Code of Honor

Last Updated: January 19, 2026

Purpose

As a global infrastructure, ManipulationNet hosts a wide range of manipulation benchmarks and evaluates robotic manipulation systems under standardized hardware kits and official evaluation software.

To preserve fairness, comparability, and scientific integrity, all participants must follow this Code of Honor.

By registering for or submitting results to ManipulationNet, you agree to the following principles.

Core Commitments

By submitting results to the benchmark leaderboard, a team affirms that:

1. Autonomy Classification Integrity: The team's submission is correctly labeled under the benchmark's autonomy category.
2. Evaluation Integrity: The standard hardware and evaluation software were used without alteration.
3. Disclosure Integrity: All disclosures made are truthful, complete, and non-misleading.

Autonomy Classification

Each submission must be reported under exactly one autonomy mode, consistent with how the system was evaluated.

Fully Autonomous

A submission is Fully Autonomous if:

• The robot completes the task without any human involvement during execution.
• All perception, state estimation, decision-making, planning, and control are performed by the system itself.
• No human input influences robot behavior once the run begins, except for: 1) emergency stops for safety; 2) mandatory physical resets explicitly required by the specific benchmark protocol.

Not permitted in Fully Autonomous mode:

• Teleoperation of any form.
• Human selection, confirmation, or correction of actions.
• Human-triggered retries, pauses, or decision-making.

Human in the Loop

A submission is Human-in-the-Loop if a human may provide high-level, non-kinematic assistance, such as:

• Task status annotation (e.g., success or failure).
• Triggering predefined autonomous recovery or termination routines.
• Selecting among a fixed set of system modes.

Not permitted in Human-in-the-Loop mode:

• Humans must not provide continuous or low-level control.
• Humans must not specify trajectories, poses, waypoints, joint commands, or velocities.
• Humans must not manually guide or correct robot motion, directly or indirectly.

If human input influences robot motion beyond these constraints, the submission must be classified as Teleoperation.

Teleoperation

A submission is Teleoperation if a human directly or indirectly controls robot actions during execution, including:

• Continuous or semi-continuous control of joints or end-effector motion.
• Manual correction of robot behavior.
• Human-driven perception–action loops.

Teleoperation submissions are permitted for specific benchmarks and will be labeled clearly on the leaderboard.

Hardware and Software Integrity

Participants must use the official ManipulationNet hardware kit exactly as provided or specified.

Participants must use the official ManipulationNet evaluation software without modification.

Disclosure for Open Science

ManipulationNet is designed to support open scientific exchange and cumulative progress.

Participants are strongly encouraged to accompany leaderboard submissions with:

• A paper, technical report, preprint, or workshop submission describing the method.
• Sufficient methodological detail to enable understanding and comparison by the community.
• Open sourced implementation, datasets and trained models.

Such accompanying materials:

• Improve interpretability of leaderboard results.
• Enable deeper technical discussion and reproducibility.
• May be linked directly from the leaderboard when available.

Disclosure of a submission is not required, but is highly encouraged as part of responsible academic benchmarking.

Fair Use and Enforcement

Violations of this Code of Honor may result in:

• Reclassification of autonomy mode.
• Removal from the leaderboard.
• Temporary or permanent disqualification.
• Public correction or retraction if necessary.

Affirmation

By participating in ManipulationNet, teams affirm that:

• Results are reported honestly and transparently.
• Benchmark hardware and evaluation software were used without modification.
• Submissions comply with both the letter and spirit of this Code of Honor.

Contact Information

For questions regarding the Code of Honor, please contact support@manipulation-net.org