This repository contains the policy checkpoint for
Bridge-WA, as presented in the paper
Bridge-WA: Predicting Where and How the World Changes for Robotic Action.
Bridge-WA is a lightweight world-action framework that distills a frozen future-change teacher into three compact priors:
A WorldBridge conditions the action transformer on these priors through multi-source attention memories and spatial-temporal biases, while the teacher model is removed at inference. By focusing action generation on where and how the scene will change, Bridge-WA suppresses nuisance appearance factors such as background, lighting, and distractors, leading to better generalization without deployment-time dense future-image generation.
For installation, training, and evaluation instructions on benchmarks such as VLABench, please refer to the official
GitHub repository.
1@article{bai2026bridgewa,
2 title = {Bridge-WA: Predicting Where and How the World Changes for Robotic Action},
3 author = {Bai, Yongjie and Wang, Hanting and Dai, Mingtong and Zhong, Qijun and Liu, Yang and Lin, Liang},
4 year = {2026}
5}
