RoboDojo Submission

Prepare a policy, open the required pull request, and submit the official evaluation application. This page explains what to include at each stage and what evaluators need to reproduce your result.

Submit an evaluation Open the RoboDojo application Read the contribution guide Review PR requirements in CONTRIBUTING.md

Before you submit

Your policy should already work with XPolicyLab’s standard adapter interface and reproduce the results documented in its policy README. Check these items first:

The policy lives under policy/<POLICY>/ and includes the required adapter files.
The policy README documents installation, checkpoints, data format, evaluation commands, and expected results.
The policy completes a local or debug evaluation with the intended checkpoint.
The pull request passes the checks described in CONTRIBUTING.md.
The checkpoint or download instructions needed to reproduce the submission are available to the evaluators.

If the model requires credentials, a private artifact, or a special environment variable, explain the access path in the application.

1. Open a pull request

Fork or clone XPolicyLab, add your policy under policy/<POLICY>/, and open a pull request following CONTRIBUTING.md.

For a new integration, use the minimal reference adapter as a starting point:

Keep the pull request focused. A reviewer should be able to find the policy entrypoint, environment requirements, checkpoint source, and evaluation command without reconstructing them from unrelated changes.

2. Submit the commit and evaluation application

Once the pull request is open, submit the exact commit you want evaluated through the official evaluation application. Include:

the XPolicyLab pull request URL and commit SHA;
the policy name and checkpoint to evaluate;
the benchmark, tasks, robot configuration, and action space;
required environment variables, model downloads, or access instructions; and
contact information for evaluation questions.

Use one application for the exact policy and checkpoint you want evaluated. If you update the adapter or checkpoint after submitting the application, mention the new commit or submit an updated application.

3. Review and leaderboard entry

The evaluation team runs the requested benchmark setup, then reviews the submitted code and reproducibility details. Once the evaluation and code review are complete, the result can be added to the RoboDojo leaderboard according to the benchmark’s submission rules.

For RoboDojo task definitions and result interpretation, see Quick Evaluation. For policy-side deployment and data contracts, see XPolicyLab.