INF-3DP: Implicit Neural Fields for Collision-Free Multi-Axis 3D Printing

Jiasheng Qu1, Zhuo Huang1, 2, Dezhao Guo1, Hailin Sun1, Aoran Lyu2,
Chengkai Dai3, Yeung Yam1,3, Guoxin Fang1, 3, *

SIGGRAPH Asia 2025 (ACM Transactions on Graphics)

1The Chinese University of Hong Kong, China.
2The University of Manchester, Manchester, United Kingdom.
3Centre for Perceptual and Interactive Intelligence, Hong Kong, China.
*Corresponding author: guoxinfang@cuhk.edu.hk

Paper(arXiv) Video(YouTube) Video(Bilibili) Code(coming soon)

We introduce a general, scalable computational framework for multi-axis 3D printing based on implicit neural fields (INFs) that unifies all stages of toolpath generation and global collision-free motion planning. Compared to explicit-representation-based methods, INF-3DP achieves up to two orders of magnitude speedup and significantly reduces waypoint-to-surface error - it can handle millions of waypoints with differentiable collision-free planning in seconds with the help of GPU-based parallel computing.

Video Presentation

Video @ bilibili

Training-free time-varying SDF for dynamic printing object modeling, enables differentiable motion planning for collision avoidance.

INF-3DP Framework

Directional Weight Score

The computational pipeline of INF-3DP. (a) Begins with representing the printing model as a point cloud for training, (b) a SDF is computed as an implicit volumetric representation of the model. (c) Guidance fields are then optimized with fabrication-aware objectives on both the surface (2-manifold) and interior (3-manifold) domains, (d) guiding Reeb-graph-based partitions to ensure printing continuity. (e) Infill and density fields are constructed for variable density infill, (f) and the optimized printing sequence field is established. (g) By evaluating the time-varying SDF, the shape of the object at each stage of printing is approximate, enabling efficient global collision detection. (h) Differentiable quaternion field optimization is then applied for collision-free and smooth motion planning. (I) The final fabrication plan, visualized as streamlines of the motion frame field, integrates all optimized fields to ensure support-free, high-quality, and collision-free multi-axis 3DP.

Physical fabrication

BibTeX


      @article{Qu2025INF3DP,
      title={INF-3DP: Implicit Neural Fields for Collision-Free Multi-Axis 3D Printing},
      author={Qu, Jiasheng and Huang, Zhuo and Guo, Dezhao and Sun, Hailin and Lyu, Aoran and  Dai, Chengkai and Yam, Yeung and Fang, Guoxin},
      journal={ACM Transactions on Graphics (TOG)},
      note={To appear in SIGGRAPH Asia 2025},
      pages={1--18},
      year={2025},
      publisher={ACM}
      }