[Feature] Habitat Rearrange tasks + Bug fixes use due to bad default initial poses

[arth-shukla]

actor/articulation building

• Set objects/articulations initial pose 5m apart (starting 10m away from origin)
• allow passing initial pose to articulation builder/urdf loader

scene builders/manip

• rearrange scene builder don't change collision groups after reconfigure
• register rearrange scene builders properly
• SceneManipulationEnv updates
• Change rearrange data source in download_asset
• replicacad replace numpy navigable position files with trimesh mesh

misc

• fetch updates
  • faster base
  • is_static requires smaller base qvel
• reset() info includes info from get_info()
• build cylinder function
• allow custom encoder in dump_json

[StoneT2000]

ok changes are good, just run and fix the errors that occur in the test (you can ignore any of the ones that complain about not enough gpu mem)

[StoneT2000]

tests

[arth-shukla]

1. fixed issues with my code and also fixed some other tests
2. added to setup.py [dev]:
   • pytest-forked plugin
     • allows pytest --forked
     • fixes RuntimeError: GPU PhysX can only be enabled once before any other code involving PhysX
   • pytest-xdist[psutil]
     • same as run.sh

only some remaining errors on unrelated tests

• download_asset does not allow output_dir, r3m not installed (on my local system)

============================= test session starts ==============================
platform linux -- Python 3.9.20, pytest-8.3.3, pluggy-1.5.0
rootdir: /home/arths/coding/sulab/ManiSkill
configfile: pyproject.toml
plugins: xdist-3.6.1, forked-1.6.0
created: 6/6 workers
6 workers [372 items]

...............................F........................................ [ 19%]
..............................................................s......... [ 38%]
........................................................................ [ 58%]
..................................s..................................... [ 77%]
........................................................................ [ 96%]
........F.F.                                                             [100%]
=================================== FAILURES ===================================
___________________________ test_download_asset_all ____________________________
[gw0] linux -- Python 3.9.20 /home/arths/mambaforge/envs/ms/bin/python3.9
@pytest.mark.slow
    def test_download_asset_all():
        from mani_skill.utils.download_asset import main, parse_args
    
>       main(parse_args(args=["all", "-y", "--quiet", "-o", "/tmp/ms2test"]))

tests/test_downloads.py:10: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

args = Namespace(uid='all', list=None, quiet=True, non_interactive=True, output_dir='/tmp/ms2test')

    def main(args):
        verbose = not args.quiet
    
        if args.list:
            downloadable_ids = []
            for k, v in assets.DATA_SOURCES.items():
                if v.source_type == args.list:
                    downloadable_ids.append(k)
            print(f"For category {args.list} the following asset UIDs are available")
            print(downloadable_ids)
            exit()
        if args.uid == "":
            print("Available asset (group) uids:")
            print(list(assets.DATA_GROUPS.keys()))
            return
        if args.uid == "all":
            if verbose:
                print("All assets will be downloaded. This may take a while.")
            uids = list(assets.DATA_SOURCES.keys())
            show_progress = True
        elif args.uid in assets.DATA_GROUPS:
            uids = assets.expand_data_group_into_individual_data_source_ids(args.uid)
            show_progress = True
        elif args.uid in assets.DATA_SOURCES:
            uids = [args.uid]
            show_progress = False
        else:
            raise KeyError("{} not found.".format(args.uid))
    
        for i, uid in enumerate(uids):
            if show_progress and verbose:
                print("Downloading assets for {}: {}/{}".format(args.uid, i + 1, len(uids)))
    
            kwargs = dict()
            kwargs["verbose"] = verbose
            kwargs["non_interactive"] = args.non_interactive
            if args.output_dir is not None:
                kwargs["output_dir"] = args.output_dir
>           output_path = download(assets.DATA_SOURCES[uid], **kwargs)
E           TypeError: download() got an unexpected keyword argument 'output_dir'

mani_skill/utils/download_asset.py:228: TypeError
_____________________ test_visualencoders_gpu[PickCube-v1] _____________________
[gw0] linux -- Python 3.9.20 /home/arths/mambaforge/envs/ms/bin/python3.9
env_id = 'PickCube-v1'

    @pytest.mark.gpu_sim
    @pytest.mark.parametrize("env_id", STATIONARY_ENV_IDS[:1])
    def test_visualencoders_gpu(env_id):
        env = gym.make(
            env_id,
            obs_mode="rgbd",
            render_mode="rgb_array",
            max_episode_steps=10,
            num_envs=16,
            sim_config=LOW_MEM_SIM_CONFIG,
        )
        assert (
            "embedding" not in env.observation_space.keys()
            and "sensor_data" in env.observation_space.keys()
        )
>       env = VisualEncoderWrapper(env, encoder="r3m")

tests/test_wrappers.py:211: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

self = <[RecursionError('maximum recursion depth exceeded while getting the str of an object') raised in repr()] VisualEncoderWrapper object at 0x77ddb81c5df0>
env = <TimeLimitWrapper<OrderEnforcing<PickCubeEnv<PickCube-v1>>>>
encoder = 'r3m', encoder_config = {}

    def __init__(self, env, encoder: Literal["r3m"], encoder_config=dict()):
        self.base_env: BaseEnv = env.unwrapped
        assert encoder == "r3m", "Only encoder='r3m' is supported at the moment"
        if encoder == "r3m":
            assert self.base_env.obs_mode in [
                "rgbd",
                "rgb",
            ], "r3m encoder requires obs_mode to be set to rgbd or rgb"
            import torchvision.transforms as T
>           from r3m import load_r3m
E           ModuleNotFoundError: No module named 'r3m'

mani_skill/utils/wrappers/visual_encoders.py:20: ModuleNotFoundError
----------------------------- Captured stderr call -----------------------------
�[33;1m2024-10-11 22:31:55,828 - mani_skill  - WARNING - initial pose not set for static object scenes-[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15]_ground, setting to default pose q=[1,0,0,0], p=[0,0,0]�[0m
_________________ test_visualencoder_flatten_gpu[PickCube-v1] __________________
[gw1] linux -- Python 3.9.20 /home/arths/mambaforge/envs/ms/bin/python3.9
env_id = 'PickCube-v1'

    @pytest.mark.gpu_sim
    @pytest.mark.parametrize("env_id", STATIONARY_ENV_IDS[:1])
    def test_visualencoder_flatten_gpu(env_id):
        env = gym.make(
            env_id,
            obs_mode="rgbd",
            render_mode="rgb_array",
            max_episode_steps=10,
            num_envs=16,
            sim_config=LOW_MEM_SIM_CONFIG,
        )
>       env = VisualEncoderWrapper(env, encoder="r3m")

tests/test_wrappers.py:243: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

self = <[RecursionError('maximum recursion depth exceeded while getting the str of an object') raised in repr()] VisualEncoderWrapper object at 0x76a0442cfac0>
env = <TimeLimitWrapper<OrderEnforcing<PickCubeEnv<PickCube-v1>>>>
encoder = 'r3m', encoder_config = {}

    def __init__(self, env, encoder: Literal["r3m"], encoder_config=dict()):
        self.base_env: BaseEnv = env.unwrapped
        assert encoder == "r3m", "Only encoder='r3m' is supported at the moment"
        if encoder == "r3m":
            assert self.base_env.obs_mode in [
                "rgbd",
                "rgb",
            ], "r3m encoder requires obs_mode to be set to rgbd or rgb"
            import torchvision.transforms as T
>           from r3m import load_r3m
E           ModuleNotFoundError: No module named 'r3m'

mani_skill/utils/wrappers/visual_encoders.py:20: ModuleNotFoundError
----------------------------- Captured stderr call -----------------------------
�[33;1m2024-10-11 22:31:56,570 - mani_skill  - WARNING - initial pose not set for static object scenes-[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15]_ground, setting to default pose q=[1,0,0,0], p=[0,0,0]�[0m
=============================== warnings summary ===============================
../../../mambaforge/envs/ms/lib/python3.9/site-packages/sapien/__init__.py:2
../../../mambaforge/envs/ms/lib/python3.9/site-packages/sapien/__init__.py:2
../../../mambaforge/envs/ms/lib/python3.9/site-packages/sapien/__init__.py:2
../../../mambaforge/envs/ms/lib/python3.9/site-packages/sapien/__init__.py:2
../../../mambaforge/envs/ms/lib/python3.9/site-packages/sapien/__init__.py:2
../../../mambaforge/envs/ms/lib/python3.9/site-packages/sapien/__init__.py:2
  /home/arths/mambaforge/envs/ms/lib/python3.9/site-packages/sapien/__init__.py:2: DeprecationWarning: pkg_resources is deprecated as an API. See https://setuptools.pypa.io/en/latest/pkg_resources.html
    import pkg_resources

-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html
=========================== short test summary info ============================
FAILED tests/test_downloads.py::test_download_asset_all
FAILED tests/test_wrappers.py::test_visualencoders_gpu[PickCube-v1]
FAILED tests/test_wrappers.py::test_visualencoder_flatten_gpu[PickCube-v1]
======= 3 failed, 367 passed, 2 skipped, 6 warnings in 189.28s (0:03:09) =======