ACDC4Robot: Automated Conversion of Description Conventions for Robots from Design to Learning

Robot description formats encode the geometry, kinematics, and physical properties required for simulation, visualization, and planning. ACDC4Robot is a Fusion 360 Add-In that exports these descriptions directly from a robot CAD assembly.

The Add-In exports URDF, SDFormat (SDF), and MJCF, and also exposes an experimental URDF+ option. Version 1.1.1 has its strongest current validation on the Fusion→MJCF→MuJoCo path. URDF, SDF, URDF+, and target-simulator behavior should be independently checked for the intended application.

The companion ACDC4Robot RobotLibrary provides reusable Fusion 360 robot models for design, simulation, and learning.

• ACDC4Robot v1.1.1 is the current supported GitHub release. Download the installable plugin and checksum from the v1.1.1 release page.
• The main branch now contains the v1.2.0 development candidate, adding Simplified Chinese UI localization contributed through pull request #14. It is available for live Fusion testing before its GitHub Release is cut.
• The Fusion→MJCF→MuJoCo path now preserves repeated occurrence identities, runs a fail-closed assembly preflight, records an export provenance report, and gives actionable errors for grounded endpoints, missing joint origins, unsupported joint types, disconnected assemblies, and unsupported geometry.
• GitHub Releases are the canonical update channel. We do not currently plan an Autodesk App Store update, so the App Store build may be older.
• Reusable robot designs are maintained separately in the ACDC4Robot RobotLibrary.

• expand live Fusion regression fixtures and automated export checks;
• improve format-specific URDF, SDF, and simulator compatibility without weakening the current MJCF validation contract;
• investigate broader nested-assembly support after reproducible fixtures are available;
• add an optional, reproducible collision-mesh workflow for contact-rich applications while keeping visual and collision assets distinct;
• continue improving documentation, examples, and community-contributed robot models through this repository and the RobotLibrary.

These are roadmap directions, not committed release dates. Issues and pull requests with reproducible Fusion fixtures are especially helpful.

Export robot description files directly from a Fusion 360 design using a GUI panel.

• Supported robot description formats:
  • URDF (Unified Robotics Description Format)
  • SDFormat (Simulation Description Format) or SDF
  • MJCF (MuJoCo Format)
  • URDF+ (experimental; not part of the v1.1.1 regression scope)
• A companion robot model library containing various robot types:
  • Robot Arm
  • Gripper
  • Mobile Robot
  • Quadruped Robot
  • Humanoid

Install the Add-In from GitHub Releases (recommended), from the legacy Autodesk App Store package, or manually from source.

Download the current installable ZIP and checksum from GitHub Releases. GitHub is the supported update channel and contains fixes that may not be present in the older Autodesk App Store package. For v1.1.1, download ACDC4Robot-1.1.1.zip and its .sha256 file, verify the checksum, unzip it, and copy the resulting top-level ACDC4Robot folder into Fusion's Add-Ins directory. Do not copy the ZIP itself or create a doubly nested ACDC4Robot/ACDC4Robot directory.

Verify the downloaded ZIP on macOS/Linux with:

shasum -a 256 ACDC4Robot-1.1.1.zip

On Windows PowerShell, use:

Get-FileHash .\ACDC4Robot-1.1.1.zip -Algorithm SHA256

An earlier Windows/macOS package remains available from the Autodesk Fusion App Store, but no immediate App Store update is planned. Prefer the GitHub release for current behavior and reproducible checksums.

Clone or download this repository and copy the Add-IN/ACDC4Robot folder into Fusion 360's Add-Ins directory, which is shown under Preferences -> General -> API -> Default Path for Scripts and Add-Ins.

Release maintainers can create a deterministic, directly installable archive with python3 scripts/build_addin_release.py. Unzip it and copy the resulting ACDC4Robot folder into the same add-in directory.

The default locations are:

Windows:

%appdata%\Autodesk\Autodesk Fusion 360\API\AddIns

Mac:

$HOME/Library/Application Support/Autodesk/Autodesk Fusion 360/API/AddIns

If Fusion uses a customized location, use the path reported under Preferences -> General -> API -> Default Path for Scripts and Add-Ins.

Restart Fusion after installing or replacing the Add-In. The v1.1.1 manifest enables Run on Startup; if ACDC4Robot is not running, press Shift+S or open UTILITIES -> Add-Ins -> Scripts and Add-Ins, find ACDC4Robot under Add-Ins -> My Add-Ins, and click Run.

The promoted ACDC4Robot command appears in the Design workspace's UTILITIES -> Add-Ins panel beside Scripts and Add-Ins. Click the command to begin exporting the active design.

The MJCF path runs a fail-closed assembly preflight before writing files, preserves repeated occurrence names, supports standard and as-built joints, and writes an acdc4robot-export-report.json provenance report. Version 1.1.1 also reports grounded endpoints, missing joint origins, and unsupported joint types with corrective messages instead of secondary Python failures. See the MJCF export guide for the supported assembly structure, verification status, and the deliberate boundary between CAD export and application-specific actuator/contact modeling.

The v1.2.0 development candidate follows Fusion's user-interface language and supports English and Simplified Chinese. Export-format and simulator choices remain stable technical values (URDF, SDFormat, MJCF, URDF+, Gazebo, PyBullet, and MuJoCo) regardless of the displayed language. Unsupported languages fall back to English. The machine-readable acdc4robot-export-report.json remains English so reports are reproducible and can be compared across computers; only user-facing labels and messages are localized.

An incompatible format/target selection is rejected with a Fusion message.

The MJCF exporter reads Fusion's internal centimetre-based API values and converts model quantities to SI units; the document's display unit does not need to be metres. Parametric and direct-modeling documents are both accepted, so disabling design history is not a prerequisite. Running the exporter on a copy remains good practice when preparing or restructuring an assembly.

• Use a flat top-level occurrence structure. Each visible occurrence must contain solid BRep geometry; nested occurrences, surface-only components, and mesh-only components fail the MJCF preflight.
• Connect the visible occurrences as one tree with rigid, revolute, or slider joints. A joint cannot connect directly to Fusion root/ground; represent the fixed structure as a component such as base_link.
• Give occurrences and joints unique names. Moving joints require a valid Fusion joint origin.
• For a self-contained design, use Break Link to make an external component internal. MJCF 1.1.1 can export referenced occurrences, but records a warning because their source revisions must be frozen for reproducibility.
• Repeated occurrences are supported by MJCF 1.1.1 and receive distinct export names. Other export formats have not been revalidated for this behavior; use Make Independent if you encounter a format-specific problem.
• Make sure all components are named with alphanumeric characters, underscores _, or hyphens -. Other characters may cause compatibility problems.
• See the Fusion 360 assembly instructions and the stricter MJCF export guide before exporting.

1. Open the prepared design in Fusion's Design workspace and click ACDC4Robot.
2. Select a description format and target environment. For the validated v1.1.1 path, select MJCF and MuJoCo.
3. Choose a dedicated parent output folder. ACDC4Robot creates or reuses a sanitized document-name subfolder, so use a clean destination to avoid mixing files from an earlier export.
4. For MJCF, correct every preflight error and export again. A successful export contains <model-name>.xml, STL visual meshes, and acdc4robot-export-report.json.
5. Load the exported model in MuJoCo and review the report, geometry, inertias, joint axes, and limits. Add actuators, sensors, joint dynamics, and explicit collision geometry in a reviewed downstream MJCF layer; ACDC4Robot does not infer them from CAD.

The examples below document earlier URDF/SDF workflows. They have not been revalidated as part of the v1.1.1 MJCF regression suite and should be tested in the current target simulator before use. The SDF closed-chain example does not imply MJCF closed-chain support; the v1.1.1 MJCF preflight requires one connected tree.

• Historical Gazebo test showing an SDFormat closed-chain mechanism

• Historical UR5e test in Gazebo

• Historical Robotiq 2F-85 gripper test in Gazebo

🤖 ACDC4Robot RobotLibrary

The companion repository contains reusable Fusion 360 robot models intended to reduce repeated assembly work and support exporter examples. Contributions of well-documented models, source revisions, and validated exports are welcome. The legacy in-repository index is retained for historical context.

If ACDC4Robot or the RobotLibrary supports your research, teaching, or robot development workflow, please cite the peer-reviewed paper:

• IEEE Xplore
• DOI: 10.1109/ICARM62033.2024.10715835
• arXiv preprint: 2312.12295

@inproceedings{qiu2024describing,
  author    = {Qiu, Nuofan and Song, Chaoyang and Wan, Fang},
  title     = {Describing Robots from Design to Learning: Towards an
               Interactive Lifecycle Representation of Robots},
  booktitle = {2024 International Conference on Advanced Robotics and
               Mechatronics (ICARM)},
  year      = {2024},
  pages     = {1081--1086},
  doi       = {10.1109/ICARM62033.2024.10715835},
  url       = {https://doi.org/10.1109/ICARM62033.2024.10715835}
}

N. Qiu, C. Song, and F. Wan, “Describing Robots from Design to Learning:
Towards an Interactive Lifecycle Representation of Robots,” in 2024
International Conference on Advanced Robotics and Mechatronics (ICARM),
Tokyo, Japan, 2024, pp. 1081–1086,
doi: 10.1109/ICARM62033.2024.10715835.

• Fusion2PyBullet
• fusion2urdf