In this study, we proposed a novel Knowledge-Informed Deep Learning (KIDL) paradigm that, to the best of our knowledge, is the first to unify behavioral generalization and traffic flow stability by systematically integrating high-level knowledge distillation from LLMs with physically grounded stability constraints in car-following modeling. Generalization is enhanced by distilling car-following knowledge from LLMs into a lightweight and efficient neural network, while local and string stability are achieved by embedding physically grounded constraints into the distillation process. Experimental results on real-world traffic datasets validate the effectiveness of the KIDL paradigm, showing its ability to replicate and even surpass the LLM's generalization performance. It also outperforms traditional physics-based, data-driven, and hybrid CFMs by at least 10.18% in terms of trajectory simulation error RMSE. Furthermore, the resulting KIDL model is proven through theoretical and numerical analysis to ensure local and string stability at all equilibrium states, offering a strong foundation for advancing AV technologies.

Researchers at Chang’an University have developed a novel combined virtual-real testing (CVRT) platform for validating autonomous vehicles. This innovative approach utilizes digital twin technology to simulate realistic scenarios and conduct parallel AEB (autonomous emergency braking) tests across various conditions. The results indicate that CVRT closely replicates real-world performance while significantly reducing test time by up to 70%. This breakthrough offers a safer, more efficient method for validating autonomous systems, with implications for scalable testing and regulation in the autonomous vehicle industry.

Socially compliant automated vehicles (SCAVs) mark a new frontier in human-centric driving automation. Integrating sensing, socially aware decision-making, safety constraints, spatial-temporal memory, and bidirectional behavioral adaptation, the proposed framework aims for AVs to interpret, learn from, and respond to human drivers. By embedding social intelligence into automated driving systems, this research paves the way for vehicles that not only drive safely but also drive socially.

Image reconstruction—the process of recovering clear images from incomplete or noisy data—has been advancing rapidly through deep learning. Yet most existing approaches rely on costly supervised training and lack theoretical transparency. A new survey maps the rise of unsupervised deep learning for image reconstruction, from traditional denoising-based priors to modern diffusion models. These methods learn structured visual information directly from unlabeled data, and have achieved impressive performance across various fields, including biomedical imaging and remote sensing. The study shows how unsupervised learning based image reconstruction unites neural network efficiency with solid mathematical foundations to achieve both interpretability and flexibility, offering a blueprint for next-generation imaging systems.

Researchers at the University of Melbourne have developed a new AI-based traffic signal control system called M²SAC that improves both fairness and efficiency at urban intersections. Unlike traditional systems focused only on cars, M²SAC accounts for pedestrians, buses, and other users. A key innovation is the phase mask mechanism, which dynamically adjusts green light timings to reduce delays. Tested on real Melbourne traffic data, the model outperformed existing methods, cutting congestion and balancing traffic flow more equitably. The approach supports smarter, fairer, and more inclusive transport systems for modern cities.

To address this challenge, researchers at Korea Advanced Institute of Science and Technology (KAIST) and Donghai Laboratory developed a new model called ProChunkFormer, which reconstructs vehicle trajectories from sparse and noisy GPS data, enabling more accurate mobility analysis and intelligent transportation planning.

The heterogeneity causes spatiotemporal inconsistencies in multimodal data, posing challenges for existing methods in multimodal feature extraction and alignment. First, in the temporal dimension, the microsecond-level temporal resolution of event data is significantly higher than the millisecond-level resolution of RGB data, resulting in temporal misalignment and making direct multimodal fusion infeasible. To address this issue, the researchers design an Event Correction Module (ECM) that temporally aligns asynchronous event streams with their corresponding image frames through optical-flow-based warping. The ECM is jointly optimized with the downstream object detection network to learn task-ware event representations.

Artificial photosynthesis is highly desired to enhance natural photosynthesis through enhanced photoelectron transfer from photocatalysts to natural photosystems. Graphitic carbon nitride (g-C₃N₄ or CN), a biocompatible organic polymer semiconductor, has emerged as a promising candidate for boosting natural photosynthesis, paving the way for sustainable agricultural technologies to increase crop yields.

Double transition metal nitrides and carbides (MXene) have garnered significant attention in the field of electromagnetic wave (EMW) absorption due to their distinctive structural properties. The design of efficient MXene-based EMW absorbers remains a formidable challenge in light of the high conductivity and strong van der Waals forces. In this work, we report for the first time the approach of the double-doping non-metal N and rare earth metal Ce-4f into Mo-MXene to construct Mo-MXene/MoO₂-N/Ce system. This process enables partial in-situ oxidation of Mo-MXene, thereby forming a heterostructure and enhancing the interface polarization. The introduction of Ce facilitates the hybridization between the 4f orbitals of rare earth Ce and the 4d orbitals of Mo, altering the electronic structure of Ce and Mo-MXene and promoting electron migration, which contributes to polarization loss. Furthermore, incorporating melamine into the precursor can induce N doping in Mo-MXene, thereby promoting dipolar polarization. Consequently, the double-doping of N and Ce enables the synergistic effects of interface polarization, dipole polarization, and conduction loss, leading to efficient EMW absorption. Therefore, at a frequency of 13.43 GHz and a matching thickness of 4.685 mm, the optimal reflection loss (R_L) value of Mo-MXene/MoO₂-N/Ce reaches -57.46 dB, which exceeds a large number of reported MXene-based absorbers. This research confirms that Mo-MXene/MoO₂-N/Ce is a promising EMW absorption material and provides valuable insights into modulating MXene-based EMW absorbers using rare earth elements.