Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.hardlightsim.com/llms.txt

Use this file to discover all available pages before exploring further.

Use Isaac Lab jobs when your workload needs a custom task package instead of the default simulation submit path.

​
What Changes Versus A Standard Simulation Job

Standard simulation submit is enough when you are sending a robot asset, a scene asset, and control/task configuration that the platform already understands. Use Isaac Lab jobs when you need to provide your own workload logic:
  • custom task code
  • custom rollout behavior
  • custom artifact production
  • Isaac Lab-specific task packaging
The submit path stays the same, but the payload adds:
  • runtime_profile="isaac_lab_rollout"
  • an isaac_lab block
  • a task package asset uploaded with asset_kind="isaac_lab_task_package"

​
Required Fields

An Isaac Lab submit must include:
  • runtime_profile="isaac_lab_rollout"
  • isaac_lab.task_module
  • isaac_lab.task_class_or_name
  • isaac_lab.task_package_asset_id or another staged task package reference
Your task package is responsible for writing the expected artifacts for the job.

​
Minimal SDK Example

import json
from pathlib import Path

import hardsim as hs

client = hs.HardsimClient.from_env()

robot_asset_id = client.upload_input_asset("./assets/franka.usdz", asset_kind="robot")
scene_asset_id = client.upload_input_asset("./assets/table_scene.usdz", asset_kind="scene")
task_package_asset_id = client.upload_input_asset(
    "./dist/my-isaac-lab-task.zip",
    asset_kind="isaac_lab_task_package",
)

job = client.submit_assets(
    robot_asset_id=robot_asset_id,
    scene_asset_id=scene_asset_id,
    robot_asset_type="usd",
    num_envs=1,
    steps=1024,
    physics_dt=0.005,
    substeps=2,
    runtime_profile="isaac_lab_rollout",
    isaac_lab={
        "task_module": "my_workloads.pick_and_lift",
        "task_class_or_name": "TabletopPickTask",
        "task_package_asset_id": task_package_asset_id,
        "task_args": {
            "force_video": True,
            "attach_distance_m": 0.08,
        },
    },
)

result = client.wait(job.job_id, poll_interval_s=2.0, timeout_s=1800.0, raise_on_error=False)
print(job.job_id, result["status"])

​
Artifact Expectations

For successful Isaac Lab jobs, the task package should produce:
  • rollout.zarr
  • render.mp4 when outputs.video=true
  • any task-specific logs you want customers to inspect
The worker also preserves structured diagnostics artifacts for failed jobs. See Artifacts and Diagnostics for the exact behavior.

​
What Customers Should Own

Hardsim runs the workload, stages assets, and returns artifacts. The workload author still owns:
  • task logic
  • robot and scene pairing
  • spawn pose and target selection
  • controller behavior
  • success criteria
If an Isaac Lab workload fails because the robot never reaches or grasps the target, that is typically a workload-definition issue, not a platform submit/runtime issue.

​
Recommendations

  • Start with a minimal task package and a simple workcell first.
  • Add outputs.video=true during bring-up so you can inspect behavior quickly.
  • Download runner.log, user_job.log, and command.stderr.log on failed jobs.
  • Move to complex warehouse scenes only after a simple tabletop workload behaves correctly.