Neural Information Processing Systems http://nips.cc/

The proposed model allows for both arbitrary alignments of the inputs and non-parametric output warpings to transform the observations. This gives rise to multiple deep Gaussian process models connected via latent generating processes.

Remote monitoring of drones has become a global objective due to emerging applications in national security and managing aerial delivery traffic. Despite their relatively small size, drones can carry significant payloads, which require monitoring, especially in cases of unauthorized transportation of dangerous goods. A drone's flight dynamics heavily depend on outdoor wind conditions and the carry-on weight, which affect the tilt angle of a drone's body and the rotation velocity of the blades. A surveillance radar can capture both effects, provided a sufficient signal-to-noise ratio for the received echoes and an adjusted postprocessing detection algorithm. Here, we conduct a systematic study to demonstrate that micro-Doppler analysis enables the disentanglement of the impacts of wind and weight on a hovering drone. The physics behind the effect is related to the flight controller, as the way the drone counteracts weight and wind differs. When the payload is balanced, it imposes an additional load symmetrically on all four rotors, causing them to rotate faster, thereby generating a blade-related micro-Doppler shift at a higher frequency. However, the impact of the wind is different. The wind attempts to displace the drone, and to counteract this, the drone tilts to the side. As a result, the forward and rear rotors rotate at different velocities to maintain the tilt angle of the drone body relative to the airflow direction. This causes the splitting in the micro-Doppler spectra. By performing a set of experiments in a controlled environment, specifically, an anechoic chamber for electromagnetic isolation and a wind tunnel for imposing deterministic wind conditions, we demonstrate that both wind and payload details can be extracted using a simple deterministic algorithm based on branching in the micro-Doppler spectra.

WFCRL also offers turbine load observations that will allow to optimize the farm performance while limiting turbine structural damages.

High Altitude Balloons (HABs) can leverage stratospheric wind layers for limited horizontal control, enabling applications in reconnaissance, environmental monitoring, and communications networks. Existing multi-agent HAB coordination approaches use deterministic methods like Voronoi partitioning and extremum seeking control for large global constellations, which perform poorly for smaller teams and localized missions. While single-agent HAB control using reinforcement learning has been demonstrated on HABs, coordinated multi-agent reinforcement learning (MARL) has not yet been investigated. This work presents the first systematic application of multi-agent reinforcement learning (MARL) to HAB coordination for distributed area coverage. We extend our previously developed reinforcement learning simulation environment (RLHAB) to support cooperative multi-agent learning, enabling multiple agents to operate simultaneously in realistic atmospheric conditions. We adapt QMIX for HAB area coverage coordination, leveraging Centralized Training with Decentralized Execution to address atmospheric vehicle coordination challenges. Our approach employs specialized observation spaces providing individual state, environmental context, and teammate data, with hierarchical rewards prioritizing coverage while encouraging spatial distribution. We demonstrate that QMIX achieves similar performance to the theoretically optimal geometric deterministic method for distributed area coverage, validating the MARL approach and providing a foundation for more complex autonomous multi-HAB missions where deterministic methods become intractable.

Within wind farms, wake effects between turbines can significantly reduce overall energy production. Wind farm flow control encompasses methods designed to mitigate these effects through coordinated turbine control. Wake steering, for example, consists in intentionally misaligning certain turbines with the wind to optimize airflow and increase power output. However, designing a robust wake steering controller remains challenging, and existing machine learning approaches are limited to quasi-static wind conditions or small wind farms. This work presents a new deep reinforcement learning methodology to develop a wake steering policy that overcomes these limitations. Our approach introduces a novel architecture that combines graph attention networks and multi-head self-attention blocks, alongside a novel reward function and training strategy. The resulting model computes the yaw angles of each turbine, optimizing energy production in time-varying wind conditions. An empirical study conducted on steady-state, low-fidelity simulation, shows that our model requires approximately 10 times fewer training steps than a fully connected neural network and achieves more robust performance compared to a strong optimization baseline, increasing energy production by up to 14 %. To the best of our knowledge, this is the first deep reinforcement learning-based wake steering controller to generalize effectively across any time-varying wind conditions in a low-fidelity, steady-state numerical simulation setting.

Infrasound are permanently recorded by the International Monitoring System (IMS) set up to detect one kiloton equivalent nuclear explosion around the world (Marty et al. 2019 [2]) and monitor the compliance of the Comprehensive Nuclear T est-Ban-T reaty (CTBT). Accurate modeling of infrasound transmission loss (TL) is essential to interpret microbarometer measurements, evaluate their detection thresholds and characterise wavefield parameters (direction of arrival, velocities, amplitudes, frequencies) and source informations (ground pressure levels associated to earthquakes, acoustic energy from man-made or volcanic explosions). TLs modeling can also help to better characterise the middle atmosphere (MA, 15 100 km) which significantly impact the infrasound propagation. The computational cost of existing numerical propagation modeling tools, such as normal modes or full-waveform simulations (parabolic equations, PEs, Waxler et al. 2021 [3]), does not currently allow the exploration of a wide parameter space (variations in atmospheric states, representation of small-scale variability, frequency and source location) for near-real time TLs predictions; making them unusable within the required CTBT operational framework. Reducing these computation times by neglecting part of the complexity of the propagation phenomenon introduces significant uncertainties in predicted TLs. For example, Le Pichon et al. 2012 [4] proposed an approach relying on heuristic modelling of wave attenuation using a semi-analytical formula mapping wind speeds in the MA to TLs at ground level. However, this method has been optimized on idealized atmospheric models neglecting range-dependent variations in the atmosphere, resulting in large errors for unfavorable initial wind conditions. Artificial intelligence methods are currently explored by Brissaud et al. 2023 [1] in the Norwegian Seismic Array (NORSAR

Safe Periodic Trochoidal Paths for Fixed-Wing UA Vs in Confined Windy Environments Jaeyoung Lim 1, David Rohr 1, Thomas Stastny 1, Roland Siegwart 1 Abstract -- Due to their energy-efficient flight characteristics, fixed-wing type uncrewed aerial vehicles (UA Vs) are useful robotic tools for long-range and duration flight applications in large-scale environments. However, flying fixed-wing UA V in confined environments, such as mountainous regions, can be challenging due to their limited maneuverability and sensitivity to uncertain wind conditions. In this work, we first analyze periodic trochoidal paths that can be used to define wind-aware terminal loitering states. We then propose a wind-invariant safe set of trochoidal paths along with a switching strategy for selecting the corresponding minimum-extent periodic path type. Finally, we show that planning with this minimum-extent set allows us to safely reach up to 10 times more locations in mountainous terrain compared to planning with a single, conservative loitering maneuver . I. INTRODUCTION Uncrewed aerial vehicles (UA Vs) have become crucial tools for information-gathering applications, such as surveying and inspection [1], search and rescue [2], and environment monitoring [3], [4]. For large-scale coverage or long-range applications, fixed-wing type UA Vs are preferred over rotary-wing type systems due to their high endurance and speed. While the wing-borne aerodynamic lift enables energy-efficient flight, it also poses challenges for operating safely.

The wind farm control problem is challenging, since conventional model-based control strategies require tractable models of complex aerodynamical interactions between the turbines and suffer from the curse of dimension when the number of turbines increases. Recently, model-free and multi-agent reinforcement learning approaches have been used to address this challenge. In this article, we introduce WFCRL (Wind Farm Control with Reinforcement Learning), the first open suite of multi-agent reinforcement learning environments for the wind farm control problem. WFCRL frames a cooperative Multi-Agent Reinforcement Learning (MARL) problem: each turbine is an agent and can learn to adjust its yaw, pitch or torque to maximize the common objective (e.g. the total power production of the farm). WFCRL also offers turbine load observations that will allow to optimize the farm performance while limiting turbine structural damages. Interfaces with two state-of-the-art farm simulators are implemented in WFCRL: a static simulator (FLORIS) and a dynamic simulator (FAST.Farm). For each simulator, $10$ wind layouts are provided, including $5$ real wind farms. Two state-of-the-art online MARL algorithms are implemented to illustrate the scaling challenges. As learning online on FAST.Farm is highly time-consuming, WFCRL offers the possibility of designing transfer learning strategies from FLORIS to FAST.Farm.