Usage
This section explains how to prepare RoboDojo for simulation evaluation. RoboDojo provides the simulator client, benchmark tasks, assets/config validation, Docker runtime, and result artifacts; policy dependencies, checkpoints, and servers live in XPolicyLab or in your own policy stack.
RoboDojo usage is organized around a typical evaluation workflow:
- Install the RoboDojo codebase and required simulation dependencies, either natively or with the optional Docker path.
- Download assets such as objects, scenes, materials, and robot embodiments.
- Configure the simulator, scene, cameras, robot, observation fields, and benchmark
env_cfg. - Run a policy through the XPolicyLab eval path, split server/client eval, Docker simulator client, smoke test, or benchmark sweep.
- Improve robustness and throughput with the Simulation Tasks tools for domain randomization and parallel environments.
After the environment is ready, continue to the task catalogs to choose task for simulation evaluation.
| Page | What it covers |
|---|---|
| Install & Download | Native dependency installation, repository setup, asset download, embodiment path updates, and optional Docker installation. |
| XPolicyLab | Policy integration, adapter contracts, deployment flow, data format, and checks. |
| Quick Evaluation | How to run eval, split server / client, smoke tests, benchmark sweeps, task protocol, and result summarization. |
| Configurations | How RoboDojo combines simulation, scene, robot, camera, observation, and benchmark env_cfg settings. |
Most experiments begin with Install & Download, XPolicyLab, and Quick Evaluation, then tune Configurations and add domain randomization or parallel environments once the baseline setup is working.