SPW-Nav: A Streaming Panoramic World Model for Language-Guided Navigation
Yunheng Liu, Ziqi Cai, Siqi Yang, Yimu Wang, Minggui Teng, Jiaming Tan, Shuchen Weng, Erwin Wu and 2 others
Alaya Lab · Peking University · Institute of Science Tokyo
Published October 6, 2026 · on the feed October 9, 2026 · ▲ 5 · arXiv 2610.08941
The abstract
Language-guided panoramic video generation benefits various downstream applications, such as interactive 3D scene exploration, virtual reality experiences, and embodied agent training. Existing panoramic generators follow predefined trajectories, and interactive world models act through low-level actions in perspective views. We propose SPW-Nav, a streaming panoramic world model that understands movement instructions and streams one minute of 2K 360-degree video in real time from a single panorama. SPW-Nav interprets each instruction in the previously generated panorama as camera motion. Spherical rotation decoupling applies rotation exactly on the sphere, pose-aligned conditioning keeps translation inputs bounded over long streams, and a multi-term memory with a few-step generator continues the scene as instructions change. We also build SPW-NavSet, panoramic videos with camera trajectories and verified instructions. Driven by language, SPW-Nav outperforms prior panoramic generators in camera-following accuracy and video quality, and supports on-the-fly instruction switching.
The Brief
2:00A short podcast summary of the paper. Written and voiced by AI.
Transcript
SPW-Nav. A Streaming Panoramic World Model for Language-Guided Navigation. By Yunheng Liu and nine others, from Alaya Lab, Peking University, and one other institution. Published on the sixth of October.
Panoramic video generated from language is useful downstream, for interactive 3D scene exploration, virtual reality, and training embodied agents. The existing tools split in an awkward way. Panoramic generators follow trajectories fixed in advance, while interactive world models act through low-level actions in perspective views rather than full panoramas. One kind sees the whole sphere but cannot be steered as it goes; the other can be steered, but only through a narrow window and a vocabulary of low-level controls rather than language.
SPW-Nav is a streaming panoramic world model that takes movement instructions in language. Starting from a single panorama, it streams one minute of 2K, three hundred sixty degree video in real time. Each instruction is interpreted inside the panorama generated so far, as a motion of the camera. Language goes in, a camera move comes out, and the next stretch of panorama follows from that move.
Three pieces make that hold up over a long stream. Spherical rotation decoupling applies rotation exactly on the sphere. Pose-aligned conditioning keeps the translation inputs bounded however long the stream runs. And a multi-term memory, paired with a few-step generator, continues the scene as the instructions change. Alongside the model, the authors build SPW-NavSet, a collection of panoramic videos paired with camera trajectories and with instructions that have been verified.
Driven by language, SPW-Nav outperforms prior panoramic generators in both camera-following accuracy and video quality, so it goes where it is told and looks better doing it. It also supports switching instructions on the fly, mid-stream.
The contribution is to turn panoramic generation from playback along a set path into something steerable: words become camera motion, applied to the full sphere, continuously, while the video is still streaming.
More with an account
Listen to every day’s papers
Daily playlists, deep dives, quests for the papers, and quizzes on their key terms.
Join free