Abstract
Autonomous Underwater Vehicles (AUVs) are seen as a safe and cost-effective platforms for performing a myriad of underwater missions. These vehicles are equipped with multiple sensors which, combined with their long endurance, can produce large amounts of data, especially when used for video capturing. These data need to be transferred to the surface to be processed and analyzed. When considering deep sea operations, where surfacing before the end of the mission may be unpractical, the communication is limited to low bitrate acoustic communications, which make unfeasible the timely transmission of large amounts of data unfeasible. The usage of AUVs as data mules is an alternative communications solution. Data mules can be used to establish a broadband data link by combining short-range, high bitrate communications (e.g., RF and wireless optical) with a Delay Tolerant Network approach. This paper presents an enhanced version of UDMSim, a novel simulation platform for data muling communications. UDMSim is built upon a new realistic AUV Motion and Localization (AML) simulator and Network Simulator 3 (ns-3). It can simulate the position of the data mules, including localization errors, realistic position control adjustments, the received signal, the realistic throughput adjustments, and connection losses due to the fast SNR change observed underwater. The enhanced version includes a more realistic AML simulator and the antenna radiation patterns to help evaluating the design and relative placement of underwater antennas. The results obtained using UDMSim show a good match with the experimental results achieved using an underwater testbed. UDMSim is made available to the community to support easy and faster evaluation of underwater data muling oriented communications solutions and to enable offline replication of real world experiments.

Abstract
This paper describes a system designed to collect water samples, from the surface down to a configurable depth, and with configurable profiles of vertical velocity. The design was intended for the analysis of suspended sediments, therefore the sampling can integrate water flow for a given depth profile, or at a specific depth. The system is based on a catamaran-shaped platform, from which a towfish is lowered to collect the water samples. The use of a surface vehicle ensures a permanent link between the operator and the full system, allowing for a proper mission supervision. All components can be remotely controlled from the control station, or programmed for fully autonomous operation. Although the main intended use is for the analysis of suspended sediments in rivers, it can easily be extended to collect water samples in other water bodies.

Abstract

Abstract
In source localization problems, the relative geometry between sensors and source will influence the localization performance. The optimum configuration of sensors depends on the measurements used for the source location estimation, how these measurements are affected by noise, the positions of the source, and the criteria used to evaluate the localization performance. This paper addresses the problem of optimum sensor placement in a plane for the localization of an underwater vehicle moving in 3D. We consider sets of sensors that measure the distance to the vehicle and model the measurement noises with distance dependent covariances. We develop a genetic algorithm and analyze both single and multi-objective problems. In the former, we consider as the evaluation metric the arithmetic average along the vehicle trajectory of the maximum eigenvalue of the inverse of the Fisher information matrix. In the latter, we estimate the Pareto front of pairs of common criteria based on the Fisher information matrix and analyze the evolution of the sensor positioning for the different criteria. To validate the algorithm, we initially compare results with a case with a known optimal solution and constant measurement covariances, obtaining deviations from the optimal less than 0.1%. Posterior, we present results for an underwater vehicle performing a lawn-mower maneuver and a spiral descent maneuver. We also present results restricting the allowed positions for the sensors.

Abstract
Solar wind forecasting is a core component of Space Weather, a field that has been the target of many novel machine-learning approaches. The continuous monitoring of the Sun has provided an ever-growing ensemble of observations, facilitating the development of forecasting models that predict solar wind properties on Earth and other celestial objects within the solar system. This enables us to prepare for and mitigate the effects of solar wind-related events on Earth and space. The performance of some simulation-based solar wind models depends heavily on the quality of the initial guesses used as initial conditions. This work focuses on improving the accuracy of these initial conditions by employing a Recurrent Neural Network model. The study's findings confirmed that Recurrent Neural Networks can generate better initial guesses for the simulations, resulting in faster and more stable simulations. In our experiments, when we used predicted initial conditions, simulations ran an average of 1.08 times faster, with a statistically significant improvement and reduced amplitude transients. These results suggest that the improved initial conditions enhance the numerical robustness of the model and enable a more moderate integration time step. Despite the modest improvement in simulation convergence time, the Recurrent Neural Networks model's reusability without retraining remains valuable. With simulations lasting up to 12 h, an 8% gain equals one hour saved per simulation. Moreover, the generated profiles closely match the simulator's, making them suitable for applications with less demanding physical accuracy.