Isaac LEAPP#

Isaac LEAPP

Lightweight Export Annotations
for Policy Pipelines

GitHub repository Download agent skill

Export the whole policy pipeline, not just the model weights.

Isaac LEAPP is NVIDIA’s library for machine learning pipeline export. LEAPP stands for Lightweight Export Annotations for Policy Pipelines. It exports the computation you already run into ONNX, TorchScript, or PyTorch ExportedProgram models. Capture preprocessing, model inference, and action processing together in one model, even when the computation spans helper functions, class methods, and supported PyTorch and NumPy operations. Choose separate models only when your deployment benefits from independent stages.

LEAPP is designed for the full policy path, including the logic that surrounds the learned model: normalization, feature transforms, recurrent state, action chunking, unit conversion, clipping, and task-specific output shaping. Without LEAPP, teams often rewrite that logic in another language, maintain a separate export pipeline, or carry the full research environment into deployment.

With LEAPP, you mark the input and output boundaries of the compute you already run. Execute the annotated pipeline with sample inputs, and LEAPP compiles the traced graph and validates exported outputs against captured Python outputs. Your code does not need to implement an RL environment interface or fit into a single forward() method. Isaac Lab also provides a native LEAPP export integration; see Ecosystem.

Get started

Read usage guide

Using the agent skill

The download includes SKILL.md and supporting guides for function patches, state, deployment semantics, and Warp. Extract the ZIP and give your agent leapp-integrate/SKILL.md, or place the leapp-integrate folder in your agent’s skills directory. Keep the references folder beside SKILL.md so the linked guides remain available. No local LEAPP checkout is needed.

Is LEAPP a Fit?#

Use LEAPP when your deployment target needs more than a single model checkpoint. It is a strong fit when you need to:

Deploy Python policy code

Move Python-based policy code from research workflows into deployment.

Capture the full pipeline

Keep preprocessing, postprocessing, constants, and tensor operations around the learned model.

Reuse existing computation

Reuse supported computation directly from your application.

Move between runtimes

Carry policy logic into another runtime, simulator, application, or deployment stack.

Choose one model or several

Export the full computation as one model, or choose separate stages with independent export backends when needed.

Describe tensor meaning

Help downstream systems connect inputs, state, commands, and outputs.

How LEAPP Fits In#

LEAPP traces the code path you actually execute. One annotated input/output boundary can capture the entire policy pipeline into one model file. Multiple boundaries let you export separate models instead. A YAML specification describes the model interfaces, semantic metadata, and any state feedback or connections between models.

The result is an export bundle that can be inspected, validated, and handed to downstream systems without requiring those systems to understand the original training project.

annotate inputs
Existing policy code
annotate outputs
export bundle
control.onnx pipeline.yaml visualization.png

Documentation Map#