[Question] Using PhysX WizardVehicle USD inside Isaac Lab (PhysxVehicleContextAPI errors)

[sw-0429]

Hi, I’m currently testing PhysX WizardVehicle inside Isaac Lab (Isaac Sim 5.0, PhysX 5.4).
My goal is to load a vehicle created with the Vehicle Wizard and use it for reinforcement learning (RL) tasks.

However, when I load the exported .usd file in my custom Isaac Lab environment (DirectRLEnv),
the following errors appear in the console:
[Error] PhysX Vehicle: vehicle is being created but the stage has no matching physics scene prim with PhysxVehicleContextAPI applied. Default values are being used but might not match the required setup for the vehicle.
[Error] PhysX Vehicle: sprung mass computation failed. Setup might be ill-conditioned. Make sure the center of mass is enclosed by wheel positions.
[Error] PhysX error: It is not allowed to add a non-GPU compatible PxConstraint to a scene that is configured for direct-GPU access. Please use a D6 joint instead.
[Error] PhysX error: PxRigidDynamic::setLinearVelocity(): it is illegal to call this method if PxSceneFlag::eENABLE_DIRECT_GPU_API is enabled!
[Error] PhysX error: PxRigidDynamic::setAngularVelocity(): it is illegal to call this method if PxSceneFlag::eENABLE_DIRECT_GPU_API is enabled!

What I Did
1.lCreated a PhysX Vehicle using the Vehicle Wizard inside Isaac Sim and Exported the vehicle as a .usd file
2. loaded this file in Isaac Lab using a configuration similar to:
robot_cfg = RigidObjectCfg(
usd_path="C:/Users/Administrator/IsaacLab/source/isaaclab_assets/isaaclab_assets/robots/Vehicle.usd",
prim_path="{ENV_REGEX_NS}/Robot",
)

3.On startup, the above errors appeared.
To fix the first one, I manually applied a PhysxVehicleContextAPI to /physicsScene:
from pxr import PhysxSchema
scene_prim = stage.GetPrimAtPath("/physicsScene")
PhysxSchema.PhysxVehicleContextAPI.Apply(scene_prim)
However, the same warnings and constraint errors still appear.

My Questions
1.Is it currently possible to load a WizardVehicle USD inside Isaac Lab environments (e.g. DirectRLEnv or ManagerBasedRLEnv)?
Or does the current Lab physics setup not support PhysxVehicleDriveStandardAPI?

2.Even if the vehicle can load,
is reinforcement learning (via PPO, SAC, etc.) supported for such PhysX Vehicles?
For example, controlling attributes like Throttle, Brake, and Steer through RL actions.

3.Do I need to disable GPU-direct (use_fabric=False) or modify the scene configuration to make PhysX Vehicles compatible with Isaac Lab?

4.Is there an official or recommended example of using PhysX Vehicle APIs (WizardVehicle) with Isaac Lab or Omniverse RL environments?

[PeterL-NV]

Hi, It is possible to load USD assets exported from the PhysX Vehicle Wizard inside Isaac Lab environments, such as DirectRLEnv or ManagerBasedRLEnv, but there are important compatibility constraints and additional steps required—especially concerning the physics scene and APIs applied to the USD assets.
For vehicles created with the PhysX Vehicle Wizard, features such as wheels, drive systems, and suspension use PhysxVehicleDriveStandardAPI. Proper integration requires the physics scene to have matching schemas and context APIs (e.g., PhysxVehicleContextAPI) configured.

Here're some recommendations:

• When loading these USD vehicles, ensure that the Isaac Lab physics scene prim (often at /physicsScene) has the required PhysxVehicleContextAPI attached using Omniverse’s USD API tools.

• Isaac Lab’s RL environments (DirectRLEnv, ManagerBasedRLEnv) use GPU-accelerated physics by default. PhysX Vehicle Wizard assets may require changes to simulation flags, such as disabling GPU-direct with use_fabric=False, to avoid constraints and initialization errors.

Isaac Lab documentation and tutorials focus primarily on robots and articulated systems; explicit guidance for WizardVehicle USD integration is not available but the flexible asset loader (RigidObjectCfg) can accept custom USD files with the proper APIs and prim paths. With that being said, currently there is no official, out-of-the-box example for PhysX WizardVehicle assets specifically in DirectRLEnv or ManagerBasedRLEnv. Usage and troubleshooting may require custom USD editing, scene API application, and physics configuration adjustments.