How to Use Task

This tutorial guides you on how to run a task.

Pre-defined Tasks

The directory internutopia_extension/tasks/__init__.py contains a list of all our pre-defined tasks:

from internutopia_extension.tasks import (
    manipulation_task,
    finite_step_task,
    single_inference_task,
)

We can also review the configuration of each task in internutopia_extension/configs/tasks/__init__.py.

How to Use Task

To use an existing task within InternUtopia, you can simply use the corresponding type of task config in the runtime configuration as following:

from internutopia.core.config import Config, SimConfig
from internutopia.core.gym_env import Env
from internutopia.core.util import has_display
from internutopia.macros import gm
from internutopia_extension import import_extensions
from internutopia_extension.configs.tasks import SingleInferenceTaskCfg

import_extensions()

headless = False
if not has_display():
    headless = True

config = Config(
    simulator=SimConfig(physics_dt=1 / 240, rendering_dt=1 / 240, use_fabric=False, headless=headless, webrtc=headless),
    task_configs=[
        SingleInferenceTaskCfg(
            scene_asset_path=gm.ASSET_PATH + '/scenes/empty.usd',
        ),
    ],
)

env = Env(config)
obs, _ = env.reset()

i = 0

while env.simulation_app.is_running():
    i += 1
    obs, _, terminated, _, _ = env.step(action={})

    if i % 1000 == 0:
        print(i)

env.close()

In InternUtopia, task_configs is a list in which each element represents one episode. This is one of the key points that distinguish InternUtopia with other platforms.

In the above example, we use SingleInferenceTaskCfg to specify the type of task, which is the simplest task that never ends. Episode specific configs are defined in SingleInferenceTaskCfg, such as scene, robots and metrics. Only scene_asset_path is specified in the example so a task with an empty scene is running in simulation.