Scalability of the Fusion 360 Gallery extrusion-based reconstruction to real-world scenarios

Determine how well the Fusion 360 Gallery approach for programmatic CAD reconstruction—specifically, recovering a sequence of extrusion operations that reproduces a target 3D shape by deciding which given sketch to extrude and to what extent—scales to more real-world scenarios beyond the limited setting studied by Willis et al. (2020).

Background

In the discussion of concurrent datasets and generative approaches for CAD, the paper contrasts its focus on parametric sketch generation with Fusion 360 Gallery, which reconstructs 3D shapes by predicting extrusion sequences. The authors note that this setting assumes sketches are provided and restricts the model’s role to selecting which sketch to extrude and the extrusion magnitude.

Given this deliberate limitation, the authors explicitly state uncertainty about the approach’s scalability to broader real-world applications where design variability, complexity, and interactions may exceed the assumptions in Fusion 360 Gallery. This raises an open question about evaluating and demonstrating the method’s performance and applicability in more realistic scenarios.

References

While \citep{willis20} is a step towards full parametric CAD generation, it is unclear how well this approach will scale to more real-world scenarios.

— Computer-Aided Design as Language  (2105.02769 - Ganin et al., 2021) in Related work — Datasets and generative models for CAD (Section 2)

Given the modest dataset of 93 traces and approximately 2,500 derived training pairs, our claims are limited to operation synthesis on held-out traces within this bounded garment domain; broader deployment in professional prototyping and small-batch production remains to be validated.

— TailorCoPilot: Enabling Agentic Pattern Making with Version-Controlled State Tracking  (2608.25462 - Sun et al., 26 Aug 2026) in Section 6, Limitations and Future Directions