NVIDIA open-sources OSMO for robotics workflows across compute clusters
NVIDIA has released OSMO as an open-source, Kubernetes-native orchestrator for physical AI workflows. The system is designed to coordinate model training on data-center GPUs, simulation on workstation-class RTX hardware, and deployment or hardware-in-the-loop testing on edge devices such as Jetson AGX Thor. Teams can describe these stages in one YAML file rather than maintaining separate schedulers and integration scripts.
OSMO registers Kubernetes clusters as backends and routes tasks according to platform labels, including gb200, rtx-pro-6000, and jetson-agx-thor. A sample workflow links an Isaac Sim task, PyTorch training on eight GB200 GPUs, and a ROS evaluation on Jetson hardware through declared inputs and outputs. The software supports serial and parallel groups, templating, retries, priorities, preemption, and GPU borrowing.
The project is licensed under Apache-2.0 and provides Helm charts, NGC containers, and a KIND-based local setup. Version 6.3.1 is the latest release identified in the article. OSMO uses NVIDIA KAI Scheduler by default and includes features such as NVLink-aware placement, interactive remote development, RBAC, OAuth2 integration, and cloud workload identity. The standalone dataset CLI is deprecated in 6.3.0 and is planned for removal in 6.4.
Written by Sakhanda Wire AI from reporting by MarkTechPost. Facts, quotes and full context are in the original article below.