Paper Overview
Field: Computer Vision (CV) Authors: Jiahao Zhang, Yeying Fan, Moitreya Chatterjee, et al. Published: 2026-09-30 arXiv: 2609.26761
Introduction
The task of 3D assembly requires translating an understanding of parts and their relationships into precise spatial arrangements. A central question: can pretrained general-purpose agents assemble objects through visual interaction, without additional assembly-specific fine-tuning?
The AssemblyWorld Environment
AssemblyWorld is an interactive 3D environment in which agents:
- Inspect rendered views and manipulate supplied rigid parts
- Are guided by images or assembly manuals when available
- Perceive part geometry through 2D views rather than direct access to mesh vertices or faces
- Have their resulting assemblies evaluated geometrically
- Evaluation of eight agent systems reveals significant capability differences.
- The strongest system achieves 80.9% part accuracy but only 59.4% full assembly success rate.
- Evaluated open-source systems lag far behind stronger closed-source peers in both execution reliability and assembly accuracy.
- Analysis of visual references, interaction trajectories, and failure cases shows that agents retain localization errors even while correcting assemblies.
AssemblyWorldBench
Built on this environment, AssemblyWorldBench comprises 100 assembly tasks across 80 objects spanning three domains:
1. Furniture 2. Industrial assembly 3. Fracture reassembly
Key Findings
Conclusion
AssemblyWorld provides a general platform for evaluating the capabilities of interactive assembly agents and for characterizing the gap between approximate structural recovery and precise reconstruction.
---
*Auto-collected on 2026-10-02.*