Add a new attachment

An attachment is part of the benchmark only when its physical geometry, simulation model, task definition, and evaluation metadata agree.

For the ADL and bimanual open calls, contact the authors before starting the design. Read the submission requirements and credit policy alongside this technical guide.

1. Reuse the mounting interface

Start from the existing press-fit geometry in the CAD repository. Do not redraw the interface from measurements when an editable source is available.

The attachment should:

• seat fully in a standard honeycomb cell;
• resist the expected manipulation load;
• remain removable without damaging the base;
• avoid interfering with neighboring cells; and
• preserve a repeatable initial state.

For an attachment wider than one cell, document exactly which adjacent cells must remain free.

2. Define the task

Write the task definition before producing the final model:

• initial state;
• required manipulation sequence;
• observable success state;
• timeout;
• number and names of intermediate stages, if any;
• permitted reset procedure; and
• likely failure modes.

A success condition must be measurable across different platforms without requiring access to internal robot state.

3. Produce the physical files

Provide:

• editable CAD source;
• STL files with descriptive names;
• a complete bill of printed components;
• print orientation and support requirements;
• assembly instructions; and
• a photograph or render of the assembled mechanism.

If the mechanism requires non-printed hardware, list the exact specification and a generally available alternative.

4. Produce the articulated model

The simulation asset should include:

• visual and collision geometry;
• named joints and links;
• joint axes, limits, and initial values;
• mass and inertia values with their derivation;
• contact and friction assumptions; and
• a smoke test that drives the mechanism through its full range.

Keep identified parameters separate from nominal defaults when they depend on a particular printed instance.

5. Extend the protocol

Propose a stable lowercase attachment_id and supply the task metadata for review. Once the attachment is accepted into a versioned benchmark extension, coordinate the following updates with the maintainers:

• the attachment reference;
• the success and timeout table;
• stage numbering, if applicable;
• trials.csv and trials.xlsx;
• logging validation; and
• the reproducibility checklist.

The current benchmark still requires all 13 conditions, with five trials each. A proposed attachment does not replace an existing condition or change the current runner's required trial count.

6. Submit the contribution

Include physical and simulation evidence in the pull request:

• photograph of the assembled attachment;
• manual actuation through the full range;
• successful fit in the HiveBoard base;
• simulation load test; and
• one completed example trial row.