There should be a way to get the current observation from our gym env and wrappers without having the need to issue an action. This is for example useful when we want to give a foundation model the current robot state in async mode which already includes the change induced by the current action.
We came up with two ideas on how to do this:
- Implement public get_obs method in all gym envs that returns the current state. This requires modify the gym interface and might cause problems with typing
- Allow "zero" action which does nothing and just returns the current observation. This would require to case check in the beginning of each step function.