Robot developers face three distinct compute challenges: training policies on data‑center GPUs, running simulation and sensor rendering on workstation‑class RTX hardware, and validating on edge devices such as Jetson AGX Thor. Each tier traditionally uses its own cluster, scheduler and custom glue scripts, creating friction when moving work between environments. NVIDIA OSMO solves this fragmentation by providing a single Kubernetes‑native workflow orchestrator that lets teams describe the entire pipeline in one YAML file and run it unchanged across all tiers.
OSMO is open‑source under the Apache‑2.0 license, ships Helm charts and container images on NGC, and offers a local quickstart that runs the full control plane on a workstation using KIND. Users register each compute tier as a Kubernetes cluster through the CLI; workflows reference logical platforms like gb200, rtx‑pro‑6000 or jetson‑agx‑thor instead of specific cluster names. The control plane routes tasks to the appropriate pool using the NVIDIA KAI Scheduler, which supports NVLink‑aware placement, per‑group timeouts and priority‑based preemption.
Key capabilities include portability across laptops, cloud EKS/AKS/GKE clusters and on‑premise or air‑gapped environments; interactive development features such as VS Code or Jupyter sessions, remote exec, port‑forward and bidirectional rsync with progress bars; robust scheduling with topology‑aware GPU allocation; built‑in security through RBAC, OAuth2 device‑code login, TLS termination and cloud workload identity; and agent integration that allows coding assistants to submit, monitor and debug pipelines.
Dataset handling has shifted to workflow‑managed outputs; the legacy osmo dataset CLI and API are deprecated and will be removed in a future release, so teams should migrate to the new approach.
OSMO has been battle‑tested on projects such as GR00T, Isaac Lab, Isaac Sim and Isaac ROS, with integrations reported for Azure and Nebius infrastructures.
AI Robotics DevOps Kubernetes OSMO NVIDIA #AI #Robotics #DevOps #Kubernetes #OSMO #NVIDIA