Scalable Geometric Fracture Assembly via Co-creation Space among Assemblers

TL;DR

This work tackles scalable geometric fracture assembly without relying on semantic information. It introduces a Co-creation Space among assemblers, where multiple agents compete for write access to a shared, bottlenecked workspace and collectively converge on consistent 6-DoF fracture poses with linear complexity in the number of assemblers. A geometric-based collision loss further mitigates ambiguities by actively separating similar fractures, and a modular geometric information routing component precedes assembly to capture geometry and relations. Across PartNet and Breaking Bad, the method achieves state-of-the-art performance and strong generalization, with ablations confirming the critical roles of the co-creation space and collision loss, and the approach offering practical implications for archaeology and robotic assembly.

Abstract

Geometric fracture assembly presents a challenging practical task in archaeology and 3D computer vision. Previous methods have focused solely on assembling fragments based on semantic information, which has limited the quantity of objects that can be effectively assembled. Therefore, there is a need to develop a scalable framework for geometric fracture assembly without relying on semantic information. To improve the effectiveness of assembling geometric fractures without semantic information, we propose a co-creation space comprising several assemblers capable of gradually and unambiguously assembling fractures. Additionally, we introduce a novel loss function, i.e., the geometric-based collision loss, to address collision issues during the fracture assembly process and enhance the results. Our framework exhibits better performance on both PartNet and Breaking Bad datasets compared to existing state-of-the-art frameworks. Extensive experiments and quantitative comparisons demonstrate the effectiveness of our proposed framework, which features linear computational complexity, enhanced abstraction, and improved generalization. Our code is publicly available at https://github.com/Ruiyuan-Zhang/CCS.

Figures (8)

• Figure 1: Comparison of Transformer-based assembly (left) and our method (right). Left: Assemblers may become confused when presented with multiple fractures to assemble simultaneously, leading to an inability to assemble any parts. Right: In our method, assemblers focus on essential fractures of parts, enabling efficient and effective task completion.
• Figure 2: The overall architecture of our framework. (a) geometric fractures feature extraction using shared parameters PointNet qi2017pointnet, (b) geometric fractures relational reasoning and information routing, (c) complete the assembly task among assemblers using the co-creation space, (d) MLP predictor for pose estimation.
• Figure 3: The illustration of geometric-based collision loss (The red circles and the blue triangles represent 2 similar fractures).
• Figure 4: Visual comparisons between our algorithm and the baseline methods on Breaking Bad (a, b) and PartNet dataset (c, d, e, f, g and h).
• Figure 5: Collision loss ablation study visualization.