Visual Foresight for Robotic Stow: A Diffusion-Based World Model from Sparse Snapshots

Lijun Zhang, Nikhil Chacko, Petter Nilsson, Ruinian Xu, Shantanu Thakar, Xibai Lou, Harpreet S. Sawhney, Zhebin Zhang, Mudit Agrawal, Bhavana Chandrashekhar, Aaron Parness · 2026

Read Visual Foresight for Robotic Stow: A Diffusion-Based World Model from Sparse Snapshots by Lijun Zhang, Nikhil Chacko, Petter Nilsson, Ruinian Xu, Shantanu Thakar, Xibai Lou, Harpreet S. Sawhney, Zhebin Zhang, Mudit Agrawal, Bhavana Chandrashekhar, Aaron Parness (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-13347.txt.

Description

Warehouse robots stow millions of items into bins, and knowing how a bin will look after a planned stow could prevent bad placements before they happen. FOREST is a world model for that foresight. It represents bin states as item-aligned masks and uses a latent diffusion transformer, conditioned on stow intent, to predict the post-stow layout from sparse snapshots.

Predicted layouts matched true layouts far better than heuristic baselines on geometric agreement. Swapping real post-stow masks for FOREST predictions cost only modest performance in load-quality checks and multi-stow reasoning, suggesting the forecasts are good enough to guide planning. Lijun Zhang leads the work with ten co-authors.

Catalog record

CREATORLijun Zhang, Nikhil Chacko, Petter Nilsson, Ruinian Xu, Shantanu Thakar, Xibai Lou, Harpreet S. Sawhney, Zhebin Zhang, Mudit Agrawal, Bhavana Chandrashekhar, Aaron Parness
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE59 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-13347

Browse the full library · About the Tower of Babel