Test-Time Spatial Reasoning for Robot Manipulation Using Generative Real-to-Sim

2026-09-27

AI Summary

This study addresses the challenge of spatial reasoning in long-horizon, multi-object robotic manipulation by proposing Simify, a training-free framework for real-time complex operations. The method reconstructs simulation assets from RGB-D images, integrating 3D generative models, vision-language models, and massively parallel physics simulation. By leveraging evolutionary search to optimize object layouts, it introduces the first training-free, test-time spatial reasoning paradigm. Experimental results demonstrate that the framework successfully executes complex rearrangement tasks involving unseen objects on physical robots, outperforming existing foundation model approaches. These findings validate the critical role of precise geometric modeling in enhancing sim-to-real transfer success rates.

Abstract

Spatial reasoning is fundamental to general robot intelligence, as it enables robots to complete long-horizon tasks involving multi-object interaction. We introduce Simify, a training-free, test-time framework that performs explicit spatial reasoning via massively parallel physics simulation. From a single RGB-D image of a scene, Simify reconstructs simulation-ready assets leveraging 3D generative models and vision-language models. Then given a task specified by a reward function (e.g., build the tallest tower), Simify launches thousands of parallel rollouts in simulation and performs an evolutionary search to optimize object arrangements, typically converging within seconds. We conduct quantitative experiments on real-robot hardware to demonstrate the ability of our framework to execute complex object rearrangement tasks end-to-end with previously unseen objects. Results show that our framework outperforms prior work on foundation models for spatial reasoning by effectively exploiting large-scale parallel simulation during inference, and also highlight the importance of complete and accurate geometry for successful sim-to-real transfer.

Problem

• Spatial Reasoning
• Robot Manipulation
• Object Rearrangement
• Sim-to-Real Transfer
• Long-horizon Tasks

Innovation

• Test-Time Spatial Reasoning
• Generative Real-to-Sim
• Massively Parallel Simulation
• Evolutionary Search
• Sim-to-Real Transfer