MEGA Hub
New

τ_0-VLA: a Hierarchical Robot Foundation Model with World-Model-Guided Test-Time Computation

Published on Aug 16 Submitted by MEGAon Aug 21#1 Paper of the day Shanghai Innovation Institute

Authors

Do you know Xiaowei Cai?You can claim authorship or link another user.Do you know Yunuo Cai?You can claim authorship or link another user.Do you know Bingao Chen?You can claim authorship or link another user.Do you know Jingxiao Chen?You can claim authorship or link another user.Do you know Zhi Chen?You can claim authorship or link another user.Do you know Siyuan Feng?You can claim authorship or link another user.Do you know Tengyu Hou?You can claim authorship or link another user.Do you know Jingshun Huang?You can claim authorship or link another user.Do you know Han Jiang?You can claim authorship or link another user.Do you know Runkun Ju?You can claim authorship or link another user.Do you know Dong Li?You can claim authorship or link another user.Do you know Mingxiang Li?You can claim authorship or link another user.Do you know Shaowei Li?You can claim authorship or link another user.Do you know Xinchen Li?You can claim authorship or link another user.Do you know Yifan Li?You can claim authorship or link another user.Do you know Yi Liu?You can claim authorship or link another user.Do you know Zhongyuan Liu?You can claim authorship or link another user.Do you know Jianlan Luo?You can claim authorship or link another user.Do you know Junwen Miao?You can claim authorship or link another user.Do you know Ruiqi Ni?You can claim authorship or link another user.Do you know Buqing Nie?You can claim authorship or link another user.Do you know Mingjie Pan?You can claim authorship or link another user.Do you know Xinlin Ren?You can claim authorship or link another user.Do you know Jianheng Song?You can claim authorship or link another user.Do you know Jiaxu Wang?You can claim authorship or link another user.Do you know Peiqi Wang?You can claim authorship or link another user.Do you know Sen Wang?You can claim authorship or link another user.Do you know Xiaoyan Wang?You can claim authorship or link another user.Do you know Dafeng Wei?You can claim authorship or link another user.Do you know Dongming Wu?You can claim authorship or link another user.Do you know Pengwei Xie?You can claim authorship or link another user.Do you know Pu Yang?You can claim authorship or link another user.Do you know Hangjian Ye?You can claim authorship or link another user.Do you know Xiangyu Yue?You can claim authorship or link another user.Do you know Jinyu Zhang?You can claim authorship or link another user.Do you know Qinglin Zhang?You can claim authorship or link another user.Do you know Xueyong Zhao?You can claim authorship or link another user.Do you know Pengfei Zhou?You can claim authorship or link another user.Do you know Yue Zhou?You can claim authorship or link another user.

Abstract

Long-horizon robot manipulation requires a robot to both execute individual skills reliably and sequence them coherently over extended tasks. Most hierarchical vision-language-action (VLA) models make each such decision with a single forward pass, leaving no mechanism to allocate additional computation to difficult or consequential choices. We introduce τ_0-VLA, a hierarchical robot foundation model that formulates high-level subtask generation as a compute-scalable inference problem through world-model-guided test-time computation. At each inference step, the high-level policy uses execution memory to generate a subtask and, when needed, searches over alternatives before committing to its output. A low-level policy then executes the generated subtask across multiple robot embodiments. The policy is trained on 40,115 hours of heterogeneous real-world data with multimodal co-training. Across in-domain and distribution-shifted settings, allocating additional test-time computation substantially improves next-subtask prediction accuracy, and these gains translate into higher closed-loop success on long-horizon robot manipulation tasks.

Community

00