Deploy Exported Policies with LEAPP

Deploy Exported Policies with LEAPP#

Isaac Lab provides LeappDeploymentEnv for running exported policies back in simulation without the training infrastructure. This is the Isaac Lab deployment path for LEAPP-exported policies and is useful for validating that the packaged policy still behaves correctly when driven through the deployment stack instead of the training stack.

Run the deployment script with the task name and the exported LEAPP .yaml file. Use the same backend extra and backend selector that you used for training and export, and pass a --viz option when you want a viewport:

Linux

uv (Recommended)

# Newton backend (kitless)
uv run --extra leapp python \
    scripts/reinforcement_learning/leapp/deploy.py \
    --task <TASK_NAME> \
    --leapp_model <PATH_TO_EXPORTED_LEAPP_YAML> \
    --viz newton_gl physics=newton_mjwarp

# OV PhysX backend
uv run --extra ovphysx,leapp python \
    scripts/reinforcement_learning/leapp/deploy.py \
    --task <TASK_NAME> \
    --leapp_model <PATH_TO_EXPORTED_LEAPP_YAML> \
    --viz kit physics=ovphysx

# Isaac Sim PhysX backend
OMNI_KIT_ACCEPT_EULA=Y ACCEPT_EULA=Y uv run --extra isaacsim,leapp python \
    scripts/reinforcement_learning/leapp/deploy.py \
    --task <TASK_NAME> \
    --leapp_model <PATH_TO_EXPORTED_LEAPP_YAML> \
    --viz kit physics=isaacsim_physx

isaaclab.sh / isaaclab.bat

# Newton backend (kitless)
./isaaclab.sh -p \
    scripts/reinforcement_learning/leapp/deploy.py \
    --task <TASK_NAME> \
    --leapp_model <PATH_TO_EXPORTED_LEAPP_YAML> \
    --viz newton_gl physics=newton_mjwarp

# OV PhysX backend
./isaaclab.sh -p \
    scripts/reinforcement_learning/leapp/deploy.py \
    --task <TASK_NAME> \
    --leapp_model <PATH_TO_EXPORTED_LEAPP_YAML> \
    --viz kit physics=ovphysx

# Isaac Sim PhysX backend
OMNI_KIT_ACCEPT_EULA=Y ACCEPT_EULA=Y ./isaaclab.sh -p \
    scripts/reinforcement_learning/leapp/deploy.py \
    --task <TASK_NAME> \
    --leapp_model <PATH_TO_EXPORTED_LEAPP_YAML> \
    --viz kit physics=isaacsim_physx

Windows

uv (Recommended)

:: Newton backend (kitless)
uv run --extra leapp python scripts\reinforcement_learning\leapp\deploy.py ^
    --task <TASK_NAME> ^
    --leapp_model <PATH_TO_EXPORTED_LEAPP_YAML> ^
    --viz newton_gl physics=newton_mjwarp

:: OV PhysX backend
uv run --extra ovphysx,leapp python scripts\reinforcement_learning\leapp\deploy.py ^
    --task <TASK_NAME> ^
    --leapp_model <PATH_TO_EXPORTED_LEAPP_YAML> ^
    --viz kit physics=ovphysx

:: Isaac Sim PhysX backend
set OMNI_KIT_ACCEPT_EULA=Y
set ACCEPT_EULA=Y
uv run --extra isaacsim,leapp python scripts\reinforcement_learning\leapp\deploy.py ^
    --task <TASK_NAME> ^
    --leapp_model <PATH_TO_EXPORTED_LEAPP_YAML> ^
    --viz kit physics=isaacsim_physx

isaaclab.sh / isaaclab.bat

:: Newton backend (kitless)
isaaclab.bat -p scripts\reinforcement_learning\leapp\deploy.py ^
    --task <TASK_NAME> ^
    --leapp_model <PATH_TO_EXPORTED_LEAPP_YAML> ^
    --viz newton_gl physics=newton_mjwarp

:: OV PhysX backend
isaaclab.bat -p scripts\reinforcement_learning\leapp\deploy.py ^
    --task <TASK_NAME> ^
    --leapp_model <PATH_TO_EXPORTED_LEAPP_YAML> ^
    --viz kit physics=ovphysx

:: Isaac Sim PhysX backend
set OMNI_KIT_ACCEPT_EULA=Y
set ACCEPT_EULA=Y
isaaclab.bat -p scripts\reinforcement_learning\leapp\deploy.py ^
    --task <TASK_NAME> ^
    --leapp_model <PATH_TO_EXPORTED_LEAPP_YAML> ^
    --viz kit physics=isaacsim_physx

What the Deployment Environment Does#

The deployment environment loads the task scene and then bypasses the training-time observation, action, reward, termination, and curriculum managers. It reads the scene objects and commands identified by the isaaclab_connection metadata in the LEAPP YAML, uses them as inputs to the exported policy pipeline, and runs that pipeline through LEAPP’s InferenceManager. The resulting outputs are written back to the mapped scene entities.

Match the Training and Export Configuration#

The deployment script rebuilds the task configuration from --task. It cannot infer the training configuration from the checkpoint or LEAPP YAML, so you must provide the same configuration selections used for training and export. The task name, LEAPP YAML, and checkpoint must describe the same policy.

Carry these settings through training, export, and deployment:

Task and Hydra overrides: use the same task name and every configuration override that changes the scene, robot, sensors, command terms, observation/action term configuration, reset behavior, or simulation timing (sim.dt and decimation).
Physics and renderer presets: pass the same backend selectors when you selected a non-default backend. For example, a policy trained and exported with physics=newton_mjwarp must also be deployed with physics=newton_mjwarp; the same applies to the untyped presets=<PRESET_NAME> form.
Seed, when applicable: use the same --seed when reproducible command sampling, reset events, or other randomized task behavior matters to your validation.

For example, append the preset after the deployment options:

uv run --extra leapp python \
    scripts/reinforcement_learning/leapp/deploy.py \
    --task Isaac-Humanoid \
    --leapp_model <PATH_TO_EXPORTED_LEAPP_YAML> \
    --viz newton_gl \
    presets=newton_mjwarp

Visualization Options#

Visualization does not change the policy or its LEAPP inputs. Omit --viz for headless execution, use --viz kit for the Omniverse Kit viewport, or select newton_gl, newton_rtx (experimental), rerun, or viser when those visualizers are installed. Use --viz none to explicitly disable all visualizers, and --max_visible_envs <COUNT> to limit displayed environments. Choose a visualizer compatible with the selected physics and renderer backend.

Note

For Direct workflow policies, see the Direct workflow export guide. Direct workflow policies are not supported by LeappDeploymentEnv.