ioailab
ioailab provides G1 robot cfgs, an IsaacLab-style task registry, action/sensor helpers, dataset refs, and action agents for IsaacLab.
make shell-gui
python examples/01_collect.py
Runtime boundary —
import ioailabis side-effect free. Task IDs are registered only through explicit imports underioailab.tasks; IsaacLab owns app launch, env construction, managers, sensors, PhysX, and stepping.
Rendered scenes
Stack cube
The loop
Every workflow is the same explicit loop: build an env, build an agent, step,
and records data through env.collect(...).
from ioailab.agents import CuroboPlannerAgent
from ioailab.envs import make_env
task_id = "GalbotG1-PickCube-v0"
env = make_env(task_id, num_envs=1)
agent = CuroboPlannerAgent.from_task(task_id)
dataset = env.collect(
agent=agent,
episodes=1,
path="data/pick_cube_demos.hdf5",
)
CuroboPlannerAgent, PolicyAgent, TeleopAgent, and TaskFlowAgent are
interchangeable at this boundary. env.collect(...) and env.evaluate(...)
remain convenience helpers for batch data generation and metrics.
Current surface
| Area | Surface |
|---|---|
| Robot cfgs | G1 articulation, arm/leg/gripper/base actions, camera activation, tensor helpers |
| Tasks | GalbotG1-PickCube-v0, -StackCube-v0, -Reach-v0, -BaseNav-v0, -PickToShelf-v0, -PickToShelf-Pick/Nav/Place-v0, -SortToShelf-v0, plus teleop/Mimic/policy variants |
| Data path | HDF5 collection, Mimic expansion, robomimic Diffusion Policy training, evaluation |
| Teleop | GP001 left-wrist/front-head RGB collection with keep/drop/exit review |
| Planning | cuRobo v2 (curobov2) agents that emit IsaacLab action tensors |
Where to go next
| Need | Page |
|---|---|
| Build a task end-to-end | Tutorial |
| Run the numbered examples | Examples |
| Understand the boundary | Architecture |
| Compose agents and task flows | Action Agents & Task Flows |
| Look up task IDs | Tasks |
| Collect, Mimic, train, export | Data & Datasets |
| Configure cameras | Sensors |
| Joint helpers and assets | Robot Reference |
| Work inside Docker | Developer Workflow |
AGENTS.md and Architecture are the design source of truth.