I’m trying to implement a solution in which a RL task is part of a bigger simulation, but I’m struggling to find a good way to do this.
I also have a standalone python script that:
1 - Launches the scene with a robot arm (Done)
2 - Do some manipulation stuff (Done)
Now, I want to:
3 - Depending on scene conditions, trigger an inference from a pre-trained environment (inference 1).
4 - Other conditions trigger inference 2.
I’m reading the new Isaac Lab Manager based environment (Isaac 4.0.0), and it looks like it is the way to go as it mentions a separation between world and environment, but I’m failing to grasp exactly how. Should I create different envs for 3) and 4) ? How would a launch this from the bigger simulation?
Any examples in which a RL inference is part of a bigger simulation?
Any pointers are appreciated!