MIT’s HardFlow algorithm steers generative AI around hard constraints without retraining
MIT researchers have developed HardFlow, a plug-and-play algorithm that lets pretrained generative AI models satisfy strict safety and physical constraints ...
MIT researchers have developed HardFlow, a plug-and-play algorithm that lets pretrained generative AI models satisfy strict safety and physical constraints ...
Researchers have built an end-to-end deep learning framework that combines 3D hand pose estimation with hand action recognition from egocentric ...
Researchers in Japan have developed a GPU-accelerated path-tracing simulator that generates realistic event camera data up to three times faster ...
Researchers in China have developed a graph-based deep reinforcement learning system that learns to schedule single-armed cluster tools processing multiple ...
A new hierarchical framework combining an obstacle-aware A* algorithm, B-spline smoothing, and fuzzy-adaptive dynamic window control lifts mobile robot obstacle ...
SUTD researchers have created ALBATROSS, a nature-inspired robot that autorotates down like a maple seed, lands on water, and then ...
A new lightweight YOLOv10-based vision system paired with an improved generative grasping network lets industrial robots detect and grip objects ...
Researchers at Shenyang University of Technology have developed a distance-aware adaptive graph neural network that estimates 6D object poses from ...
Researchers at the University of Western Australia have developed a fully proprioceptive control framework that lets the Unitree G1 humanoid ...
A new conceptual framework proposes that AI and robotics for older adults should be judged by whether they expand later-life ...
© 2026 Scienmag - Science Magazine
© 2026 Scienmag - Science Magazine