescue missions in occluded environments.

Multi-agent systems (MAS) are a promising solution for autonomous exploration tasks in hazardous or remote environments, such as planetary surveys. In such settings, communication among agents is essential to ensure collaborative task execution, yet conventional approaches treat exploration and communication as decoupled subsystems. This work presents a novel framework that tightly integrates semantic communication into the MAS exploration process, adapting communication strategies to the exploration methodology to improve overall task performance. Specifically, we investigate the application of semantic joint source-channel coding (JSCC) with over-the-air computation (AirComp) for distributed function computation for the application of cooperative subsurface imaging using the adapt-then-combine full waveform inversion (ATC-FWI) algorithm. Our results demonstrate that semantic JSCC significantly outperforms classical point-to-point and standard JSCC methods, especially in high-connectivity networks. Furthermore, incorporating side information at the receiving agent enhances communication efficiency and imaging accuracy, a feature previously unexplored in MAS-based exploration. We validate our approach through a use case inspired by subsurface anomaly detection, showing measurable improvements in imaging performance per agent. This work underscores the potential of semantic communication in distributed multi-agent exploration, offering a communication-aware exploration paradigm that achieves task-relevant performance gains.

Future space exploration missions envision multiagent networks autonomously conducting subsurface exploration. Achieving this requires distributed subsurface imaging. Previously, we introduced the adapt-then-combine full waveform inversion (ATCFWI) in both the time and frequency domains, enabling each receiver in a seismic network to generate high-resolution subsurface images locally through data exchange with neighboring receivers. To further enhance imaging quality, we explore the regularization of local cost functions. Specifically, we incorporate Tikhonov and total variation (TV) regularization. Tikhonov regularization promotes smooth structural recovery, while TV regularization enhances edge sharpness and contrast. We integrate both methods into ATCFWI to enable robust imaging in a distributed fashion. Through numerical evaluations, we demonstrate that ATCFWI, combined with these regularization techniques, significantly improves imaging performance across all receivers. Code for our proposed method is available on https://github.com/bshin/fd-atcfwi.

The increasing frequency and severity of forest fires, driven by climate change and human activity, necessitate accurate and efficient modeling of such events. This paper proposes a method leveraging prior knowledge in the form of a stationary, inhomogeneous advection-diffusion Partial Differential Equation (PDE), combined with decentralized, over-thenetwork sparse Bayesian learning (SBL), to enable a multiagent system to localize smoke sources and reconstruct spatial smoke concentration based on collected smoke concentration samples. The approach employs Green’s functions for solving the PDE and nonlinear optimization for grid-less estimation of sources based on cooperatively acquired data. A novel iterative initialization strategy is introduced to enhance convergence and promote sparsity in source localization. Simulations validate the method’s effectiveness.

This research introduces a novel approach to seismic exploration on the Moon and Mars, employing autonomous robotic swarms equipped with seismic sensing and processing hardware. By relying on probabilistic inference methods, we aim to survey large surface areas to both autonomously identify and map subsurface features such as lava tubes and ice deposits. These are crucial for future human habitats and potential in-situ resource utilization. This endeavor presents unique challenges due to the communication limitations and uncertainties inherent in remote, autonomous operations. To address these challenges, we adopt a distributed approach with robotic swarms, where each rover processes seismic data and shares the results with other rovers in its vicinity, contending with imperfect communication links. Thus, the swarm is used as a distributed computing network. The decisions made within the network are based on probabilistic modeling of the underlying seismic inference problem. A key innovation in this respect is the use of factor graphs to integrate uncertainties and manage inter-rover communications. This framework enables each rover to generate a localized subsurface map and autonomously decide on strategic changes in the seismic network topology, either exploring new areas or repositioning to enhance measurement accuracy of targeted underground regions. The vision is to implement this approach on a distributed factor graph, allowing for a coordinated, probabilistic analysis of seismic data across the swarm. This strategy represents a significant departure from traditional static seismic sensor arrays, offering a dynamic and adaptable solution for planetary exploration. The first step towards realizing this vision involves implementing a Kalman filter for the one-dimensional linear heterogeneous wave equation. This has been achieved by reformulating finite difference schemes for wave propagation simulation into a state-space description. The resulting linear continuous n-th order system can be explicitly solved and rewritten into a discrete state space model that can be used in the standard Kalman filter recursion. However, the standard Kalman filter is limited due to its assumption that both model and process noise are Gaussian. With factor graphs, this limitation can be overcome, enabling a more robust and versatile analysis. Several simulation results will be shown to demonstrate the performance of these approaches. We intend to extend the approach to higher-dimensional problems, implementing distributed versions of the Kalman filter and factor graph with simulated, non-perfect communication links. Eventually, the seismic inverse problems will be solved in these frameworks. Successfully achieving these objectives could greatly enhance our capabilities in extraterrestrial exploration, paving the way for more informed and efficient future space missions.

Swarms of drones offer increased sensing aperture. When these swarms mimic natural behaviors, sampling is enhanced by adapting the aperture to local conditions. We demonstrate that this enables detection and tracking of heavily occluded targets. Object classification in conventional aerial images generalizes poorly due to occlusion randomness and is inefficient even under minimal occlusion. In contrast, anomaly detection applied to synthetic-aperture integral images remains robust in dense vegetation and independent of pre-trained classes. Our autonomous, centralized swarm searches for unknown or unexpected occurrences, tracking them while continuously adapting its sampling pattern to optimize local viewing conditions. We achieved average positional accuracies of 0.39 m with average precisions of 93.2% and average recalls of 95.9%. Here, adapted particle swarm optimization considers detection confidences and predicted target appearance. We present a new confidence metric that identifies the most abnormal targets and show that sensor noise can be effectively included in the synthetic aperture process, removing the need for costly optimization of high-dimensional parameter spaces. Finally, we provide a hardware-software framework enabling low-latency transmission and fast processing of video and telemetry data. Although our field experiments involved six drones, ongoing technological advances will soon enable larger, faster swarms for military and civil applications.

Ground motion observations on planetary objects are a prerequisite for a detailed understanding of their interior structure and evolution. The imaging of the near surface structure - in particular on the Moon - has strong practical implications. First, the race is on to detect ice-bearing rocks near the surface from which water could be extracted and used as a resource for crewed missions. Second, due to the substantial bombardment of the lunar surface with meteorites and the lack of an atmosphere, observatories or habitats may have to be built underground. It has been proposed that cavities from ancient lava flows below the lunar surface could be used to place infrastructure. Current mission plans for geophysical exploration focus on static seismic sensors/arrays that would be restricted to the area they can explore. With the NEPOS project we want to go beyond these restrictions and develop concepts for mobile seismic arrays that work in an autonomous way using robotic technology. The scientific challenges include the understanding of wavefield effects of icy rocks and caves in a strongly scattering environment, the provision of optimal source-receiver configurations to detect them, as well as an integrated data-processing workflow from observation to subsurface image including the quantification of uncertainties. In order to solve these challenges, we first developed a Digital Twin for wave propagation in the strongly heterogeneous lunar crust to generate synthetic seismic data using the spectral element code SALVUS. We compared the synthetic seismograms to data from the Apollo 17 Lunar Seismic Profiling Experiment (LPSE) and find that their main characteristics coincide. We further generated synthetic seismograms for a variety of network configurations and subsurface heterogeneities, which will be used to test appropriate imaging methods for the lunar subsurface structure. Due to the presence of strongly scattering media ambient noise tomography seems to be a promising method, as was already shown in previous studies. We apply seismic interferometry to LPSE data, as well as to our synthetic seismograms, to reconstruct Green’s functions, which give us information on the subsurface properties.

In this paper, we present a complete analytic probability based description of mobile-to-mobile uncorrelated scatter channels. We provide a theoretical proof that the proposed probability based description is equivalent to the correlation based description introduced by Bello and Matz. This equivalence is evaluated through a comparison of the hybrid characteristic probability density function with the correlation based description of a measured generic mobile-to-mobile channel, both of which can be obtained directly either from theory or from measurement data. The comparison confirms the similarity between the probability based and correlation based description qualitatively and quantitatively. Thus, the proposed probabilistic description complements the common correlation based description providing a comprehensive theoretical description of arbitrary uncorrelated scatter channels.

This work presents a multi-agent system for the adaptive detection and localization of gas sources, based on sequentially collected concentration measurements. Gas distribution is modeled using stationary, non-homogeneous advection-diffusion Partial Differential Equation (PDE), driven by a superposition of unknown, arbitrarily located Dirac measures that represent the gas sources. To estimate both the number and locations of these sources, sparse Bayesian learning (SBL) is integrated with a combine-then-adapt gradient optimization strategy, resulting in an adaptive inversion algorithm capable of estimating both the source support and the concentration field in 2D. Unlike current approaches that require access to the full dataset, the proposed method operates adaptively, enabling real-time estimation achieving a similar estimation accuracy. This results in a more robust and efficient solution for gas source localization in complex environments using realistic sensors, while also offering greater flexibility in data collection.

Airflow is the key transport mechanism for airborne substances like gas or particulate matter. It is of great interest in many applications ranging from evacuation planning to analyzing indoor ventilation systems. However, accurately determining a spatial map of the airflow is difficult and time-consuming since environmental parameters and boundary conditions are often unknown. This work introduces a novel adaptive sampling strategy for mobile robots. The strategy allows multiple mobile robots with anemometers to autonomously collect airflow measurements and generate a two-dimensional spatial map of the airflow field. Using a Domain-knowledge Assisted Exploration approach, the robots respond in real-time to the measurements already taken and determine the most informative locations online for further measurements. We incorporate the Navier-Stokes Partial Differential Equations to fuse the collected data with model assumptions. By casting the airflow model into a probabilistic framework, we can quantify uncertainties in the airflow field and develop an intelligent, uncertainty-driven exploration strategy inspired by optimal experimental design principles. This strategy combines an estimated uncertainty map with a rapidly exploring random tree path planner. Additionally, using the Navier-Stokes equations allows us to interpolate spatially between measurements in a physics-informed way, enabling us to construct a more accurate airflow map. We implemented and evaluated the proposed concept in simulations and experiments in a laboratory environment, where five mobile robots explore artificially generated airflow fields. The results indicate that our approach can correctly estimate the airflow and show that the proposed adaptive exploration strategy gathers information more efficiently than a predefined sampling pattern.

We propose a guidance strategy to optimize real-time synthetic aperture sampling for occlusion removal with drones by pre-scanned point-cloud data. Depth information can be used to compute visibility of points on the ground for individual drone positions in the air. Inspired by Helmholtz reciprocity, we introduce reciprocal visibility to determine the dual situation - the visibility of potential sampling position in the air from given points of interest on the ground. The resulting visibility map encodes which point on the ground is visible by which magnitude from any position in the air. Based on such a map, we demonstrate a first greedy sampling optimization.

Finding sources or leaks of airborne material in Chemical, Biological, Radiological, or Nuclear (CBRN) accidents is crucial for effective disaster response. This paper makes use of sparse Bayesian learning (SBL) to cooperatively estimate source locations based on measurements by multiple robots or a sensor network. The SBL approach facilitates the identification of sparse source support, indirectly providing information about the number of sources and their locations. To achieve this, we introduce a novel method that includes a trained surrogated model for the gas dispersion process described by a Partial Differential Equation (PDE). Namely, a Physics-Guided Neural Network (PGNN) is employed to approximate a parameterized Green's function of the PDE. The obtained approximation is integrated into a gradient-based optimization process. The proposed method allows estimating super-resolution arbitrary source locations, eliminating constraints to a specific grid. Further, the newly proposed PGNN surrogate model comes with the advantage that the approach can be extended to cases where no analytic Green's function is available. Simulation results demonstrate the effectiveness of the proposed approach, showcasing its potential for enhanced airborne material detection in CBRN scenarios.

We present a GPU-accelerated, computationally efficient seismic imaging approach implemented in CUDA for use on NVIDIA's edge device, the Jetson Nano. The presented implementation in-tends to enable fast seismic imaging in autonomous exploration tasks on robotic platforms where electrical power and weight are limiting factors. For the imaging method we consider traveltime tomography, a seismic imaging technique based on first-arrival trav-eltimes. To this end, we rely on the fast iterative method that solves the eikonal equation in a parallel fashion and provides first-arrival traveltime maps. Furthermore, we employ a gradient-based ray tracer to reconstruct wave paths through the subsurface. We im-plement ray tracer and parts of the tomography update in a parallel fashion for efficient computation on the GPU of the Jetson Nano. We demonstrate the imaging capability of our implementation for synthetic as well as real seismic data. We also show that our edge device implementation achieves imaging results in a comparable time range as a state-of-the-art geophysical inversion tool running on a powerful desktop CPU.

We present a 3D seismic exploration and imaging survey conducted by robotic platforms in a hardware-in-the-loop system. To this end, we integrate the adapt-then-combine full waveform inversion (ATC-FWI) over a network of mobile rovers in the ROS2 framework. The ATC-FWI allows for distributed subsurface imaging in a multi-agent network, i.e., each rover obtains a 3D subsurface image via data exchange with other rovers in the network. We demonstrate the capability of our system by performing multiple surveys using a synthetic subsurface model with an anomaly. The rovers acquire seismic data over different measurement areas and perform distributed imaging to reconstruct the subsurface. We show that the rovers are able to image the anomaly and to enhance their subsurface image over multiple measurement stages in different areas.

Imaging the subsurface of extra-terrestrial bodies and planets has gained significant attention in recent space exploration missions. Multi-agent systems that autonomously perform subsurface imaging have been proposed. There, agents collaborate to image the subsurface by leveraging their wireless connections, enabling each agent to obtain an estimate of the subsurface image. However, traditional subsurface imaging techniques rely on a single entity for data collection and inversion, making them centralized schemes that limit direct availability of subsurface images at the agents. In this article, we propose the joint use of distributed seismic imaging techniques based on traveltime tomography and full waveform inversion, namely, the distributed traveltime tomography (D-TOMO) and the adapt-then-combine full waveform inversion (ATC-FWI). Combined in a sequential manner these techniques allow each agent to infer high-resolution subsurface images by exchanging data with their neighboring agents starting from a simple initial subsurface model. Unlike existing decentralized seismic imaging methods, our proposed scheme is fully distributed and provides flexibility without the need of anchor nodes or a full mesh topology. We demonstrate that ATC-FWI can recover high-frequency components based on a low-resolution subsurface image provided by D-TOMO in the initial stage. To assess the imaging performance, we employ a synthetic model, the SEG/EAGE salt model as well as real data from field measurements conducted over a tunnel.

The conservation of hydrological resources involves continuously monitoring their contamination. A multiagent system composed of autonomous surface vehicles is proposed herein to efficiently monitor the water quality. To achieve a safe control of the fleet, the fleet policy should be able to act based on measurements and fleet state. It is proposed to use local Gaussian processes and deep reinforcement learning to jointly obtain effective monitoring policies. Local Gaussian processes, unlike classical global Gaussian processes, can accurately model the information in a dissimilar spatial correlation which captures more accurately the water quality information. A deep convolutional policy is proposed, that bases the decisions on the observation on the mean and variance of this model, by means of an information gain reward. Using a double deep Q‐learning algorithm, agents are trained to minimize the estimation error in a safe manner thanks to a Consensus‐based heuristic. Simulation results indicate an improvement of up to 24% in terms of the mean absolute error with the proposed models. Also, training results with 1–3 agents indicate that our proposed approach returns 20% and 24% smaller average estimation errors for, respectively, monitoring water quality variables and monitoring algae blooms, as compared to state‐of‐the‐art approaches.