|

NVIDIA Open-Sources OSMO: One YAML Orchestrates Physical AI Training, Simulation, and Robot Testing

ByRicardo

Robot builders do not need one compute downside. They have 3. A coverage is skilled on GB200 or H100 clusters, examined in Isaac Sim on RTX GPUs, then validated on a Jetson mounted inside an actual robotic. Each tier has its personal cluster, its personal scheduler, and its personal glue scripts. NVIDIA OSMO is NVIDIA’s reply to that fragmentation: an open-source, Kubernetes-native workflow orchestrator that lets a crew describe the entire pipeline in a single YAML file and run it throughout each tier with out touching infrastructure code.

Deployable? Yes. OSMO is Apache-2.0 licensed, ships Helm charts and containers on NGC, and has a local quickstart that runs the complete management airplane on a workstation with KIND.

The Three Computer Problem

NVIDIA frames bodily AI as a three computer problem. Training occurs on data-center GPUs. Simulation, physics, and sensor rendering occur on workstation-class RTX {hardware}. Deployment and hardware-in-the-loop (HIL) testing occur on edge units equivalent to Jetson AGX Thor, normally on premises. Each tier normally will get its personal tooling, and the handoffs are the place customized scripts accumulate.

OSMO treats all 3 as backends of 1 management airplane. Each backend is a Kubernetes cluster registered by the CLI. Workflows by no means title a cluster. They title a platform (for instance gb200, rtx-pro-6000, or jetson-agx-thor) and OSMO routes the duty to a pool that provides it.

What a Workflow Looks Liake

The Repo’s canonical instance is 3 duties cAhained by information:A

  • simulation runs an Isaac Sim container on rtx-pro-6000
  • train-policy runs a PyTorch container on gb200 with 8 GPUs, taking the simulation job’s output as enter
  • evaluate-thor runs a ROS app on jetson-agx-thor, consuming the skilled coverage and writing outcomes to a named dataset

Dependencies come from inputs, persistence from outputs, and placement from platform. The user guide covers serial and parallel job teams, Jinja templating for parameterized workflows, retry insurance policies, and HIGH/NORMAL/LOW priorities with preemption and GPU borrowing throughout swimming pools.

Key Capabilities

  • Portability: The similar YAML runs on a laptop computer (Docker/KIND) or on EKS, AKS, GKE, on-premise, or air-gapped clusters. The 6.3.0 launch added a multi-provider deploy-k8s.sh that provisions OSMO on Azure AKS, AWS EKS, microk8s, or any current cluster, with storage wiring for MinIO, Azure Blob, AWS S3, or bring-your-own S3.
  • Interactive improvement: Developers can launch VS Code, Jupyter, or SSH classes on a distant GPU node, exec into operating duties, port-forward providers, and rsync recordsdata in each instructions. Version 6.3.0 added osmo workflow rsync obtain with a stay progress bar.
  • Scheduling: OSMO makes use of the NVIDIA KAI Scheduler by default. Release 6.2.8 added NVLink topology-aware placement for multi-GPU duties. Release 6.3.0 made exec_timeout and queue_timeout per group, so a stalled simulation group not kills sibling coaching teams.
  • Data: The mission describes content-addressable datasets with deduplication that it claims can reduce storage by 10 to 100x. Note that the standalone osmo dataset CLI and /datasets API had been deprecated in 6.3.0 and are slated for elimination in 6.4, with workflow-managed dataset outputs because the alternative.
  • Security and identification: Since 6.2.8 OSMO ships an RBAC authorization sidecar, OAuth2 proxy integration with device-code login, and identity-provider person mapping. Release 6.3.0 added TLS termination on the Envoy gateway and cloud workload identification (Azure Workload Identity, AWS IRSA/Pod Identity) so providers not mount storage keys as Kubernetes Secrets. Release 6.3.1 tightened the default osmo-user function to the default pool.
  • Agent integration: The repository ships an AGENTS.md, a skills listing, and an MCP deployment guide. At GTC 2026 NVIDIA mentioned OSMO integrates with Claude Code, OpenAI Codex, and Cursor so coding brokers can submit, monitor, and debug pipelines.

Interactive Explainer: Watch OSMO Route One Workflow Across 3 Compute Tiers

Press Run workflow to see how OSMO schedules the README instance job by job. Click any tier or step quantity to examine what occurs there.

Key Takeaways

  • OSMO orchestrates coaching, simulation, and edge HIL duties from one YAML throughout heterogeneous Kubernetes clusters
  • Apache-2.0, Helm charts on NGC, newest launch 6.3.1 (June 2026), native KIND quickstart accessible
  • KAI Scheduler default, NVLink-aware placement, per-group timeouts, RBAC and OAuth2 in-built
  • Battle-tested on GR00T, Isaac Lab, Isaac Sim, and Isaac ROS; Azure and Nebius integrations exist
  • Dataset CLI is deprecated in 6.3 and eliminated in 6.4; plan migrations accordingly

Check out the GitHub, Documentation, Releases, Cookbook and NVIDIA OSMO page. All credit score goes to the researcher of this mission. Also, be at liberty to comply with us on Twitter and don’t neglect to affix our 150k+ML SubReddit and Subscribe to our Newsletter. Wait! are you on telegram? now you can join us on telegram as well.

Need to companion with us for selling your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar and so forth.? Connect with us

The put up NVIDIA Open-Sources OSMO: One YAML Orchestrates Physical AI Training, Simulation, and Robot Testing appeared first on MarkTechPost.

Similar Posts