Does it support RL Dict Observation?

[shaoxiang]

I saw it in wrappers/rl_games only support gym.spaces.Box, But I need to use gym.spaces.Dict.
Any one have any good methods or suggestions?

    def observation_space(self) -> gym.spaces.Box:
        """Returns the :attr:`Env` :attr:`observation_space`."""
        # note: rl-games only wants single observation space
        policy_obs_space = self.unwrapped.single_observation_space["policy"]
        if not isinstance(policy_obs_space, gymnasium.spaces.Box):
            raise NotImplementedError(
                f"The RL-Games wrapper does not currently support observation space: '{type(policy_obs_space)}'."
                f" If you need to support this, please modify the wrapper: {self.__class__.__name__},"
                " and if you are nice, please send a merge-request."
            )
        # note: maybe should check if we are a sub-set of the actual space. don't do it right now since
        #   in ManagerBasedRLEnv we are setting action space as (-inf, inf).
        return gym.spaces.Box(-self._clip_obs, self._clip_obs, policy_obs_space.shape)

[Mayankm96]

The observation manager returns a dictionary (as you can see from the code). If RL-Games accepts a dictionary, the above function should be modified to do that. Unfortunately, I do not have the bandwidth to test this out without a concrete example available.