Download
OSMO is a developer-first orchestration platform designed to scale complex physical AI workflows across heterogeneous computing environments, including cloud GPUs, simulation clusters, and edge devices. It was originally built internally at NVIDIA to support robotics and embodied AI systems, where workflows span multiple stages such as data generation, training, simulation, and hardware testing. The platform addresses what NVIDIA refers to as the “three computer problem” by unifying these stages into a single pipeline defined through simple YAML configurations. It enables users to orchestrate tasks across Kubernetes clusters, automatically managing dependencies, scheduling, and resource allocation without requiring deep infrastructure expertise.
Features
- Unified orchestration of training, simulation, and hardware workflows
- YAML-based pipeline definition without custom scripting
- Distributed execution across Kubernetes clusters and edge devices
- Support for reinforcement learning and synthetic data generation
- Interactive development via Jupyter, VSCode, and SSH
- Hardware-in-the-loop validation and benchmarking
Project Samples
