search

metadriverse / metadrive

MetaDrive: Open-source driving simulator
https://metadriverse.github.io/metadrive/
Apache License 2.0
671 stars 100 forks source link

verify_headless_installation fails #709

Open mimi89999 opened 2 months ago

mimi89999 commented 2 months ago

Hello,

I noticed that running fails with the following error:

(.venv) michel@debian:/tmp$ python -m metadrive.examples.verify_headless_installation
[INFO] Environment: MetaDriveEnv
[INFO] MetaDrive version: 0.4.2.3
[INFO] Sensors: [lidar: Lidar(), side_detector: SideDetector(), lane_line_detector: LaneLineDetector()]
[INFO] Render Mode: none
[INFO] Horizon (Max steps per agent): None
[INFO] Assets version: 0.4.2.3
[INFO] Known Pipes: glxGraphicsPipe
[INFO] Start Scenario Index: 0, Num Scenarios : 1
[INFO] Episode ended! Scenario Index: 0 Reason: out_of_road.
Bullet physics world is launched successfully!
[INFO] Environment: MetaDriveEnv
[INFO] MetaDrive version: 0.4.2.3
[INFO] Sensors: [lidar: Lidar(), side_detector: SideDetector(), lane_line_detector: LaneLineDetector(), rgb_camera: RGBCamera(256, 256), depth_camera: DepthCamera(256, 256)]
[INFO] Render Mode: offscreen
[INFO] Horizon (Max steps per agent): None
Traceback (most recent call last):
  File "<frozen runpy>", line 198, in _run_module_as_main
  File "<frozen runpy>", line 88, in _run_code
  File "/tmp/.venv/lib/python3.11/site-packages/metadrive/examples/verify_headless_installation.py", line 11, in <module>
    verify_installation(args.cuda, args.camera)
  File "/tmp/.venv/lib/python3.11/site-packages/metadrive/tests/test_installation.py", line 84, in verify_installation
    capture_headless_image(cuda)
  File "/tmp/.venv/lib/python3.11/site-packages/metadrive/tests/test_installation.py", line 14, in capture_headless_image
    env = MetaDriveEnv(
          ^^^^^^^^^^^^^
  File "/tmp/.venv/lib/python3.11/site-packages/metadrive/envs/metadrive_env.py", line 103, in __init__
    super(MetaDriveEnv, self).__init__(config)
  File "/tmp/.venv/lib/python3.11/site-packages/metadrive/envs/base_env.py", line 299, in __init__
    self.agent_manager = self._get_agent_manager()
                         ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/tmp/.venv/lib/python3.11/site-packages/metadrive/envs/base_env.py", line 396, in _get_agent_manager
    return VehicleAgentManager(init_observations=self._get_observations())
                                                 ^^^^^^^^^^^^^^^^^^^^^^^^
  File "/tmp/.venv/lib/python3.11/site-packages/metadrive/envs/base_env.py", line 393, in _get_observations
    return {DEFAULT_AGENT: self.get_single_observation()}
                           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/tmp/.venv/lib/python3.11/site-packages/metadrive/envs/base_env.py", line 661, in get_single_observation
    o = ImageStateObservation(self.config) if img_obs else LidarStateObservation(self.config)
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/tmp/.venv/lib/python3.11/site-packages/metadrive/obs/image_obs.py", line 26, in __init__
    self.img_obs = ImageObservation(config, config["vehicle_config"]["image_source"], config["norm_pixel"])
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/tmp/.venv/lib/python3.11/site-packages/metadrive/obs/image_obs.py", line 61, in __init__
    self.state = np.zeros(self.observation_space.shape, dtype=np.float32 if self.norm_pixel else np.uint8)
                          ^^^^^^^^^^^^^^^^^^^^^^
  File "/tmp/.venv/lib/python3.11/site-packages/metadrive/obs/image_obs.py", line 67, in observation_space
    sensor_cls = self.config["sensors"][self.image_source][0]
                 ~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^
  File "/tmp/.venv/lib/python3.11/site-packages/metadrive/utils/config.py", line 225, in __getitem__
    self._check_and_raise_key_error(item)
  File "/tmp/.venv/lib/python3.11/site-packages/metadrive/utils/config.py", line 211, in _check_and_raise_key_error
    raise KeyError(
KeyError: "'main_camera' does not exist in existing config. Please use config.update(...) to update the config. Existing keys: dict_keys(['lidar', 'side_detector', 'lane_line_detector', 'rgb_camera', 'depth_camera'])."

After switching to the rgb camera, I get a different error:

(.venv) michel@debian:/tmp$ python -m metadrive.examples.verify_headless_installation --camera rgb
[INFO] Environment: MetaDriveEnv
[INFO] MetaDrive version: 0.4.2.3
[INFO] Sensors: [lidar: Lidar(), side_detector: SideDetector(), lane_line_detector: LaneLineDetector()]
[INFO] Render Mode: none
[INFO] Horizon (Max steps per agent): None
[INFO] Assets version: 0.4.2.3
[INFO] Known Pipes: glxGraphicsPipe
[INFO] Start Scenario Index: 0, Num Scenarios : 1
[INFO] Episode ended! Scenario Index: 0 Reason: out_of_road.
Bullet physics world is launched successfully!
[INFO] Environment: MetaDriveEnv
[INFO] MetaDrive version: 0.4.2.3
[INFO] Sensors: [lidar: Lidar(), side_detector: SideDetector(), lane_line_detector: LaneLineDetector(), rgb_camera: RGBCamera(256, 256), depth_camera: DepthCamera(256, 256)]
[INFO] Render Mode: offscreen
[INFO] Horizon (Max steps per agent): None
[INFO] Assets version: 0.4.2.3
Xlib:  extension "XFree86-DGA" missing on display ":0".
[INFO] Known Pipes: glxGraphicsPipe
[WARNING] You are using too large buffer! The height is 256, and width is 256. It may lower the sample efficiency! Consider reducing buffer size or use cuda image by set [image_on_cuda=True]. (base_camera.py:46)
[WARNING] You are using too large buffer! The height is 256, and width is 256. It may lower the sample efficiency! Consider reducing buffer size or use cuda image by set [image_on_cuda=True]. (base_camera.py:46)
Assertion failed: _current_shader_context != nullptr at line 6939 of panda/src/glstuff/glGraphicsStateGuardian_src.cxx
Traceback (most recent call last):
  File "/tmp/.venv/lib/python3.11/site-packages/metadrive/component/sensors/depth_camera.py", line 142, in _dispatch_compute
    self.engine.graphicsEngine.dispatch_compute(
AssertionError: _current_shader_context != nullptr at line 6939 of panda/src/glstuff/glGraphicsStateGuardian_src.cxx
Traceback (most recent call last):
  File "<frozen runpy>", line 198, in _run_module_as_main
  File "<frozen runpy>", line 88, in _run_code
  File "/tmp/.venv/lib/python3.11/site-packages/metadrive/examples/verify_headless_installation.py", line 11, in <module>
    verify_installation(args.cuda, args.camera)
  File "/tmp/.venv/lib/python3.11/site-packages/metadrive/tests/test_installation.py", line 86, in verify_installation
    capture_headless_image(cuda, image_source="rgb_camera")
  File "/tmp/.venv/lib/python3.11/site-packages/metadrive/tests/test_installation.py", line 33, in capture_headless_image
    env.reset()
  File "/tmp/.venv/lib/python3.11/site-packages/metadrive/envs/base_env.py", line 519, in reset
    self.lazy_init()  # it only works the first time when reset() is called to avoid the error when render
    ^^^^^^^^^^^^^^^^
  File "/tmp/.venv/lib/python3.11/site-packages/metadrive/envs/base_env.py", line 406, in lazy_init
    initialize_engine(self.config)
  File "/tmp/.venv/lib/python3.11/site-packages/metadrive/engine/engine_utils.py", line 12, in initialize_engine
    cls.singleton = cls(env_global_config)
                    ^^^^^^^^^^^^^^^^^^^^^^
  File "/tmp/.venv/lib/python3.11/site-packages/metadrive/engine/base_engine.py", line 88, in __init__
    self.warmup()
  File "/tmp/.venv/lib/python3.11/site-packages/metadrive/engine/base_engine.py", line 741, in warmup
    self.taskMgr.step()
  File "/tmp/.venv/lib/python3.11/site-packages/direct/task/Task.py", line 504, in step
    self.mgr.poll()
  File "/tmp/.venv/lib/python3.11/site-packages/metadrive/component/sensors/depth_camera.py", line 142, in _dispatch_compute
    self.engine.graphicsEngine.dispatch_compute(
AssertionError: _current_shader_context != nullptr at line 6939 of panda/src/glstuff/glGraphicsStateGuardian_src.cxx
QuanyiLi commented 2 months ago

For the first error, you should explicitly declare that you need the main_camera as a sensor by putting it in the sensor config:

"sensors":{"main_camera": ()
    ...                 
}

So you can access it even in offscreen mode.

The second error is raised by the depth camera because the GPU doesn't support OpenGL > 4.0 and the compute shader is not available. It usually happens for users using an integrated graphics card.

mimi89999 commented 2 months ago

My GPU should support OpenGL 4.6. Is there any specific extension that is required that I could check for?

QuanyiLi commented 3 weeks ago

Sorry for the late reply. You would be better to try some compute shader that is irrelevant to MetaDrive and Panda3D to figure out whether it is the OpenGL/hardware problem. Then we can do further debugging on Panda3D and MetaDrive side.