Access/Modify simulation state when using Orbit BaseEnv

[JChunX]

I am trying to port a MPPI controller from Isaacgym to Orbit. The controller itself relies on some pretty fine-grained control over environment states for parallelized rollouts. eg. using set_actor_root_state_tensor

I know that I can do something similar in Orbit by using the Scene APIs, but I would also want to take advantage of the modularity and randomization capabilities of BaseEnv. Reading the documentation, it looks like my best option would be to add terms to ActionsCfg that directly control joint positions and velocities for the robot, but I don't know if this will also work for all dynamic prims in the scene. Any suggestions would be awesome!

[Mayankm96]

Hi @JChunX ,

I think I'd need to understand this a bit more before recommending a path.

In your case, do you want to use the environment for performing rollouts for your controller? In that case, your MPPI controller will wrap around the environment and call the step as needed. I don't think you would implement the MPPI algorithm as an action term itself since the action space should match the space of your controller's output.