Spaces showcase tasks for Isaac Lab

This project is a showcase for the definition and use of the different observation and action spaces available at Isaac Lab.

Currently, only the Isaac Lab's Direct workflow supports the definition of other spaces for observation and action other than Box. See Isaac Lab's documentation for more information.

Models defined for the tasks are just an example of what can be done with skrl, visit skrl's Model instantiators for more details.

Setup

Install (optional, because the scripts will automatically add the project to the PYTHONPATH if it is not installed).

python -m pip install -e source/spaces_showcase

List available tasks.

python scripts/list_envs.py

Showcased spaces

Cartpole task

Train a task for a specific observation and action space type.

python scripts/skrl/train.py --task Spaces-Showcase-<OBSERVATION>-<ACTION>-Cartpole-Direct-v0 --headless

		action space
		Box	Discrete	MultiDiscrete (2)
observation space	Box	x	x	x
	Discrete (1)	x	x	x
	MultiDiscrete (1)	x	x	x
	Dict	x	x	x
	Tuple	x	x	x

Cartpole-Camera task

Train a task for a specific observation and action space type.

python scripts/skrl/train.py --task Spaces-Showcase-<OBSERVATION>-<ACTION>-Cartpole-Camera-Direct-v0 --enable_cameras --headless

		action space
		Box	Discrete	MultiDiscrete (2)
observation space	Box	x	x	x
	Discrete (1)	-	-	-
	MultiDiscrete (1)	-	-	-
	Dict	x	x	x
	Tuple	in progress	in progress	in progress

---

Notes: The following features are currently supported only on the skrl's develop branch:

⁽¹⁾ One-hot encoding of Discrete and MultiDiscrete observation spaces.

⁽²⁾ MultiDiscrete action space (MultiCategoricalMixin model instantiator).