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

2026-08-17

AI Summary

This study addresses the absence of critical decision-making compute allocation mechanisms in hierarchical models for long-horizon manipulation by proposing τ0-VLA. This method introduces test-time compute into hierarchical Vision-Language-Action architectures for the first time, leveraging world models to guide reasoning and execution memory. By reformulating high-level subtask generation as a scalable dynamic search problem, it achieves decoupling between high- and low-level policies while enabling multimodal collaborative training. Experimental results demonstrate that this mechanism significantly improves both subtask prediction accuracy and closed-loop success rates in long-horizon tasks. These findings validate the effectiveness of on-demand compute allocation in enhancing robotic decision-making capabilities for complex, extended operations.

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.

Problem

• Long-horizon robot manipulation
• Hierarchical VLA models
• Test-time computation
• Subtask generation

Innovation

• Test-Time Computation
• Hierarchical VLA
• World Model
• Compute-Scalable Inference
• Long-Horizon Manipulation