Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

RoboStereo: Dual-Tower 4D Embodied World Models for Unified Policy Optimization

Ruicheng Zhang, Guangyu Chen, Zunnan Xu, Zihao Liu, Zhizhou Zhong, Mingyang Zhang, Jun Zhou, Xiu Li

Abstract

Scalable Embodied AI faces fundamental constraints due to prohibitive costs and safety risks of real-world interaction. While Embodied World Models (EWMs) offer promise through imagined rollouts, existing approaches suffer from geometric hallucinations and lack unified optimization frameworks for practical policy improvement. We introduce RoboStereo, a symmetric dual-tower 4D world model that employs bidirectional cross-modal enhancement to ensure spatiotemporal geometric consistency and alleviate physics hallucinations. Building upon this high-fidelity 4D simulator, we present the first unified framework for world-model-based policy optimization: (1) Test-Time Policy Augmentation (TTPA) for pre-execution verification, (2) Imitative-Evolutionary Policy Learning (IEPL) leveraging visual perceptual rewards to learn from expert demonstrations, and (3) Open-Exploration Policy Learning (OEPL) enabling autonomous skill discovery and self-correction. Comprehensive experiments demonstrate RoboStereo achieves state-of-the-art generation quality, with our unified framework delivering >97% average relative improvement on fine-grained manipulation tasks.

RoboStereo: Dual-Tower 4D Embodied World Models for Unified Policy Optimization

Abstract

Scalable Embodied AI faces fundamental constraints due to prohibitive costs and safety risks of real-world interaction. While Embodied World Models (EWMs) offer promise through imagined rollouts, existing approaches suffer from geometric hallucinations and lack unified optimization frameworks for practical policy improvement. We introduce RoboStereo, a symmetric dual-tower 4D world model that employs bidirectional cross-modal enhancement to ensure spatiotemporal geometric consistency and alleviate physics hallucinations. Building upon this high-fidelity 4D simulator, we present the first unified framework for world-model-based policy optimization: (1) Test-Time Policy Augmentation (TTPA) for pre-execution verification, (2) Imitative-Evolutionary Policy Learning (IEPL) leveraging visual perceptual rewards to learn from expert demonstrations, and (3) Open-Exploration Policy Learning (OEPL) enabling autonomous skill discovery and self-correction. Comprehensive experiments demonstrate RoboStereo achieves state-of-the-art generation quality, with our unified framework delivering >97% average relative improvement on fine-grained manipulation tasks.
Paper Structure (25 sections, 18 equations, 12 figures, 7 tables, 2 algorithms)

This paper contains 25 sections, 18 equations, 12 figures, 7 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Embodied World Models and Geometric Consistency
  4. Policy Optimization and Test-Time Scaling for VLAs
  5. Method
  6. Architecture Design
  7. Training Strategy for World Model
  8. World Model-Based Policy Optimization
  9. Experiments
  10. Experimental Setup
  11. Video Quality Evaluation
  12. Policy Optimization Evaluation
  13. Ablation Study
  14. Conclusion
  15. Inference Efficiency Analysis
  16. ...and 10 more sections

Figures (12)

  • Figure 1: (a) Qualitative comparison of RoboStereo against SOTA EWMs. (b) Quantitative comparison of unified policy optimization framework against traditional paradigms.
  • Figure 2: RoboStereo Architecture. Symmetric dual DiT towers (a) process RGB and XYZ pointmaps via bidirectional cross-attention for visual-geometric fusion (b) and a Gaussian head for flexible-viewpoint rendering. Dual-path action-conditioned timestep embedding mechanism (c) ensures precise frame-level trajectory control.
  • Figure 3: Illustration of Test-Time Policy Augmentation (TTPA).
  • Figure 3: Task success rates (%) under different optimization paradigms in the MimicGen simulation benchmark.
  • Figure 4: Illustration of Imitative-Evolutionary Policy Learning (IEPL).
  • ...and 7 more figures