Sensor2Sensor: Cross-Embodiment Sensor Conversion for Autonomous Driving

Researchers unveiled Sensor2Sensor, a method that converts monocular dashcam video into structured autonomous driving logs.

The system generates paired multi-view camera images and LiDAR point clouds using reconstructed training data and diffusion models.

It leverages 4D Gaussian Splatting to create realistic multi-sensor sequences from in-the-wild driving footage.

The approach could help AV teams train and test systems on far more diverse real-world scenarios than fleet logs alone.