Gymwrapper: support both gym and gymnasium, and support dict_obs and …

[youliangtan]

As described in: #497 , reopen after v1.5 release

• Support both gymnasium and gym interfaces. A runtime check will ensure gym>=0.26.0 (which uses terminate and truncate).
• Support Flatten and Dict Observation. GymWrapper(..., flatten_obs=True), previously only return flatten obs, which is not always desired.
• Ensure that the .observation_space specifies the correct dtype and shape.

Quick RUn

import robosuite as suite
from robosuite.wrappers.gym_wrapper import GymWrapper
import cv2

if __name__ == '__main__':
    # create environment instance
    env = suite.make(
        env_name="Lift",  # try with other tasks like "Stack" and "Door"
        robots="Panda",   # try with other robots like "Sawyer" and "Jaco"
        has_renderer=True,
        has_offscreen_renderer=True,
        use_camera_obs=True,
    )

    # convert to gym/gymnasium environment
    env = GymWrapper(
        env,
        keys=["robot0_proprio-state", "object-state", "agentview_image"],
        flatten_obs=False, # default is True, we will not flatten the observation
    )
    print(" -> action space:", env.action_space)
    print(" -> observation space:", env.observation_space)
    obs, info = env.reset()

    for i in range(1000):
        # action = np.random.randn(env.robots[0].dof)  # sample random action
        action = env.action_space.sample()
        obs, reward, done, trunc, info = env.step(action)  # take action in the environment
        print(obs.keys(), obs["agentview_image"].shape)

        # Optional camera view viz
        cv2.imshow("Agent View", cv2.cvtColor(obs["agentview_image"], cv2.COLOR_RGB2BGR))
        if cv2.waitKey(1) & 0xFF == ord('q'): break

        env.render()  # render on display

[kevin-thankyou-lin]

Looks good to me:

can we flip the cv2 imshow image?

[kevin-thankyou-lin]

Sorry, just realized that script isn't part of the code base. PR looks good