Ecosystem

LEAPP sits between training stacks and deployment stacks. Upstream projects export a LEAPP bundle from the code they already run. Downstream runtimes load that bundle instead of rebuilding observation, action, and state handling by hand.

Isaac Lab

Export a trained reinforcement learning policy from Isaac Lab as a LEAPP bundle. The export keeps the observation preprocessing, action postprocessing, and recurrent state used during training.

https://isaac-sim.github.io/IsaacLab/release/3.0.0/source/policy_deployment/05_leapp/exporting_policies_with_leapp.html

Isaac ROS Deploy

Run a LEAPP bundle on ROS 2. The runtime loads the bundle, runs inference, maps policy terms onto topics or ros2_control, and can gate outputs through a safety controller.

https://nvidia-isaac-ros.github.io/repositories_and_packages/isaac_ros_deploy/index.html