Mandatory lane changes at intersections often lead to intricate conflicts and traffic oscillations.  The advent of connected and autonomous vehicles (CAVs) is expected to mitigate these disruptions by coordinating acceleration and lane-change behaviors. Addressing this, researchers developed SS-MA-PPO, a novel Multi-Agent Reinforcement Learning (MARL) framework that assists CAVs in coordinating these critical decisions. Evaluated against a real-world dataset from Langfang, this method significantly improves traffic efficiency compared with traditional models and other Multi-Agent Reinforcement Learning baselines.

Researchers at EPFL have developed a deep-learning framework that dramatically improves vehicle re-identification in large-scale drone traffic monitoring. By combining visual features with traffic-based travel-time predictions grounded in shockwave theory, the system can reliably match vehicles that appear nearly identical from a bird’s-eye view. Tested on one of the largest multi-UAV traffic datasets, the approach boosts ReID accuracy by 36.8 percent, enabling more robust urban vehicle tracking.

Autonomous driving requires real-time interaction between vehicles and infrastructure to ensure safety and efficiency. However, current V2X system focused on simulation environments or constrained testbeds, overlooking critical aspects, including the scalability of the autonomous driving environment, as well as infrastructures, vehicles, and control data exchange. To address these issues, this study proposes mOS, a modular edge-intelligent framework validated through real-world intersection deployments and metaverse-based mixed-reality simulations. Built on a containerized and extensible architecture, mOS enables dynamic coordination among vehicles, infrastructure, and virtual entities. Experimental results demonstrate enhanced safety, responsiveness, and scalability while overcoming the limitations of conventional J2735-based V2X systems. Operating over commercial 5G with acceptable latency, the framework provides a cost-effective and practical platform for next-generation intelligent transportation systems.

Understanding traffic scenes under adverse conditions such as rain, fog, night-time illumination, and motion blur remains a major challenge for intelligent transportation systems. Researchers at Tsinghua University propose TrafficPerceiver, a multimodal large language model framework that unifies traffic scene understanding and target-oriented segmentation under natural language instructions. By incorporating reinforcement learning based on group-relative policy optimization, the framework improves robustness and interpretability in complex real-world traffic environments.

Researchers have unlocked the secrets behind the extraordinary maneuverability of the black ghost knifefish, a freshwater species known for its effortless forward, backward, and hovering movements. By systematically analyzing the fin’s unique morphology and wave-like motion, the team has established a comprehensive kinematic database that could bridge the gap between biological propulsion and bio-inspired robotic design, potentially revolutionizing underwater vehicle performance in complex environments.