Failed to initialize OpenGL on evaluation

[erick-alv]

Hello, I am trying to setup the project in my own system. I am not using Docker since my mujoco key ist not valid for that. I have Ubuntu 18.04 and installed CUDA 10.0 and cudnn 7.6.5. However I am getting following error:

Doing evaluation of initial policy on 10 trials with randomly sampled tasks
====== finished initialization ======
Found 4 GPUs for rendering. Using device 1.
libEGL warning: DRI2: failed to create dri screen
libEGL warning: Not allowed to force software rendering when API explicitly selects a hardware device.
libEGL warning: DRI2: failed to create dri screen
Could not initialize EGL
Traceback (most recent call last):
File "/home/erick/RL/meld/meld/scripts/run_meld.py", line 1023, in
app.run(main)
File "/home/erick/anaconda3/envs/meld/lib/python3.6/site-packages/absl/app.py", line 299, in run
_run_main(main, args)
File "/home/erick/anaconda3/envs/meld/lib/python3.6/site-packages/absl/app.py", line 250, in _run_main
sys.exit(main(argv))
File "/home/erick/RL/meld/meld/scripts/run_meld.py", line 1020, in main
train_eval(FLAGS.root_dir, FLAGS.experiment_name, seed=FLAGS.seed)
File "/home/erick/anaconda3/envs/meld/lib/python3.6/site-packages/gin/config.py", line 1032, in wrapper
utils.augment_exception_message_and_reraise(e, err_str)
File "/home/erick/anaconda3/envs/meld/lib/python3.6/site-packages/gin/utils.py", line 49, in augment_exception_message_and_reraise
six.raise_from(proxy.with_traceback(exception.traceback), None)
File "", line 3, in raise_from
File "/home/erick/anaconda3/envs/meld/lib/python3.6/site-packages/gin/config.py", line 1009, in wrapper
return fn(*new_args, **new_kwargs)
File "/home/erick/RL/meld/meld/scripts/run_meld.py", line 809, in train_eval
log_image_observations=log_image_observations,
File "/home/erick/RL/meld/meld/utils/eval_summaries.py", line 93, in perform_eval_and_summaries_meld
time_step = env.reset()
File "/home/erick/RL/meld/agents/tf_agents/environments/py_environment.py", line 141, in reset
self._current_time_step = self._reset()
File "/home/erick/RL/meld/meld/environments/video_wrapper.py", line 22, in _reset
time_step = self._env.reset()
File "/home/erick/RL/meld/agents/tf_agents/environments/py_environment.py", line 141, in reset
self._current_time_step = self._reset()
File "/home/erick/RL/meld/agents/tf_agents/environments/wrappers.py", line 99, in _reset
return self._env.reset()
File "/home/erick/RL/meld/agents/tf_agents/environments/py_environment.py", line 141, in reset
self._current_time_step = self._reset()
File "/home/erick/RL/meld/agents/tf_agents/environments/gym_wrapper.py", line 177, in _reset
observation = self._gym_env.reset()
File "/home/erick/RL/meld/agents/tf_agents/environments/py_environment.py", line 141, in reset
self._current_time_step = self._reset()
File "/home/erick/RL/meld/meld/environments/gym_wrappers.py", line 153, in _reset
return self._modify_observation(observation)
File "/home/erick/RL/meld/meld/environments/gym_wrappers.py", line 124, in _modify_observation
image = self._env.sim.render(**self._render_kwargs)[::-1, :, :]
File "mjsim.pyx", line 156, in mujoco_py.cymj.MjSim.render
File "mjsim.pyx", line 158, in mujoco_py.cymj.MjSim.render
File "mjrendercontext.pyx", line 45, in mujoco_py.cymj.MjRenderContext.init
File "mjrendercontext.pyx", line 113, in mujoco_py.cymj.MjRenderContext._setup_opengl_context
File "opengl_context.pyx", line 130, in mujoco_py.cymj.OffscreenOpenGLContext.init
RuntimeError: Failed to initialize OpenGL
In call to configurable 'train_eval' (<function train_eval at 0x7f81002c5a60>)

I read the issue openai/mujoco-py#187 to try some of the suggestion but none solves the error. As far as I understand, when calling self._env.sim.render(**self._render_kwargs) LD_PRELOAD should not have a value since some of the parameters are passed. I tried with an empty LD_PRELOAD and with LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libGLEW.so:/usr/lib/nvidia-xxx/libGL.so (for that I created the empty folder /usr/lib/nvidia-xxx since my system does not have one) but it does not work. Do you have any suggestion?

[newuhe]

I have the same problem. Have you solved it? @erick-alv

[mingyang-wang26]

Same problem here and also tried to change the LD_PRELOAD value but still can not fix it.

[NirViaje]

Seems like I solved this problem in #7 , the following is the guideline:

install the newest MuJoCo and mujoco-py, to enable headless rendering (in my case)

• https://github.com/openai/mujoco-py#install-mujoco

install the dedicated version of tensorflow 1.15 for CUDA11 (if you are on A100 or rtx3xxx in my case)

• https://github.com/NVIDIA/tensorflow