alirezaaminzadeh/nesting-placement-policy

Nesting Placement Policy

Policy network for learning-to-guide-search in 2D industrial nesting.

Architecture

The model does not output final layouts. It predicts:

1. Next piece candidate — priority scoring over remaining pieces
2. Best orientation — upright vs. 90° rotation
3. Promising region — spatial bias for heuristic placement
4. Strategy recommendation — solver selection under instance features

A classical heuristic (Skyline / Guillotine) completes placement.

Training Data

Features derived from clallier/nesting-tasks-2d task distributions.

Intended Use

• Guide nesting search in MDF, glass, steel, fabric, leather, PCB, print, packaging
• Recommend solver strategy based on layout state features
• Pair with NestForge Space for interactive planning

Limitations

• Rectangle abstraction of irregular shapes
• Policy weights are lightweight feature-based model (not end-to-end layout generation)
• Best suited as search guidance, not standalone optimizer

License

Apache-2.0