Exploration

A team of researchers has taken a step beyond traditional reinforcement learning approaches to multimodal reasoning, addressing a core challenge in how artificial intelligence assigns credit to individual tokens in video-based tasks. The problem is f

Autonomous driving systems face an unrelenting adversary: the real world’s unpredictability. Weather, lighting, and environmental anomalies can render even the most sophisticated sensors useless, turning what should be a routine drive into a high-sta

Researchers at the University of Trento and Intel Labs have quietly dismantled one of the last major barriers in automated X-ray security screening, a breakthrough that could reshape how airports, borders, and critical infrastructure detect threats.

A new study exposes a blind spot in how autonomous driving systems fuse camera and LiDAR data, revealing that the reliability of each sensor is not merely a function of distance or weather, as commonly assumed, but is instead dictated by the specific

In a breakthrough for autonomous systems and dynamic scene understanding, a team of researchers has developed MESSENGER, a memory-enhanced sequential scene flow estimation model that redefines how machines interpret motion in 3D environments. Scene f

A team of researchers has quietly developed a framework that could reshape how machines interpret video data—a breakthrough with immediate implications for defence and security applications. The work, led by Yudong Han and collaborators from Sun Yat-

UAV photogrammetry has quietly revolutionised high-precision terrain monitoring in sectors where millimetre-level accuracy matters—mining, civil infrastructure, and, increasingly, defence and security. Yet while the hardware—drones, cameras, GNSS rec

The latest research out of South Korea’s Gwangju Institute of Science and Technology introduces a radical shift in how machines learn to recognize objects in images—one that could quietly upend everything from satellite surveillance to battlefield re

A team of researchers has just exposed a critical flaw in how artificial intelligence models are taught to forget unwanted visual domains—such as cartoon-style images or stylized medical diagrams—and their findings could have significant implications

Underground mines are some of the most hostile environments for spatial data collection, where GPS signals vanish and every centimetre of rock movement can signal risk. Yet the last decade has seen a quiet revolution in how miners map these spaces, w