The world is being quietly rearranged by people who write very long documents.


The title they went with Out-of-Sight Embodied Agents: Multimodal Tracking, Sensor Fusion, and Trajectory Forecasting Noisy translates that to

AI learns to track people and vehicles even when cameras can't see them


Researchers improved a machine learning system that predicts where pedestrians and vehicles are heading when they're hidden from camera view — blocked by buildings, out of frame, or obscured by occlusion. This matters because autonomous vehicles, robots, and security systems need to know where unseen objects are going to avoid collisions and stay safe in the real world, where camera coverage is always incomplete.
For the first time, this work uses camera geometry and positioning data to clean up noisy sensor readings of out-of-sight objects, rather than requiring perfect visual information — closing a genuine gap between what vision systems assume in the lab and what they actually face on streets and in warehouses where people and vehicles disappear regularly.

If you insist
Read the original →