Introducing a novel harmonic loss function for V2X cooperative 3D object detection that achieves task consistency while maintaining real-time performance. A deep dive into the method and experimental results.

How we combine RGB imagery, semantic segmentation, and ego-vehicle speed using attention mechanisms and recurrent neural networks to predict pedestrian crossing behavior in real-time.

Extending frustum-based point cloud methods to handle far-distance objects, a critical challenge for autonomous driving perception systems operating at highway speeds.