Hey everyone,
My team got tired of wrestling with Kubernetes and headless Docker containers just to run parallel simulations. We were acting like DevOps engineers instead of focusing on robotics.
So, we built a Python SDK (Hardsim) that turns headless simulation scaling into a simple API call. It handles the GPU provisioning, state tracking, and returning artifacts.
import hardsim as hs
# 1. Upload assets
robot_asset_id = hs.upload_input_asset("./assets/robot.urdf", asset_kind="robot")
scene_asset_id = hs.upload_input_asset("./assets/scene.usd", asset_kind="scene")
# 2. Trigger headless compute
job = hs.submit_assets(
robot_asset_id=robot_asset_id,
scene_asset_id=scene_asset_id,
robot_asset_type="urdf",
num_envs=64,
steps=2000
)
# 3. Wait and retrieve artifacts
result = hs.wait(job.job_id)
paths = hs.download(job.job_id, "./outputs")
You can check out the platform (and find the docs) here: hardlightsim.com
If anyone wants to run an actual workload and try to break the system, shoot me an email at sudhar@hardlightsim.com. I’ll manually drop some compute credits into your account so you can test the execution speed on us.
Curious, how are you all currently orchestrating your parallel sim workers?