Isaac Lab on AWS: From Simulation to Registered Policy

This article introduces Isaac Lab on AWS, an AWS CDK solution that provides a complete physical AI development environment from simulation to policy registration, eliminating weeks of infrastructure setup.

• Golden AMI pre-installs NVIDIA Isaac Sim, Isaac Lab, GPU drivers, and remote desktop for immediate productivity
• Two deployment topologies: Instance Mode for individual researchers with direct GPU access, Fleet Mode for multi-user self-service platform with auto-scaling
• Integrated Amazon SageMaker training plane with project-as-input architecture—no Docker rebuilds between desktop iteration and scaled training
• Automatic MLflow experiment tracking and model registration with provenance tags for governance
• Three bundled samples demonstrate full workflow: SO-ARM101 reach-grasp-lift, teleoperation with physical leader arm, and Unitree H1 rough-terrain locomotion
• Network security includes private VPC endpoints, encryption at rest/transit, and in Fleet Mode, Cognito MFA and DCV Connection Gateway brokering
• Idle-aware resource reclaim minimizes costs; golden AMI build is one-time cost skipped on subsequent deploys

Isaac Lab on AWS eliminates infrastructure overhead for physical AI teams, enabling robotics engineers to focus on policy development rather than cloud configuration.