Various marine animals possess the ability to track their preys and navigate dark aquatic environments using hydrodynamic sensing of the surrounding flow. In the present study, a deep-learning model is applied to a biomimetic sensor for underwater position detection of a wake-generating body. The sensor is composed of a bundle of spatially-distributed optical fibers that act as artificial seal-like whiskers and interact with the body’s wake in the form of time-variant (bending) deflections. Supervised learning is employed to relate the vibrations of the artificial whiskers to the position of an upstream cylinder. The labeled training data are prepared based on the processing and reduction of the recorded bending responses of the artificial whiskers while the cylinder is placed at various locations. An iterative training algorithm is performed on two neural-network models while using the 10-fold cross-validation technique. The models are able to predict the coordinates of the cylinder in the two-dimensional (2D) space with a high degree of accuracy. The current implementation of the sensor can passively sense the wake generated by the cylinder at Re ≃ 6000 and estimate its position with an average error smaller than the characteristic diameter D of the cylinder and for inter-distances (in the water tunnel) up to 25-times D.

Like many other surveillance systems, security and safety are mainly what concern people these days. It can be used in two methods, watching threats in real-time or searching for an event that already happened. Surveillance systems can be defined in many different ways like violence or any abnormal behaviour detection. In this paper, we propose a surveillance system for detecting abnormal behaviours on campus using deep learning models. It also includes how YOLO, CNN, and LSTM are used to apply concepts like feature extraction, object detection, action detection and identification. YOLO algorithm was used in the smoking detection and the playing cards achieving the accuracy of 90% and 93.8% respectively. As for the CNN-LSTM model, we achieved an accuracy of 93.5%.

This article investigates the formation of spontaneous coordination in a row of flexible 2D flaps (artificial cilia) in a chamber filled with a high viscous liquid (Re = 0.12). Each flap is driven individually to oscillate by a rotary motor with the root of the flap attached to its spindle axle. A computer-vision control loop tracks the flap tips online and toggles the axle rotation direction when the tips reach a pre-defined maximum excursion. This is a vision-controlled implementation of the so-called “geometric clutch” hypothesis. When running the control loop with the flaps in an inviscid reference situation (air), they remain in their individual phases for a long term. Then, the flaps are studied in the chamber filled with a highly viscous liquid, and the same control loop is started. The flexible flaps now undergo bending due to hydrodynamic coupling and come, after a maximum of 15 beats, into a synchronous metachronal coordination. The study proves in a macroscopic lab experiment that viscous coupling is sufficient to achieve spontaneous synchronization, even for a symmetric cilia shape and beat pattern.

Inventory management is a core part of supply chains, and over the years it has been increasingly challenged by the need to balance economic performance with environmental considerations. While prior reinforcement learning (RL) studies have incorporated carbon emissions indirectly through cost penalties or regulatory constraints, this work addresses an existing gap by treating emissions as an independent optimization objective. This study examines RL as an adaptive decision‑making approach for inventory optimization with two objectives: maximizing profit and minimizing carbon emissions. The problem is formulated as a Markov Decision Process, and four RL algorithms Proximal Policy Optimization (PPO), Phasic Policy Gradient (PPG), Advantage Actor‑Critic (A2C), and Double Deep Q‑Network (DDQN) are evaluated under identical experimental conditions. Carbon emissions are explicitly modeled in the reward function rather than embedded within operating costs. The results show that PPG achieves the highest profitability with only a modest increase in emissions, while DDQN converges faster but yields lower profit overall. Sensitivity analysis indicates that reward weighting strongly influences policy behavior, with PPO providing the most stable trade‑off between profitability and emissions.

Formation control in multi-robot systems (MRS) is essential for collaborative transport, environmental surveillance, material handling, and distributed monitoring. A major challenge in MRS is maintaining predefined formations or cooperative task execution when individual robots experience operational faults, potentially isolating them from the group. In mission-critical scenarios, preserving the number of operational robots is crucial for task success. To address this, we propose a Robot Replacement approach framework for differential wheeled mobile robots. This approach isolates faulty robots and dynamically replaces them with pre-deployed spares, ensuring uninterrupted formation tasks. A graph theory-based framework models inter-robot communication and formation topology, enabling decentralized coordination. The proposed techniques were implemented in a MATLAB/Simulink simulation environment. The simulated robots are equipped with LiDAR, an inertial measurement unit (IMU), and wheel encoders for navigation. Simulation results demonstrate that the framework successfully maintains the target formation and task continuity during robot failures by dynamically integrating replacements with minimal disruption.