Papers

Recent advances in real-time interactive text-driven motion generation have enabled humanoids to perform diverse behaviors. However, kinematics-only generators often exhibit physical hallucinations, producing motion trajectories that are physically infeasible to track with a downstream motion tracking controller or unsafe for real-world deployment. These failures often arise from the lack of explicit physics-aware objectives for real-robot execution and become more severe under out-of-distribution (OOD) user inputs. Hence, we propose SafeFlow, a text-driven humanoid whole-body control framework that combines physics-guided motion generation with a 3-Stage Safety Gate driven by explicit risk indicators. SafeFlow adopts a two-level architecture. At the high level, we generate motion trajectories using Physics-Guided Rectified Flow Matching in a VAE latent space to improve real-robot executability, and further accelerate sampling via Reflow to reduce the number of function evaluations (NFE) for real-time control. The 3-Stage Safety Gate enables selective execution by detecting semantic OOD prompts using a Mahalanobis score in text-embedding space, filtering unstable generations via a directional sensitivity discrepancy metric, and enforcing final hard kinematic constraints such as joint and velocity limits before passing the generated trajectory to a low-level motion tracking controller. Extensive experiments on the Unitree G1 demonstrate that SafeFlow outperforms diffusion- and retargeting-based baselines in success rate, physical compliance, and inference speed while preserving motion diversity, with consistent gains across three downstream tracking controllers.

This paper presents an open-source Software-in-the-Loop (SIL) simulation platform designed for autonomous Ackerman vehicle research and education. The proposed framework focuses on simplicity, while making it easy to work with small-scale experimental setups, such as the XTENTH-CAR platform. The system was designed using open source tools, creating an environment with a monocular camera vision system to capture stimuli from it with minimal computational overhead through a sliding window based lane detection method. The platform supports a flexible algorithm testing and validation environment, allowing researchers to implement and compare various control strategies within an easy-to-use virtual environment. To validate the working of the platform, Model Predictive Control (MPC) and Proportional-Integral-Derivative (PID) algorithms were implemented within the SIL framework. The results confirm that the platform provides a reliable environment for algorithm verification, making it an ideal tool for future multi-agent system research, educational purposes, and low-cost AGV development. Our code is available at https://github.com/shantanu404/monosim.git.

Single-snapshot FDA-MIMO-GPR requires clutter models that account for dispersive-medium uncertainty, yet the statistical link between complex-medium characterization and clutter covariance analysis has remained unclear. This paper develops a propagation-side statistical framework that maps random perturbations of the relaxation spectrum to complex permittivity, complex wavenumber, steering-vector perturbation, medium-induced clutter covariance, and total clutter covariance. Within this framework, the effects of medium uncertainty on effective rank, effective clutter-subspace dimension, and target–clutter separability are characterized through a KL-based modal decomposition and a subspace-projection analysis. Numerical validation uses five literature-informed dielectric families to define physically traceable prior scenarios, a controlled random-field model to exercise the main propagation chain, and gprMax-based full-wave FDTD snapshots for an independent solver-level consistency check. Monte Carlo closure shows stage-wise numerical consistency, identifies steering linearization as the dominant approximation-sensitive step, and supports a weak perturbation regime with a bounded extension into a moderate regime. In a representative whitening-and-detection benchmark, the structured covariance model raises AUC from 0.593 for a diagonal baseline to 0.753, while prior-mismatch experiments indicate gradual rather than abrupt degradation. These results provide an explicit and interpretable interface for embedding complex-medium uncertainty into FDA-MIMO-GPR clutter analysis within a first-order, propagation-dominated setting.

Collaboration is a central theme in multi-robot systems as tasks and demands increasingly require capabilities that go beyond what any one individual robot possesses. Yet, despite extensive work on cooperative control and coordinated behaviors, the terminology surrounding collective multi-robot interaction remains inconsistent across research communities. In particular, cooperation, coordination, and collaboration are often treated interchangeably, without clearly articulating the differences among them. To address this gap, we propose definitions that distinguish and relate cooperation, coordination, and collaboration in multi-robot systems, highlighting the support of new capabilities in collaborative behaviors, and illustrate these concepts through representative examples. Building on this taxonomy, different frameworks for collaboration are reviewed, and technical challenges and promising future research directions are identified for collaborative multi-robot systems.

Semantic communication has emerged as a promising paradigm for improving transmission efficiency and task-level reliability, yet most existing reliability-enhancement approaches rely on retransmission strategies driven by semantic fidelity checking that require additional check codewords solely for retransmission triggering, thereby incurring substantial communication overhead. In this paper, we propose S3CHARQ, a Joint Source-Channel-Check Coding framework with hybrid automatic repeat request that fundamentally rethinks the role of check codewords in semantic communications. By integrating the check codeword into the JSCC process, S3CHARQ enables JS3C, allowing the check codeword to simultaneously support semantic fidelity verification and reconstruction enhancement. At the transmitter, a semantic fidelity-aware check encoder embeds auxiliary reconstruction information into the check codeword. At the receiver, the JSCC and check codewords are jointly decoded by a JS3C decoder, while the check codeword is additionally exploited for perceptual quality estimation. Moreover, because retransmission decisions are necessarily based on imperfect semantic quality estimation in the absence of ground-truth reconstruction, estimation errors are unavoidable and fundamentally limit the effectiveness of rule-based decision schemes. To overcome this limitation, we develop a reinforcement learning-based retransmission decision module that enables adaptive, sample-level retransmission decisions, effectively balancing recovery and refinement information under dynamic channel conditions. Experimental results demonstrate that compared with existing HARQ-based semantic communication systems, the proposed S3CHARQ framework achieves a 2.36 dB improvement in the 97th percentile PSNR, as well as a 37.45% reduction in outage probability.

Analog beamforming holds great potential for future terahertz (THz) communications due to its ability to generate high-gain directional beams with low-cost phase shifters. However, conventional analog beamforming may suffer substantial performance degradation in wideband systems due to the beam squint effect. Instead of relying on high-cost true-time delayers, we propose an efficient six-dimensional movable antenna (6DMA) architecture to mitigate the beam-squint effect. In particular, we study a wideband wide-beam coverage problem in this paper, aiming to maximize the minimum beamforming gain over a given range of azimuth/elevation angles and frequencies by jointly optimizing the analog beamforming vector, the MA positions within a two-dimensional (2D) region, and the three-dimensional (3D) rotation angles of the antenna array. However, this problem is non-convex and intractable to solve optimally due to the coupling of the spatial and frequency domains and that of the antenna weights, positions and rotation. To tackle this problem, we first derive an optimal solution to it in a special case with azimuth or elevation angle coverage only. It is shown that rotating a uniform linear array (ULA) is sufficient to achieve global optimality and eliminate beam-squint effects. While for other general cases, an alternating optimization (AO) algorithm is proposed to obtain a high-quality suboptimal solution, where the antennas' beamforming weights, positions, and rotation angles are alternately optimized by combining successive convex approximation (SCA), sequential update with Gibbs sampling (GS), and hybrid coarse- and fine-grained search. Simulation results demonstrate that our proposed scheme can significantly outperform conventional antenna arrays without antenna movement or rotation, thus offering a cost-effective solution for wideband transmission over THz bands.

Active incoherent millimeter-wave (AIM) imaging is a recently developed technique that has been shown to generate fast millimeter-wave imaging using sparse apertures and Fourier domain sampling. In these systems, spatial frequency sampling is determined by cross-correlation between antenna pairs, making array geometry an important aspect that dictates the field of view (FOV) and image quality. This work investigates the impact of array redundancy and spatial sampling diversity on AIM image reconstruction performance. We present a comparative study of three receive array configurations, including one simple circular design and two arrays obtained through optimization strategies designed to maximize unique spatial samples while preserving system resolution and FOV. Performance is evaluated using the image-domain metrics of structural similarity index (SSIM) and peak sidelobe level (PSL), enabling a quantitative assessment of reconstruction fidelity and artifact suppression. We perform experimental validation using a 38-GHz AIM imaging system, implementing a 24-element receive array within a 48-position reconfigurable aperture. Results demonstrate that optimized array configurations improve spatial sampling efficiency and yield measurable gains in reconstruction quality compared to a conventional circular array, highlighting the importance of array design for AIM imaging systems.

Deep neural networks (DNNs), particularly those using Rectified Linear Unit (ReLU) activation functions, have achieved remarkable success across diverse machine learning tasks, including image recognition, audio processing, and language modeling. Despite this success, the non-convex nature of DNN loss functions complicates optimization and limits theoretical understanding. In this paper, we highlight how recently developed convex equivalences of ReLU NNs and their connections to sparse signal processing models can address the challenges of training and understanding NNs. Recent research has uncovered several hidden convexities in the loss landscapes of certain NN architectures, notably two-layer ReLU networks and other deeper or varied architectures. This paper seeks to provide an accessible and educational overview that bridges recent advances in the mathematics of deep learning with traditional signal processing, encouraging broader signal processing applications.

The fading-memory (FM) property captures the progressive loss of influence of past inputs on a system's current output and has originally been formalized by Boyd and Chua in an operator-theoretic framework. Despite its importance for systems approximation, reservoir computing, and recurrent neural networks, its connection with state-space notions of nonlinear stability, especially incremental ones, remains understudied. This paper introduces a state-space definition of FM. In state-space, FM can be interpreted as an extension of incremental input-to-output stability ($\delta$IOS) that explicitly incorporates a memory kernel upper-bounding the decay of past input differences. It is also closely related to Boyd and Chua's FM definition, with the sole difference of requiring uniform, instead of general, continuity of the memory functional with respect to an input-fading norm. We demonstrate that incremental input-to-state stability ($\delta$ISS) implies FM semi-globally for time-invariant systems under an equibounded input assumption. Notably, Boyd and Chua's approximation theorems apply to $\delta$ISS state-space models. As a closing application, we show that, under mild assumptions, the state-space model of a current-driven memristive device possesses the FM property.

Since the introduction of Masked Autoencoders, various improvements to masking techniques have been explored. In this paper, we rethink masking strategies for audio representation learning using masked prediction-based self-supervised learning (SSL) on general audio spectrograms. While recent informed masking techniques have attracted attention, we observe that they incur substantial computational overhead. Motivated by this observation, we propose dispersion-weighted masking (DWM), a lightweight masking strategy that leverages the spectral sparsity inherent in the frequency structure of audio content. Our experiments show that inverse block masking, commonly used in recent SSL frameworks, improves audio event understanding performance while introducing a trade-off in generalization. The proposed DWM alleviates these limitations and computational complexity, leading to consistent performance improvements. This work provides practical guidance on masking strategy design for masked prediction-based audio representation learning.

In this paper, we present DROP, high-Density Relocation-free sequential OPerations in automated valet parking. DROP addresses the challenges in high-density parking & vehicle retrieval without relocations. Each challenge is handled by jointly providing area-efficient layouts and relocation-free parking & exit sequences, considering accessibility with relocation-free sequential operations. To generate such sequences, relocation-free constraints are formulated as explicit logical conditions expressed in boolean variables. Recursive search strategies are employed to derive the logical conditions and enumerate relocation-free sequences under sequential constraints. We demonstrate the effectiveness of our framework through extensive simulations, showing its potential to significantly improve area utilization with relocation-free constraints. We also examine its viability on an application problem with prescribed operational order. The results from all experiments are available at: https://drop-park.github.io.

Generative AI is envisioned to have a crucial impact on next generation mobile networking, making the sixth generation (6G) system considerably more autonomous, flexible, and adaptive than its predecessors. By leveraging their natural language processing and code generation capabilities, AI agents enable novel interactions and services between networks and vertical applications. A particularly promising and interesting use case is the customization of connectivity services for vertical applications by generating new customized processing blocks based on text-based service requests. More specifically, AI agents are able to generate code for a new function block that handles user plane traffic, allowing it to inspect and decode a protocol data unit (PDU) and perform specified actions as requested by the application. In this study, we investigate the code generation problem for generating such customized processing blocks on-demand. We evaluate various factors affecting the accuracy of the code generation process in this context, including model selection, prompt design, and the provision of a code template for the agent to utilize. Our findings indicate that AI agents are capable of generating such blocks with the desired behavior on-demand under suitable conditions. We believe that exploring the code generation for network-specific tasks is a very interesting problem for 6G and beyond, enabling networks to achieve a new level of customization by generating new capabilities on-demand.

Text-to-Speech (TTS) models are significantly more numerically fragile than Large Language Models (LLMs) due to their continuous waveform generation and perceptual sensitivity to small numerical perturbations. While aggressive precision reduction techniques such as BlockFloat8 (BFP8) and low-fidelity (LoFi) compute have been widely adopted in language models, applying similar strategies to TTS systems often results in audible artifacts, phase instability, and spectral distortion. In this work, we present Lightning V2, a production-grade TTS model co-optimized for Tenstorrent hardware. Through precision-aware architectural design and hardware-software co-optimization, we achieve over 95% LoFi computational fidelity and more than 80% BlockFloat8 deployment without measurable degradation in audio quality. Leveraging Tenstorrent's Network-on-Chip (NoC), distributed SRAM, and deterministic execution model, we reduce memory movement and redundant weight fetches, enabling efficient low-precision inference. Compared to an NVIDIA L40S baseline, Lightning V2 achieves approximately 4x lower on-prem accelerator cost at equivalent throughput, while maintaining production audio fidelity. Our results demonstrate that precision co-design, combined with hardware-aware optimization, can fundamentally reshape the economics of real-time speech inference.

As power systems advance toward net-zero targets, behind-the-meter renewables are driving rapid growth in distributed energy resources (DERs). Virtual power plants (VPPs) increasingly coordinate these resources to support power distribution network (PDN) operation, with EV charging stations (EVCSs) emerging as a key asset due to their strong impact on local voltages. However, in practice, VPPs must make operational decisions with only partial visibility of PDN states, relying on limited, aggregated information shared by the distribution system operator. This work proposes a safety-enhanced VPP framework for coordinating multiple EVCSs under such realistic information constraints to ensure voltage security while maintaining economic operation. We develop Transformer-assisted Lagrangian Multi-Agent Proximal Policy Optimization (TL-MAPPO), in which EVCS agents learn decentralized charging policies via centralized training with Lagrangian regularization to enforce voltage and demand-satisfaction constraints. A transformer-based embedding layer deployed on each EVCS agent captures temporal correlations among prices, loads, and charging demand to improve decision quality. Experiments on a realistic 33-bus PDN show that the proposed framework reduces voltage violations by approximately 45% and operational costs by approximately 10% compared to representative multi-agent DRL baselines, highlighting its potential for practical VPP deployment.

The fast motion of Low Earth Orbit (LEO) satellites causes the propagation channel to vary rapidly, and its behavior is strongly shaped by the surrounding environment, especially at low elevation angles where signals are highly susceptible to terrain blockage and other environmental effects. Existing studies mostly rely on assumed statistical channel distributions and therefore ignore the influence of the actual geographic environment. In this paper, we propose an environment-aware channel modeling method for air-to-ground wireless links. We leverage real environmental data, including digital elevation models (DEMs) and land cover information, together with ray tracing (RT) to determine whether a link is line-of-sight (LOS) or non-line-of-sight (NLOS) and to identify possible reflection paths of the signal. The resulting obstruction and reflection profiles are then combined with models of diffraction loss, vegetation absorption, and atmospheric attenuation to quantitatively characterize channel behavior in realistic geographic environments. Since RT is computationally intensive, we use RT-generated samples and environmental features to train a scalable diffusion model that can efficiently predict channel performance for arbitrary satellite and ground terminal positions, thereby supporting real-time decision-making. In the experiments, we validate the proposed model with measurement data from both cellular and LEO satellite links, demonstrating its effectiveness in realistic environments.

Accurate and timely crop yield estimation is critical for global food security, agricultural policy, and farm management. The Copernicus Sentinel-2 satellite constellation, with high spatial, temporal, and spectral resolution, has transformed agricultural monitoring by enabling field- and sub-field-scale analysis. This review synthesizes recent advances in Sentinel-2-based crop yield estimation. A key trend is the shift from regional models to high-resolution field-level assessments driven by three main approaches: (i) empirical models using vegetation indices combined with machine and deep learning methods such as Random Forest and Convolutional Neural Networks; (ii) integration of process-based crop growth models (e.g., WOFOST, SAFY) via data assimilation of Sentinel-2-derived variables like Leaf Area Index (LAI); and (iii) data fusion techniques combining Sentinel-2 optical data with Sentinel-1 SAR to mitigate cloud-related limitations. The review shows that machine learning, deep learning, and hybrid modeling frameworks can explain substantial within-field yield variability across crops and regions. However, performance remains constrained by limited ground-truth data, cloud-induced gaps, and challenges in model transferability across years and locations. Future directions include tighter integration of multi-modal data and improved in-season observations to support robust, operational decision-making in precision agriculture and sustainable intensification.

During human motor skill training and physical rehabilitation, there is an inherent trade-off between task difficulty and user performance. Characterizing this trade-off is crucial for evaluating user performance, designing assist-as-needed (AAN) protocols, and assessing the efficacy of training protocols. In this study, we propose a novel human-in-the-loop (HiL) Pareto optimization approach to characterize the trade-off between task performance and the perceived challenge level of motor learning or rehabilitation tasks. We adapt Bayesian multi-criteria optimization to systematically and efficiently perform HiL Pareto characterizations. Our HiL optimization employs a hybrid model that measures performance with a quantitative metric, while the perceived challenge level is captured with a qualitative metric. We demonstrate the feasibility of the proposed HiL Pareto characterization through a user study. Furthermore, we present the utility of the framework through three use cases in the context of a manual skill training task with haptic feedback. First, we demonstrate how the characterized trade-off can be used to design a sample AAN training protocol for a motor learning task and to evaluate the group-level efficacy of the proposed AAN protocol relative to a baseline adaptive assistance protocol. Second, we demonstrate that individual-level comparisons of the trade-offs characterized before and after the training session enable fair evaluation of training progress under different assistance levels. This evaluation method is more general than standard performance evaluations, as it can provide insights even when users cannot perform the task without assistance. Third, we show that the characterized trade-offs also enable fair performance comparisons among different users, as they capture the best possible performance of each user under all feasible assistance levels.

District heating systems (DHSs) require coordinated economic dispatch and temperature regulation under uncertain operating conditions. Existing DHS operation strategies often rely on disturbance forecasts and nominal models, so their economic and thermal performance may degrade when predictive information or model knowledge is inaccurate. This paper develops a data-driven online control framework for DHS operation by embedding steady-state economic optimality conditions into the temperature dynamics, so that the closed-loop system converges to the economically optimal operating point without relying on disturbance forecasts. Based on this formulation, we develop a Data-Enabled Policy Optimization (DeePO)-based online learning controller and incorporate Adaptive Moment Estimation (ADAM) to improve closed-loop performance. We further establish convergence and performance guarantees for the resulting closed-loop system. Simulations on an industrial-park DHS in Northern China show that the proposed method achieves stable near-optimal operation and strong empirical robustness to both static and time-varying model mismatch under practical disturbance conditions.

Underwater observatories have recently emerged as an efficient solution for marine biodiversity monitoring. The primary objective of this work is to enable efficient and cost-effective data muling from underwater sensors by investigating the use of optical wireless communications to transmit data from the underwater sensors to an aerial node close to the water surface, such as an unmanned aerial vehicle (UAV). More specifically, we utilize a direct water-to-air (W2A) optical communication link between the sensor node equipped with an LED emitter and the UAV equipped with an ultra-sensitive receiver, i.e., a silicon photo-multiplier. As a main contribution, we develop a comprehensive Monte Carlo-based ray-tracing algorithm to characterize this complex channel. This framework rigorously incorporates the impact of air bubbles modeled through the Mie scattering theory, a realistic sea surface representation derived from the JONSWAP spectrum, and an analytical derivation of the channel loss resulting from UAV instability under wind-induced perturbations. Furthermore, we conduct a comprehensive analysis of the W2A channel, examining the influence of key parameters such as wind speed, transmitter configurations, and receiver characteristics. The end-to-end performance evaluation demonstrates the practical feasibility of the proposed approach, achieving a bit-error rate of $10^{-3}$ at a data rate of 1 Mbps for a transmitter depth of 47 m and wind speeds up to 13 m/s.

Deep spatially selective filters achieve high-quality enhancement with real-time capable architectures for stationary speakers of known directions. To retain this level of performance in dynamic scenarios where only the speakers' initial directions are given, accurate, yet computationally lightweight tracking algorithms become necessary. Assuming a frame-wise causal processing style, temporal feedback allows for leveraging the enhanced speech signal to improve tracking performance. In this work, we investigate strategies to incorporate the enhanced signal into lightweight tracking algorithms and autoregressively guide deep spatial filters. Our proposed Bayesian tracking algorithms are compatible with arbitrary deep spatial filters. To increase the realism of simulated trajectories during development and evaluation, we develop a synthetic data generation framework based on the social force model. Results validate that the autoregressive incorporation significantly improves the accuracy of our Bayesian trackers, resulting in superior enhancement with none or only negligibly increased computational overhead. Real-world recordings complement these findings and demonstrate the generalizability of our methods to unseen acoustic conditions.

Speech Emotion Recognition (SER) in real-world scenarios remains challenging due to severe class imbalance and the prevalence of spontaneous, natural speech. While recent approaches leverage self-supervised learning (SSL) representations and multimodal fusion of speech and text, most existing methods apply supervision only at the final classification layer, limiting the discriminative power of intermediate representations. In this work, we propose Crab (Contrastive Representation and Multimodal Aligned Bottleneck), a bimodal Cross-Modal Transformer architecture that integrates speech representations from WavLM and textual representations from RoBERTa, together with a novel Multi Layer Contrastive Supervision (MLCS) strategy. MLCS injects multi-positive contrastive learning signals at multiple layers of the network, encouraging emotionally discriminative representations throughout the model without introducing additional parameters at inference time. To further address data imbalance, we adopt weighted cross-entropy during training. We evaluate the proposed approach on three benchmark datasets covering different degrees of emotional naturalness: IEMOCAP, MELD, and MSP-Podcast 2.0. Experimental results demonstrate that Crab consistently outperforms strong unimodal and multimodal baselines across all datasets, with particularly large gains under naturalistic and highly imbalanced conditions. These findings highlight the effectiveness of Multi Layer Contrastive Supervision as a general and robust strategy for SER. Official implementation can be found in https://github.com/AI-Unicamp/Crab.

We introduce Echoes, a new dataset for music deepfake detection designed for training and benchmarking detectors under realistic and provider-diverse conditions. Echoes comprises 4,468 tracks (131 hours of audio) spanning multiple genres (pop, rock, electronic), and includes content generated by ten popular AI music generation systems. To prevent shortcut learning and promote robust generalization, the dataset is deliberately constructed to be challenging, enforcing semantic-level alignment between spoofed audio and bona fide references. This alignment is achieved by conditioning generated audio samples directly on bona-fide waveforms or song descriptors. We evaluate Echoes in a cross-dataset setting against three existing AI-generated music datasets using state-of-the-art Wav2Vec2 XLS-R 2B representations. Results show that (i) Echoes is the hardest in-domain dataset; (ii) detectors trained on existing datasets transfer poorly to Echoes; (iii) training on Echoes yields the strongest generalization performance. These findings suggest that provider diversity and semantic alignment help learn more transferable detection cues.

We address target interception in contested environments in the presence of multiple defenders whose interception capability is limited by finite ranges. Conventional methods typically impose conservative stand-off constraints based on maximum engagement distance and neglect the interceptors' actuator limitations. Instead, we formulate safety constraints using defender-induced engagement zones. To account for actuator limits, the vehicle model is augmented with input saturation dynamics. A time-varying safe-set tightening parameter is introduced to compensate for transient constraint violations induced by actuator dynamics. To ensure scalable safety enforcement in multi-defender scenarios, a smooth aggregate safety function is constructed using a log-sum-exp operator combining individual threat measures associated with each defender's capability. A smooth switching guidance strategy is then developed to coordinate interception and safety objectives. The attacker pursues the target when sufficiently distant from threat boundaries and progressively activates evasive motion as the EZ boundaries are approached. The resulting controller relies only on relative measurements and does not require knowledge of defender control inputs, thus facilitating a fully distributed and scalable implementation. Rigorous analysis provides sufficient conditions guaranteeing target interception, practical safety with respect to all defender engagement zones, and satisfaction of actuator bounds. An input-constrained guidance law based on conservative stand-off distance is also developed to quantify the conservatism of maximum-range-based safety formulations. Simulations with stationary and maneuvering defenders demonstrate that the proposed formulation yields shorter interception paths and reduced interception time compared with conventional methods while maintaining safety throughout the engagement.

Adversarial training is a defense method that trains machine learning models on intentionally perturbed attack inputs, so they learn to be robust against adversarial examples. This paper develops a robust voltage control framework for distribution networks with high penetration of distributed energy resources (DERs). Conventional voltage control methods are vulnerable to strategic cyber attacks, as they typically consider only random or black-box perturbations. To address this, we formulate white-box adversarial attacks using Projected Gradient Descent (PGD) and train a deep reinforcement learning (DRL) agent adversarially. The resulting policy adapts in real time to high-impact, strategically optimized perturbations. Simulations on DER-rich networks show that the approach maintains voltage stability and operational efficiency under realistic attack scenarios, highlighting the effectiveness of gradient-based adversarial DRL in enhancing robustness and adaptability in modern distribution system control.

Existing cell-free integrated sensing and communication (CF-ISAC) beamforming algorithms predominantly rely on classical optimization techniques, which often entail high computational complexity and limited scalability. Meanwhile, recent learning-based approaches have difficulty capturing the global interactions and long-range dependencies among distributed access points (APs), communication users, and sensing targets. To address these limitations, we propose the first Set Transformer-based CF-ISAC beamforming framework (STCIB). By exploiting attention mechanisms, STCIB explicitly models global relationships among network entities, naturally handles unordered input sets, and preserves permutation invariance across APs, users, and targets. The proposed framework operates in an unsupervised manner, eliminating the need for labeled training data, and supports three design regimes: (i) sensing-centric, (ii) communication-centric, and (iii) joint ISAC optimization. We benchmark STCIB against a convolutional neural network (CNN) baseline and two state-of-the-art optimization algorithms: the convex-concave procedure algorithm (CCPA) and augmented Lagrangian manifold optimization (ALM-MO). Numerical results demonstrate that STCIB consistently outperforms the CNN, achieving substantially higher ISAC performance with only a negligible increase in runtime. For instance, in regime (iii), at $\eta$=0.4, STCIB improves the sensing and communication sum rates by 14.8 % and 31.6 %, respectively, relative to the CNN, while increasing runtime by only 0.26 %. Compared with CCPA and ALM-MO, STCIB offers significantly lower computational cost while maintaining modest performance gains. In regime (i), for a 3.0 bps/Hz communication threshold, the runtime of STCIB is only 0.1 % and 0.3 % of that required by CCPA and ALM-MO, respectively, while improving the sensing sum rate by 4.45 % and 5.9 %.

We consider a quantum switch with a finite number of quantum memory registers that aims to serve multipartite entanglement requests among $N$ users. We propose scheduling policies that aim to optimize the average number of requests served per unit time by efficiently utilizing the switch's available memory. To measure the performance of the scheduling policies, we employ the newly introduced metric of age of entanglement establishment (AoEE). We formulate the scheduling problem in a restless multi-armed bandit (RMAB) framework. We show that the scheduling of entanglement requests is indexable. Subsequently, we find a closed-form expression of the Whittle index for all possible request-age pairs. By modeling the Whittle index of each request as its reward and its cardinality as its cost, we formulate the memory-constrained scheduling problem as a $0$-$1$ knapsack problem and solve it via dynamic programming. Furthermore, we consider two low-complexity sequential greedy policies that leverage two different modified Whittle indices.

Transcranial ultrasound stimulation (TUS) offers non-invasive deep-brain neuromodulation with high spatial precision, but reliably generating complex multi-target acoustic fields through the skull remains challenging. Here, we introduce a physics-aware hologram technique that directly generates fabrication-ready holographic implementations while preserving consistency between numerical field synthesis and physical acoustic realization. The method enables single-, dual-, and tri-focal transcranial stimulation patterns and was validated through in silico simulations, ex vivo skull measurements, and in vivo experiments. Compared with representative state-of-the-art methods, the proposed approach improved focal reconstruction and energy confinement at intended targets while reducing off-target acoustic leakage. In a neuropathic-pain mouse model, simultaneous bilateral stimulation of thalamic nuclei reduced c-Fos expression and showed preliminary improvements in pain-related behavioral responses. These findings support the use of fabrication-consistent holographic design for spatially localized and reproducible multi-target transcranial neuromodulation.

Compounding error, where small prediction mistakes accumulate over time, presents a major challenge in learning-based control. A common remedy is to train multi-step predictors directly instead of rolling out single-step models. However, it is unclear when the benefits of multi-step predictors outweigh the difficulty of learning a more complex model. We provide the first quantitative analysis of this trade-off for linear dynamical systems. We study three predictor classes: (i) single step models, (ii) multi-step models, and (iii) single step models trained with multi-step losses. We show that when the model class is well-specified and accurately captures the system dynamics, single-step models achieve the lowest asymptotic prediction error. On the other hand, when the model class is misspecified due to partial observability, direct multi-step predictors can significantly reduce bias and improve accuracy. We provide theoretical and empirical evidence that these trade-offs persist when predictors are used in closed-loop control.

This work studies the synthesis of active perception policies for predictive safety monitoring in partially observable stochastic systems. Operating under strict sensing and communication budgets, the proposed monitor dynamically schedules sensor queries to maximize information gain about the safety of future states. The underlying stochastic dynamics are captured by a labeled hidden Markov model (HMM), with safety requirements defined by a deterministic finite automaton (DFA). To enable active information acquisition, we introduce minimizing k-step Shannon conditional entropy of the safety of future states as a planning objective, under the constraint of a limited sensor query budget. Using observable operators, we derive an efficient algorithm to compute the k-step conditional entropy and analyze key properties of the conditional entropy gradient with respect to policy parameters. We validate the effectiveness of the method for predictive safety monitoring through a dynamic congestion game example.

Changing the transmission system topology is an efficient and costless lever to reduce congestion or increase exchange capacities. The problem of finding the optimal switch states within substations is called Optimal Substation Reconfiguration (OSR), and may be framed as a Mixed Integer Linear Program (MILP). Current state-of-the-art optimization techniques come with prohibitive computing times, making them impractical for real-time decision-making. Meanwhile, deep learning offers a promising perspective with drastically smaller computing times, at the price of an expensive training phase and the absence of optimality guarantees. In this work, we frame OSR as an Amortized Optimization problem, where a Graph Neural Network (GNN) model – our data being graphs – is trained in a self-supervised way to improve the objective function. We apply our approach to the maximization of the exchange capacity between two areas of a small-scale 12-substations system. Once trained, our GNN model improves the exchange capacity by 10.2% on average compared to the all connected configuration, while a classical MILP solver reaches an average improvement of 15.2% with orders-of-magnitude larger computing times.