warmae/AIM-3D-Eulerian-trail

3D Voxel Infill Path Planning (RL Model)

This model takes a 3D voxel occupancy map as input and generates a continuous toolpath that fills all empty space. No CAD model is required — the policy is trained via reinforcement learning with intrinsic curiosity.

---

Features

• CAD-less volumetric infill path generation
• Reinforcement learning policy (26 discrete 3D actions)
• Supports any binary voxel grid (1: empty / 0: solid)

---

Quick Start

import torch

model = torch.load("model.pt")
model.eval()

# voxel_map: torch.Tensor [D, H, W]
trajectory = model.generate_path(voxel_map)

---

Citation

If you use this model, please cite:

Yang et al., CAD-Less Volumetric Path Generation for Lunar 3D Printing via Intrinsic Curiosity Motivation, 2025.

---

License

Research use only.