Usage
Use this section as the practical path from a fresh checkout to a running RoboDojo evaluation.
RoboDojo owns the simulator side: benchmark tasks, assets, environment configs, evaluation client, Docker runtime, and result files. XPolicyLab owns the policy side: model dependencies, checkpoints, policy servers, adapters, and the observation/action contract between model and environment.
Recommended path
Section titled “Recommended path”- Start with Install & Download to install RoboDojo, download assets, update embodiment paths, and optionally download datasets.
- Read XPolicyLab if you need to run, adapt, train, or deploy a policy.
- Use Quick Evaluation to run a smoke test, a single task, a capability-dimension sweep, split server/client evaluation, Docker client evaluation, or a complete benchmark.
- Check Configurations when you need to change robot setup, cameras, scenes, observation fields, or evaluation configs.
- After the baseline works, choose tasks from the simulation and real-task catalogs, then add domain randomization or parallel environments when needed.
Official leaderboard evaluation
Section titled “Official leaderboard evaluation”If you would like to submit your policy for evaluation and inclusion on the official RoboDojo leaderboard, please complete the official evaluation application.
| Page | Use it for |
|---|---|
| Install & Download | Native install, assets, RoboDojo datasets (including HDF5 format), embodiment path updates, and optional Docker setup. |
| XPolicyLab | Policy adapter concepts, integrated policies, data/checkpoint layout, deployment flow, and model integration. |
| Quick Evaluation | Smoke tests, single-task eval, capability-dimension and full benchmark sweeps, split server/client mode, Docker client mode, and result outputs. |
| Configurations | How env_cfg combines simulation, scene, robot, camera, observation, and benchmark settings. |