The team led by Ziqi Cai released SPW-Nav, a streaming panoramic world model that turns language directions into live 360‑degree video. It works from one starting panorama and can follow instructions like “turn left” or “move forward” while keeping video quality high. The system uses spherical rotation and pose‑aligned conditioning to avoid drift over long streams. They also released SPW-NavSet, a dataset of panoramic videos with verified camera trajectories and instructions.
Why it matters
It could let VR apps create on‑the‑fly tours from a single photo, reducing the need for pre‑recorded footage.
