LinkForge Core

The platform-independent Intermediate Representation (IR) engine for robot descriptions.

What Is LinkForge Core?

Writing and maintaining URDF or SRDF by hand is fragile: inertia values are guessed, collision geometries drift, and physics bugs surface only after a simulator crash (or worse, on hardware). LinkForge Core solves this by treating your robot as source code with physical constraints, not a static XML document.

It provides a mathematically rigorous, zero-dependency Intermediate Representation (IR) engine with hardened physical validation, scientific inertia solvers (Mirtich / Sylvester), and lossless round-trip translation between URDF, XACRO, and SRDF.

Installation

pip install linkforge-core

Zero external dependencies. No Blender, no ROS installation, no C++ compilation required.

Why LinkForge Core?

• Physically Guaranteed Sim Stability: Zero-mass links or unphysical inertia tensors cause simulators like Gazebo or Isaac Sim to crash. LinkForge Core uses the Mirtich algorithm (Divergence Theorem) to calculate exact inertia properties from geometries, validated against Sylvester's Criterion to ensure physical validity.
• Standardized & Namespaced Assembly: Easily compile complex robots, merge multiple sub-assemblies (e.g. attaching a gripper to an arm), and apply joint prefixing and limits programmatically using the fluent Composer API.
• Hardened Sandboxed Security: Safely parse untrusted third-party robot descriptions. LinkForge Core blocks path-traversal attacks and restrains file reading to designated package boundaries.
• Light & Portable: Zero external dependencies. No C++ compilation required, making it highly portable across standard Python environments, CI/CD pipelines, and HPC clusters.

Quickstart

LinkForge Core exposes a flat, curated public API. No nested import paths required.

from linkforge.core import RobotBuilder, box, cylinder

# Initialize the assembly builder
builder = RobotBuilder("forge_arm")

# Define the base link (root of the robot)
builder.link("base_link") \
    .visual(box(0.2, 0.2, 0.1)) \
    .collision() \
    .mass(5.0) \
    .root()

# Define and connect the upper arm link via a revolute joint
builder.link("upper_arm", parent="base_link") \
    .visual(cylinder(0.05, 0.8)) \
    .collision() \
    .mass(2.5) \
    .revolute(
        axis=(0, 0, 1),
        limits=(-3.14, 3.14),
        effort=50.0,
        velocity=2.0
    ) \
    .commit()

# Compile to production-ready, validated URDF XML
urdf_xml = builder.export_urdf()

Key Capabilities

Parse, Validate & Compile (Full Round-Trip)

Ingest existing URDF, XACRO, or SRDF files into the IR, validate them, and compile back out to any target format without data loss:

from linkforge.core import read_urdf, validate_robot, write_urdf, write_xacro

# Ingest and auto-resolve package:// paths
robot = read_urdf("my_robot.urdf")

# Perform kinematic, structural, and physical checks
result = validate_robot(robot)

if result.is_valid:
    # Compile back to URDF or XACRO
    write_urdf(robot, "my_robot_validated.urdf")
    write_xacro(robot, "my_robot.xacro")
else:
    for issue in result.errors:
        print(f"  [{issue.code.name}] {issue.message} on {issue.affected_objects}")

MoveIt 2 & SRDF Semantic Composition

Programmatically compose MoveIt 2 planning groups, end-effectors, and self-collision matrices for SRDF without manually editing XML:

from linkforge.core import RobotBuilder, read_srdf, write_srdf

builder = RobotBuilder("my_arm")
# ... define links and joints ...

# Add a kinematic planning group via base/tip chain shorthand
builder.semantic.group(
    "arm",
    base_link="base_link",
    tip_link="wrist_link",
)

# Define an end-effector
builder.semantic.end_effector(
    name="gripper",
    group="gripper_group",
    parent_link="wrist_link",
    parent_group="arm",
)

# Export both URDF + SRDF in one pass
urdf_xml = builder.export_urdf()
srdf_xml = builder.export_srdf()

# Or parse and re-export an existing SRDF
semantic = read_srdf("my_robot.srdf")
write_srdf(semantic, "my_robot_updated.srdf")

Exact Solid-Body Inertia Solver

Compute principal moments of inertia and Center of Mass offsets for primitives or complex triangle meshes, hardened with local origin conditioning for floating-point accuracy:

from linkforge.core import Box, Vector3, calculate_inertia

geometry = Box(size=Vector3(1.0, 0.5, 0.3))
# Automatically computes exact ixx, iyy, izz principal moments
inertia = calculate_inertia(geometry, mass=10.0)

Sensor Suite

Define cameras, LiDAR, IMU, GPS, and force/torque sensors with configurable noise models directly in the IR:

from linkforge.core import RobotBuilder

builder = RobotBuilder("sensor_bot")
# ... define links ...

# Add a LiDAR sensor to a link
builder.link("lidar_link", parent="base_link") \
    .lidar("front_lidar", range_min=0.1, range_max=30.0, samples=360) \
    .commit()

# Add a camera sensor
builder.link("camera_link", parent="base_link") \
    .camera("front_camera", width=1280, height=720) \
    .commit()

Headless Use in CI / ML Pipelines

linkforge-core has zero GUI dependencies, making it ideal for headless environments:

# ci_validate.py — run in any CI/CD pipeline or HPC cluster
from linkforge.core import read_urdf, validate_robot
import sys

robot = read_urdf("robot.urdf")
result = validate_robot(robot)

if not result.is_valid:
    for issue in result.errors:
        print(f"ERROR: [{issue.code.name}] {issue.message}")
    sys.exit(1)

print("✓ Robot model validated successfully.")

# In your CI pipeline:
pip install linkforge-core
python ci_validate.py

Resources & Documentation

• Extensive Documentation: Read the tutorials and how-to guides at linkforge.readthedocs.io.
• Open Source Repository: View source, open issues, and join discussions on GitHub.
• License: Standard open-source Apache-2.0 License.