Fix compatibility with gymnasium >= 1.0.0

[traversaro]

The change in Farama-Foundation/Gymnasium#868 broke the code in gym-xarm, resulting in the error:

TypeError: MujocoRenderer.render() got an unexpected keyword argument 'camera_name'

when executed.

This PR fixes the problem, by keeping compatibility with gymnasium 0.29.1 .

Fix #9 .
Fix huggingface/lerobot#474 .

[traversaro]

Even with this fix, some tests are failing with gymnasium==1.2.0 and gymnasium-robotics==1.4.1 (the same test pass successfully with older versions, such as the one installed by default by lerobot).

(lerobot) traversaro@IITBMP014LW012:~/lerobotws/gym-xarm$ pytest
============================================================================================================================ test session starts ============================================================================================================================
platform linux -- Python 3.10.18, pytest-8.4.1, pluggy-1.6.0
rootdir: /home/traversaro/lerobotws/gym-xarm
configfile: pyproject.toml
plugins: mock-serial-0.0.1, cov-6.2.1, timeout-2.4.0
collected 3 items

tests/test_env.py FFF                                                                                                                                                                                                                                                 [100%]

================================================================================================================================= FAILURES ==================================================================================================================================
________________________________________________________________________________________________________________________ test_env[XarmLift-v0-state] ________________________________________________________________________________________________________________________

env_task = 'XarmLift-v0', obs_type = 'state'

    @pytest.mark.parametrize(
        "env_task, obs_type",
        [
            ("XarmLift-v0", "state"),
            ("XarmLift-v0", "pixels"),
            ("XarmLift-v0", "pixels_agent_pos"),
        ],
    )
    def test_env(env_task, obs_type):
        env = gym.make(f"gym_xarm/{env_task}", obs_type=obs_type)
>       check_env(env.unwrapped)

tests/test_env.py:18:
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
../../miniforge3/envs/lerobot/lib/python3.10/site-packages/gymnasium/utils/env_checker.py:420: in check_env
    check_step_determinism(env)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

env = <gym_xarm.tasks.lift.Lift object at 0x773c5d453fa0>, seed = 123

    def check_step_determinism(env: gym.Env, seed=123):
        """Check that the environment steps deterministically after reset.

        Note: This check assumes that seeded `reset()` is deterministic (it must have passed `check_reset_seed`) and that `step()` returns valid values (passed `env_step_passive_checker`).
        Note: A single step should be enough to assert that the state transition function is deterministic (at least for most environments).

        Raises:
            AssertionError: The environment cannot be step deterministically after resetting with a random seed,
                or it truncates after 1 step.
        """
        if env.spec is not None and env.spec.nondeterministic is True:
            return

        env.action_space.seed(seed)
        action = env.action_space.sample()

        env.reset(seed=seed)
        obs_0, rew_0, term_0, trunc_0, info_0 = env.step(action)
        seeded_rng: np.random.Generator = deepcopy(env.unwrapped._np_random)

        env.reset(seed=seed)
        obs_1, rew_1, term_1, trunc_1, info_1 = env.step(action)

        assert (
            env.unwrapped._np_random.bit_generator.state  # pyright: ignore [reportOptionalMemberAccess]
            == seeded_rng.bit_generator.state
        ), "The `.np_random` is not properly been updated after step."

>       assert data_equivalence(
            obs_0, obs_1
        ), "Deterministic step observations are not equivalent for the same seed and action"
E       AssertionError: Deterministic step observations are not equivalent for the same seed and action

../../miniforge3/envs/lerobot/lib/python3.10/site-packages/gymnasium/utils/env_checker.py:216: AssertionError
_______________________________________________________________________________________________________________________ test_env[XarmLift-v0-pixels] ________________________________________________________________________________________________________________________

env_task = 'XarmLift-v0', obs_type = 'pixels'

    @pytest.mark.parametrize(
        "env_task, obs_type",
        [
            ("XarmLift-v0", "state"),
            ("XarmLift-v0", "pixels"),
            ("XarmLift-v0", "pixels_agent_pos"),
        ],
    )
    def test_env(env_task, obs_type):
        env = gym.make(f"gym_xarm/{env_task}", obs_type=obs_type)
>       check_env(env.unwrapped)

tests/test_env.py:18:
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
../../miniforge3/envs/lerobot/lib/python3.10/site-packages/gymnasium/utils/env_checker.py:420: in check_env
    check_step_determinism(env)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

env = <gym_xarm.tasks.lift.Lift object at 0x773c4e9974f0>, seed = 123

    def check_step_determinism(env: gym.Env, seed=123):
        """Check that the environment steps deterministically after reset.

        Note: This check assumes that seeded `reset()` is deterministic (it must have passed `check_reset_seed`) and that `step()` returns valid values (passed `env_step_passive_checker`).
        Note: A single step should be enough to assert that the state transition function is deterministic (at least for most environments).

        Raises:
            AssertionError: The environment cannot be step deterministically after resetting with a random seed,
                or it truncates after 1 step.
        """
        if env.spec is not None and env.spec.nondeterministic is True:
            return

        env.action_space.seed(seed)
        action = env.action_space.sample()

        env.reset(seed=seed)
        obs_0, rew_0, term_0, trunc_0, info_0 = env.step(action)
        seeded_rng: np.random.Generator = deepcopy(env.unwrapped._np_random)

        env.reset(seed=seed)
        obs_1, rew_1, term_1, trunc_1, info_1 = env.step(action)

        assert (
            env.unwrapped._np_random.bit_generator.state  # pyright: ignore [reportOptionalMemberAccess]
            == seeded_rng.bit_generator.state
        ), "The `.np_random` is not properly been updated after step."

        assert data_equivalence(
            obs_0, obs_1
        ), "Deterministic step observations are not equivalent for the same seed and action"
        if not data_equivalence(obs_0, obs_1, exact=True):
            logger.warn(
                "Step observations are not equal although similar given the same seed and action"
            )

>       assert data_equivalence(
            rew_0, rew_1
        ), "Deterministic step rewards are not equivalent for the same seed and action"
E       AssertionError: Deterministic step rewards are not equivalent for the same seed and action

../../miniforge3/envs/lerobot/lib/python3.10/site-packages/gymnasium/utils/env_checker.py:224: AssertionError
__________________________________________________________________________________________________________________ test_env[XarmLift-v0-pixels_agent_pos] ___________________________________________________________________________________________________________________

env_task = 'XarmLift-v0', obs_type = 'pixels_agent_pos'

    @pytest.mark.parametrize(
        "env_task, obs_type",
        [
            ("XarmLift-v0", "state"),
            ("XarmLift-v0", "pixels"),
            ("XarmLift-v0", "pixels_agent_pos"),
        ],
    )
    def test_env(env_task, obs_type):
        env = gym.make(f"gym_xarm/{env_task}", obs_type=obs_type)
>       check_env(env.unwrapped)

tests/test_env.py:18:
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
../../miniforge3/envs/lerobot/lib/python3.10/site-packages/gymnasium/utils/env_checker.py:420: in check_env
    check_step_determinism(env)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

env = <gym_xarm.tasks.lift.Lift object at 0x773c4eaf79a0>, seed = 123

    def check_step_determinism(env: gym.Env, seed=123):
        """Check that the environment steps deterministically after reset.

        Note: This check assumes that seeded `reset()` is deterministic (it must have passed `check_reset_seed`) and that `step()` returns valid values (passed `env_step_passive_checker`).
        Note: A single step should be enough to assert that the state transition function is deterministic (at least for most environments).

        Raises:
            AssertionError: The environment cannot be step deterministically after resetting with a random seed,
                or it truncates after 1 step.
        """
        if env.spec is not None and env.spec.nondeterministic is True:
            return

        env.action_space.seed(seed)
        action = env.action_space.sample()

        env.reset(seed=seed)
        obs_0, rew_0, term_0, trunc_0, info_0 = env.step(action)
        seeded_rng: np.random.Generator = deepcopy(env.unwrapped._np_random)

        env.reset(seed=seed)
        obs_1, rew_1, term_1, trunc_1, info_1 = env.step(action)

        assert (
            env.unwrapped._np_random.bit_generator.state  # pyright: ignore [reportOptionalMemberAccess]
            == seeded_rng.bit_generator.state
        ), "The `.np_random` is not properly been updated after step."

>       assert data_equivalence(
            obs_0, obs_1
        ), "Deterministic step observations are not equivalent for the same seed and action"
E       AssertionError: Deterministic step observations are not equivalent for the same seed and action

../../miniforge3/envs/lerobot/lib/python3.10/site-packages/gymnasium/utils/env_checker.py:216: AssertionError
============================================================================================================================= warnings summary ==============================================================================================================================
tests/test_env.py::test_env[XarmLift-v0-state]
tests/test_env.py::test_env[XarmLift-v0-pixels_agent_pos]
  /home/traversaro/miniforge3/envs/lerobot/lib/python3.10/site-packages/gymnasium/utils/env_checker.py:120: UserWarning: WARN: Using `env.reset(seed=123)` observations are not equal although similar.
    logger.warn(

tests/test_env.py::test_env[XarmLift-v0-state]
tests/test_env.py::test_env[XarmLift-v0-pixels_agent_pos]
  /home/traversaro/miniforge3/envs/lerobot/lib/python3.10/site-packages/gymnasium/utils/env_checker.py:124: UserWarning: WARN: Using `env.reset(seed=123)` then `env.reset()` observations are not equal although similar.
    logger.warn(

-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html
========================================================================================================================== short test summary info ==========================================================================================================================
FAILED tests/test_env.py::test_env[XarmLift-v0-state] - AssertionError: Deterministic step observations are not equivalent for the same seed and action
FAILED tests/test_env.py::test_env[XarmLift-v0-pixels] - AssertionError: Deterministic step rewards are not equivalent for the same seed and action
FAILED tests/test_env.py::test_env[XarmLift-v0-pixels_agent_pos] - AssertionError: Deterministic step observations are not equivalent for the same seed and action
======================================================================================================================= 3 failed, 4 warnings in 1.82s =======================================================================================================================

[jadechoghari]

Hi @traversaro thanks for your PR, is it compatible with gymnasium >= 1.0.0, ? or are you encountering issues. otherwise we are planning to depreciate gym-xarm in LeRobot

[traversaro]

It wanted to be compatible with gymnaisum 1.0.0, but there was the test failure in #12 (comment) . I suspect it was not a regression, but rather the environment tests on gymnasium that became more strict. Anyhow I was only interested in this as it was a transitive dependency of gymnasium[all], if that is going to be dropped, I would not be interested in it anymore.