physical-ai-toolchain

Training documentation for reinforcement learning with Isaac Lab and behavioral cloning with LeRobot. Both frameworks run on Azure ML and NVIDIA OSMO platforms.

📖 Training Guides

Guide Description
Isaac Lab Training RL training with SKRL and RSL-RL backends on Azure ML and OSMO
LeRobot Training Behavioral cloning with ACT and Diffusion policies
Experiment Tracking MLflow and WANDB setup, model registration, checkpoint flows
MLflow Integration SKRL metric logging internals, metric filtering, and troubleshooting

⚖️ Platform Comparison

Aspect Azure ML OSMO
Submission az ml job create osmo workflow submit
Orchestration Azure ML compute targets OSMO workflow engine + KAI Scheduler
Experiment tracking MLflow (managed) MLflow + WANDB (credential injection)
Dataset injection Azure ML datastores OSMO buckets (base64 or dataset upload)
Model registration az ml model create Via MLflow or post-training script
Monitoring Azure ML Studio OSMO UI Dashboard

🚀 Quick Start

Isaac Lab RL training on Azure ML:

./scripts/submit-azureml-training.sh --task Isaac-Velocity-Rough-Anymal-C-v0

LeRobot behavioral cloning on OSMO:

./scripts/submit-osmo-lerobot-training.sh -d lerobot/aloha_sim_insertion_human

🤖 Crafted with precision by ✨Copilot following brilliant human instruction, then carefully refined by our team of discerning human reviewers.