Modified FrankaCabinet Example with Actor’s RigidBody set to nan

Setup:

I substitures the original Franka arm with a self-designed 7DOF one, and modify numObservations, numActions and robot-related parameters with the IsaacGymEnvs.

Problem:

ValueError: The parameter loc has invalid values with CPU pipeline. With GPU pipeline, there are more problems related with memory allocation.

What I’ve tried

The default dof configuration has been tested with jointmonkey.py. After printing tensors, I found that after using gym.acquire_rigid_body_state_tensor, tensors related to the robot arm are filled with “nan”.

Hi there,

The ValueError generally occurs when the policy receives NaNs in the observation or reward buffers. It could be possible that certain actions or reset states you are applying to your robot are causing incorrect behaviours in simulation.