COMPAS Forge 🛠️

Note

An Open-Source, High-Performance Rust-Backed Geometry Verification, Robotic Swept-Collision & Preflight Assembly Clearance Engine for the COMPAS Framework

COMPAS Forge is an open-source, high-performance geometry verification and digital fabrication preflight suite developed to bridge the gap between computational design environments (such as Rhino, Grasshopper, and Blender) and real-world physical manufacturing.

Bound to the COMPAS ecosystem, this library provides microsecond-precision topological and physical validation checks, optimizing CAD models before exporting them to robotic paths or CNC machinery.

This is an open-source, research-oriented library designed for academic collaboration. We invite researchers, roboticists, and computational designers to contribute, extend fabrication profiles, and integrate advanced geometric solvers.

---

Key SOTA Architectural Features

• Zero-Copy Memory Shared FFI (Pillar 1): Directly transmutes flat contiguous C-compatible memory layouts (PyBuffer) between Python and parallel Rust threads without copy-pasting, eliminating $O(N)$ string serialization overhead.
• Continuous Collision Detection (CCD) with Rotational Sub-Stepping (Pillar 2): Supports simultaneous translation and rotation (helical sweep toolpaths) using piecewise temporal sub-stepping, evaluating exact Time of Impact (TOI) down to 6 decimal places in microseconds.
• Parallel Assembly Contact Solver (Pillar 3): Combines parallel R-Tree broad-phase spatial filters with a robust 2D Sutherland-Hodgman Polygon Clipper to extract exact contact areas, centroids, and normals across thousands of adjacent blocks in milliseconds.
• Universal COMPAS 1.x / 2.x Schema Parser: Embedded dual-format JSON deserializer that natively resolves both nested arrays and string-keyed maps, solving version compatibility breaks seamlessly.
• Transparent Plugin Acceleration (Pillar 4): Integrated with COMPAS core plugin auto-discovery. Standard queries like .is_manifold() and .is_closed() are transparently intercepted and solved by the Rust engine.

---

Scaling Performance Benchmark

We evaluated the performance of the transparent plugin acceleration by querying .is_closed() on dense parametric geodesic dome structures. COMPAS pure-Python execution times are compared against COMPAS Forge's Rust FFI memory shared engine:

Mesh Facets (Count)	Pure Python (ms)	Rust Forge FFI (ms)	Speedup Factor
400	0.476 ms	0.031 ms	15.35x
1,600	1.975 ms	0.070 ms	28.41x
3,600	4.169 ms	0.129 ms	32.29x
6,400	7.576 ms	0.215 ms	35.27x
10,000	11.240 ms	0.362 ms	31.06x
22,500	27.878 ms	0.737 ms	37.85x
40,000	49.254 ms	1.173 ms	41.99x

At 40,000 facets, pure-Python COMPAS takes ~49 ms to resolve watertightness, dropping the viewport frame rate below the interactive threshold (~20 FPS). COMPAS Forge processes the same geometry in 1.17 ms (~850 Hz), enabling lag-free real-time manipulation in CAD viewports (Rhino 8 / Grasshopper).

---

Installation

Standard Installation

Once released, COMPAS Forge can be installed from PyPI:

pip install compas-forge

Installation from Source

Requirements:

• Rustc 1.96+
• Python 3.9+

git clone https://github.com/moaminmo90/compas-forge.git
cd compas-forge
pip install -e .

---

Python API Usage Guide

1. Transparent COMPAS Core Acceleration

Simply install compas_forge. Native COMPAS queries are transparently routed to the Rust core under the hood:

from compas.datastructures import Mesh

mesh = Mesh.from_json("dense_model.json")

# Executed natively in Rust in microseconds!
if mesh.is_manifold() and mesh.is_closed():
    print("Mesh is valid and watertight.")

2. Rotational Continuous Collision Detection (CCD)

Evaluate continuous swept trajectories of simultaneous translations and rotations:

import compas_forge

# Pose: [x, y, z, qx, qy, qz, qw]
pose_a_start = [-2.5, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]
pose_a_end   = [ 2.5, 0.0, 0.0, 0.0, 0.0, 0.707, 0.707] # Rotates 90 deg

result = compas_forge.check_swept_collision_zero_copy(
    mesh_a, pose_a_start, pose_a_end,
    mesh_b, pose_b_start, pose_b_end
)

if result["has_collision"]:
    print(f"Collision at Time of Impact: {result['time_of_impact']:.6f} s")

3. Assembly Contact Solver (Sutherland-Hodgman)

Extract exact contact polygons, centroids, and areas from touching blocks:

import compas_forge

blocks = {"block_0": mesh_0, "block_1": mesh_1}
contacts = compas_forge.compute_assembly_contacts_zero_copy(blocks, tolerance=0.005)

for contact in contacts:
    print(f"Contact between {contact['block_a']} and {contact['block_b']}")
    print(f"Contact Area: {contact['area_m2']} m² | Centroid: {contact['centroid']}")

---

CLI Usage Guide

# 1. Run general diagnostics
python -m compas_forge check my_mesh.json

# 2. Run continuous swept-collision
python -m compas_forge swept mesh_a.json -2.5,0,0,0,0,0,1 2.5,0,0,0,0,0.7,0.7 mesh_b.json 0,0,0,0,0,0,1 0,0,0,0,0,0,1

# 3. Resolve assembly contacts
python -m compas_forge contacts block_1.json block_2.json --tolerance 0.005

---

Examples & Reproducibility Sandbox

We provide ready-to-run educational scripts inside the examples/ directory to demonstrate and replicate our SOTA algorithms instantly:

1. Zero-Copy Mesh Healing (examples/example_zero_copy_healing.py): Repairs 1,000s of vertex duplicates and face normal directions in-memory and reconstructs a healthy COMPAS Mesh in milliseconds.
2. Rotational Sweep Collision (examples/example_continuous_collision.py): Runs continuous swept collision detection on two moving, rotating objects.
3. Voussoir Arch Assembly Solver (examples/example_assembly_contacts.py): Synthesizes a 30-block voussoir arch and solves 29 contact interfaces with centroids and normal vectors in milliseconds.

To run any example:

python examples/example_zero_copy_healing.py

To regenerate the scaling benchmark chart:

python generate_benchmarks.py

---

Mathematical Formulations & Algorithms

1. Mesh Volume via Gauss's Divergence Theorem

$$ V = \frac{1}{6} \sum_i \mathbf{p}_0 \cdot \left(\mathbf{p}_1 \times \mathbf{p}_2\right) $$

2. Best-Fit Newell Plane & Planarity Deviation

$$ n_x = \sum_{i=0}^{N-1} (y_i - y_{i+1})(z_i + z_{i+1}) $$ $$ n_y = \sum_{i=0}^{N-1} (z_i - z_{i+1})(x_i + x_{i+1}) $$ $$ n_z = \sum_{i=0}^{N-1} (x_i - x_{i+1})(y_i + y_{i+1}) $$

The maximum perpendicular distance $d_{max}$ of any vertex $\mathbf{v}i$ to the centroid plane $\mathbf{c}$ is evaluated as: $$ d{max} = \max_i \left|(\mathbf{v}_i - \mathbf{c}) \cdot \mathbf{n}\right| $$

3. Topological Invariants (Euler & Genus)

$$ \chi = V - E + F $$ $$ g = \frac{2 - \chi}{2} $$

---

Author

• GitHub: https://github.com/moaminmo90
• LinkedIn: https://www.linkedin.com/in/moaminmo90

---

License

Licensed under the MIT License.