Environments

Robot environments

Complete the Synthetic first episode before installing a simulator. ORA stays in its locked host environment; each simulator runs under its own explicit interpreter.

Environment Use it for Start here
LIBERO / LIBERO-Pro Panda tasks and source-paired robustness cases Install and run one task
RoboTwin Dual-arm manipulation in clean or randomized scenes Install and run one task
RoboDojo Isaac Sim manipulation tasks Install and run one task
RoboCasa365 Kitchen tasks for a mobile manipulator (PandaOmron), MuJoCo on the CPU Install and run one task
RoboLab Isaac Sim tabletop tasks for a DROID Franka with a Robotiq gripper Install and run one task

On a single-GPU Linux machine, ora setup --env and ora run --env do the whole sequence below for one task; see Run ORA on one GPU.

Setup sequence

  1. Install the pinned simulator and assets using its guide.
  2. Register the host and required perception services using common setup.
  3. Run environment-specific doctor discovery.
  4. Obtain the source-bound controller, seed, service, and provider qualification required by the task.
  5. Compose a fresh RunSpec, inspect it, and explicitly launch the episode.

Installation is not qualification. Each environment guide covers interpreter variables, source pins, assets, and one-task commands. The support matrix records what was checked.

Results

Only the environment's official boundary supplies the manipulation verdict. A model's statement that it finished is not a success label. Synthetic reports official_success: null and has no benchmark score.

Source: docs/environments/README.md