Supported Tasks#
The official task environments are bundled in the embodichain wheel under
the embodichain_tasks import package. Import-backed tasks register during
package discovery; configuration-defined Task Program tasks register when
their gym config is loaded. No second package installation is needed.
Run a task by passing one of its gym configuration files to the unified CLI:
embodichain run-env \
--gym_config embodichain_tasks/configs/tasks/manipulation/tableware/pour_water/task.cobotmagic.yaml
Use --preview to inspect a configured environment without starting a data
generation run. See Running Environments with run-env for all launch options and
Expert Data Generation for the dataset workflow.
Environment catalog#
Domain |
Environment ID |
Example gym config |
|---|---|---|
Classic control |
|
|
Manipulation |
|
|
Manipulation |
|
|
Manipulation |
|
|
Special |
|
|
Special |
|
|
Tableware |
|
|
Tableware |
|
|
Tableware |
|
|
Manipulation |
|
|
Manipulation |
|
|
Manipulation |
|
|
Tableware |
|
|
Tableware |
|
|
Tableware |
|
|
Tableware |
|
|
Tableware |
|
|
The value of id inside a conventional gym config must match a registered
environment ID. A supported configuration-defined Task Program deployment
declares environment.component,
task_program.{program,integration,execution_policy}, and
embodiment.component; loading it registers its free id against the
common EmbodiedEnv. A pure env.yaml component has environment_id
but no runnable id, so task discovery does not list it. Discovery is based
on the top-level id schema rather than an env* filename prefix.
When adding a task, update this catalog together with its runnable config.