The exploration-exploitation dilemma is a long-standing topic in deep reinforcement learning. In recent research, a noise-driven enhancement for exploration algorithm has proposed for UAV autonomous navigation. This algorithm introduces a differentiated exploration noise control strategy based on the global navigation training hit rate and the specific situations encountered by the UAV in each episode. Furthermore, it designs a noise dual experience replay buffer to amplify the distinct effects of noisy and deterministic experiences. This approach reduces the computational cost associated with excessive exploration and mitigates the problem of the navigation policy converging to a local optimum.

Silicon carbide (SiC) and silicon nitride (Si₃N₄) powders are critical raw materials for advanced ceramics. However, traditional synthesis methods face four major challenges: difficulty in achieving SiC nanosizing, difficulty in realizing Si₃N₄ high purification, the need for external energy input for the weakly exothermic Si-C reaction, and the requirement of adding large amounts of diluents to enable the combustion synthesis of the strongly exothermic Si-N₂ reaction. Recently, a research team utilized the difference in heat release between the Si-N₂ and Si-C reactions. By means of chemical furnace encapsulation, the strong heat release from the Si-N₂ reaction was used to induce the combustion synthesis of the weakly exothermic Si-C reaction system. Through the regulation of the combustion reaction temperature field and the partial pressure of CO reducing gas, β-SiC powders with an average particle size of only 30 nm and high-purity pink β-Si₃N₄ powders with an oxygen content as low as 0.46 wt% were successfully synthesized. Their work is published in the journal Industrial Chemistry & Materials on 10 October.

University of Tennessee, Knoxville researchers Jindong Tan and Dr. Gregory Mancini developed a wireless robotic imaging device for minimally invasive surgery, integrating AI for enhanced optics and precision. Funded by a $1 million grant from the National Science Foundation, the device reduces incisions, improves outcomes, and expands surgical possibilities in tight spaces like joints and the brain, fostering collaborative innovation in medical technology.

The balance between electricity supply and demand requires advanced technologies and precise management, especially given the growing presence of renewable sources. Researchers are working on new energy storage and control strategies to ensure a stable and secure energy supply, preventing blackouts like the one that occurred on April 28, 2025.

The project “Management of Renewable Systems with Storage and Converter Control to Contribute to the Operation of the Future Power System”, led by Professors Emilio Pérez Soler and Ignacio Peñarrocha Alós, members of the Electricity, Electronics and Automation Research Group, is progressing toward its goal of integrating renewable energies through the development of advanced control strategies and their experimental validation on a platform that allows real-time testing of the combined operation of batteries, converters and controllers.

So far, the research team has developed predictive models related to the electricity market, such as daily market prices and services aimed at regulating the power system frequency. These models have been used to define a strategy based on deep reinforcement learning, enabling optimal participation of grid-connected storage systems in various electricity markets. Regarding lithium-ion batteries, new techniques have been developed to estimate their state of charge and health, improving both performance and lifespan.

Fashion trend prediction has traditionally relied on experts’ intuition and creativity, as well as big-data analysis for incorporating insights on consumer behaviors. However, this method remains inaccessible to fashion students and small brands. Now, researchers show how AI can supplement expert knowledge with data-based insights, enabling individuals to predict trends with more accuracy and nuance. To this end, the researchers developed a novel prompting technique to enable ChatGPT to provide more accurate and specific responses.

New developments in the aerospace industry require novel materials that can withstand extremely high temperatures. Recently, a team of scientists from Hanbat National University has come up with a stable, superior oxidation-resistant coating layer on a TiTaNbMoZr high-entropy alloy using a sequential two-step B–Si pack cementation process. This next-generation materials science technology is expected to revolutionize the defense and aerospace sectors.