Supported Tasks#
Task environments combine a simulation scene, robot, and task logic behind a Gymnasium-compatible interface.
The official task environments are bundled in the embodichain wheel under
the embodichain_tasks import package. This section follows the task-first
layout under embodichain_tasks/configs/tasks so that documentation,
configuration paths, and list-task output use the same hierarchy.
Find a Task#
Use the catalog commands to inspect the installed task set:
embodichain list-task
embodichain list-task --category manipulation
embodichain show-task embodichain_tasks:repeated_pick_place
Preview a task with its runnable JSON or YAML configuration:
embodichain run-task \
--gym_config embodichain_tasks/configs/tasks/manipulation/repeated_pick_place/task.ur5.yaml \
--preview
Configuration-backed environments run through embodichain run-task. Tasks
that exist only as learning environments or Python registrations are identified
on their category page. See Task Catalog for catalog metadata and
static gallery export, and Running Tasks with run-task for launch options.
See Reinforcement Learning Training to train an RL policy.
Task hierarchy#
Domain |
Configuration hierarchy |
Scope |
|---|---|---|
|
Compact control and learning environments. |
|
|
Velocity-tracking tasks for legged robots. |
|
|
RL, Task Program, and environment-only manipulation tasks. |
|
|
Standalone examples and specialized differentiable environments. |
The top-level id in a runnable JSON or YAML configuration is the Gym
environment ID. A reusable env.yaml component may instead expose only
environment_id and is not runnable by itself. Import-backed tasks register
during package discovery; configuration-defined Task Program tasks register
when their runnable task configuration is loaded.
Create a Task#
Start from the
task template.
Use an EmbodiedEnv subclass for custom Python behavior, or compose a
configuration-defined Task Program from reusable components. Third-party
task packages are discovered through the embodichain.tasks entry point.
See Embodied Environments for environment authoring and
Configure and Run an Embodied Task Program for Task Programs. The
official task package README
describes package layout and registration.