The cooperation of multiple unmanned aerial vehicle (multi‐UAV) systems in adversarial and uncertain environments, particularly for multitarget pursuit, has gained significant attention. However, existing pursuit methods face challenges, such as limited adaptability, high communication demands, and reduced interpretability, which weaken their effectiveness in decentralized settings with low system redundancy. To address these challenges, this article proposes a self‐organized multitarget pursuit approach based on hierarchical probabilistic graphical models (PGM) for multi‐UAV systems. Specifically, a scalable hierarchical PGM structure is designed to facilitate a direct mapping from environmental perception to action decision‐making for individual UAVs. To estimate the probability distribution parameters of the designed PGM's nodes, a density‐driven probability parameter estimation method that relies on local perceptual information is proposed, which employs smoothed particle hydrodynamics for local density estimation. A series of simulations and real‐world experiments validates the scalability, adaptability, and robustness of the proposed approach in dynamic, obstacle‐cluttered environments.

Fluid Regulators In article number 2401069, Johannes T. B. Overvelde, Danqing Liu, and co-workers demonstrate a liquid crystal network (LCN)-based responsive fluid regulators incorporated in fluidic circuits that can be manipulated through multi-physical stimuli approach, where the authors showcase the regulating behavior by altering its fluidic resistance in response to temperature. The experimental work is verified by finite element modeling for the thermal actuation. Image credit: Charlotte Bording.

Deep Learning–Powered Robust Tactile Perception In article number 2400909, Suck Won Hong, Gyeonghwa Heo, Jusouk Yoon, and co-workers introduce the convergence of engineering, human touch, deep learning, and robotics through laser-induced graphene-based electronic skin (e-skin). As a human finger engages with the e-skin interface, the robotic hand interprets the input as tactile data, transforming human-robot interaction and enhancing precision in intelligent interfaces for medical device operation and beyond.

Synthetic microrobots have gained significant attention due to their potential in various applications in biomedicine and lab‐on‐a‐chip technologies. As a fundamental requirement, microrobots must navigate in 3D, effectively counteracting gravity to execute their tasks. However, locomotion at small scales presents numerous counterintuitive behaviors, primarily governed by the interactions between the microrobot's body and its surrounding boundaries. In this study, the locomotion of surface‐rolling microrobots is investigated in 3D, particularly focusing on their ability to climb walls. Through a combination of experiments and computational fluid dynamics analyzes, it is demonstrated that the influence of gravity plays a secondary role in enabling surface‐rolling microrobots to climb walls. Instead, locomotion capability in 3D settings is primarily determined by interactions with surrounding boundaries. The fundamental principles of surface‐rolling locomotion in 3D spaces is elucidated and a design strategy aimed at optimizing fluid flow for efficient propulsion in future applications is proposed.

Reservoir computing (RC) provides a route to use physical systems for computation and machine learning. Owing to their inherent nonlinearity, nanomechanical systems constitute an interesting technology to serve as reservoir. While RC platforms are built using microelectromechanical systems, the energy efficiency, response time, and footprint of these systems can be significantly improved by using nanoscale devices. Herein, the use of nanoelectromechanical systems (NEMS) is investigated, which can be used in RC, utilizing inherent nonlinearities and the fading memory effect from the transient response of NEMS. The smaller size and higher operating frequencies of NEMS enable faster processing rates compared to micromechanical systems, while their compact footprint, low power consumption, and ability to operate under ambient conditions simplify integration into practical applications. In modified national institute of standards and technology (MNIST) handwritten digit–recognition test, this system achieves 90% accuracy with a 3.3 μs processing time per pixel. Also the effect of driving frequency and amplitude on NEMS classification accuracy is investigated using experiments and simulations and it is shown that no significant dependency in any of the parameters is observed. Herein, an estimate for energy consumption of core NEMS RC system on MNIST data is provided. These results highlight the potential for various applications that require efficient and fast information processing in resource‐constrained environments.

Deep learning faces challenges like limited data, vanishing gradients, high parameter counts, and long training times. This article addresses two key issues: 1) data scarcity in ophthalmology and 2) vanishing gradients in deep networks. To overcome data limitations, an image processing‐based data generation method is proposed, expanding the dataset size by 12x. This approach enhances model training and prevents overfitting. For vanishing gradients, a deep neural network is introduced with optimized weight updates in initial layers, enabling the use of more and deeper layers. The proposed methods are validated using the retinal fundus multi‐disease image database dataset, a limited and imbalanced ophthalmology dataset available on the Grand Challenge website. Results show a 10% improvement in model accuracy compared to the original dataset and a 5% improvement over the benchmark reported on the website.

Medical ultrasound (US) has been widely used in clinical practices due to its merits of being low cost, real time, and radiation free. However, its capability to reveal the underlying tissue properties remains underexplored. A physics‐constrained learning framework is studied to reversely approximate tissue property representations from multiple B‐mode images acquired with varying dynamic ranges. First, an extractor network is used to generate property maps, that is, attenuation coefficient α, reflection coefficient β, border probability ρb, scattering density ρs, scattering amplitude ϕ, and one perturbation p map characterizing the variations caused by varying dynamic range. The α − ϕ maps are loosely regularized by rendering them forward to realistic US images using ray‐tracing simulator. To further enforce the physics constraints, a ranking loss is introduced to align the disparity between two estimated p maps with the discrepancy between two distinct inputs. Due to the missing ground truth α − ϕ maps, alternatively, the method is validated by evaluating the consistency between the feature maps inferred from distinct images. The results demonstrate that the proposed method can robustly extract consistent intermediate maps from images. Furthermore, one potential downstream application is showcased to perform realistic US augmentation by introducing specific noise into the physics‐inspired α − ϕ maps.

Matrix‐vector multiplications are a fundamental building block of artificial intelligence; this essential role has motivated their implementation in a variety of physical substrates, from memristor crossbar arrays to photonic‐integrated circuits. Yet their realization in soft‐matter intelligent systems remains elusive. Herein, A reprogrammable elastic metamaterial that computes matrix‐vector multiplications using floppy modes—deformations with near‐zero stored elastic energy is experimentally demonstrated. Floppy modes allow to program complex deformations without being hindered by the natural stiffness of the material; but their practical application is challenging, as their existence depends on global topological properties of the system. To overcome this challenge, a continuously parameterized unit cell design with well‐defined compatibility characteristics is introduced. This unit cell is then combined to form arbitrary matrix‐vector multiplications that can even be reprogrammed after fabrication. The results demonstrate that floppy modes can act as key enablers for embodied intelligence, smart micro electro mechanical systems(MEMS) devices, and in‐sensor edge computing.

Sensor arrays are ubiquitous. They capture images in digital cameras, record the swipes of the fingers on the screens of the phones and tablets, or map pressure distribution over an area. Soft capacitive sensor arrays have been proposed to make electronic pressure‐sensing skins capable of identifying the location and intensity of touch. However, large arrays of those sensors remain challenging to produce, as they require high‐resolution patterning of electrodes and routing of long and thin electrical connections. These two tasks remain difficult or costly for the high‐resistivity compliant electrodes of soft and stretchy capacitive sensors. Instead of relying on the complex patterning of arrays to provide location resolution, a plain, unstructured sensor with a single pair of electrodes is used and relied on computation to infer pressure location and amplitude from multifrequency sensing signals. Herein, a machine learning–based approach, which enables us to identify pressure location on a continuous 1D sensor split into 5 sensing zones with an accuracy greater than 97%, is proposed. A regression algorithm calculates the force amplitude with a mean absolute error of 5.2%.