SIGSEGV on env.step for multiple renderings.

[chasebos91]

Hey there,
I'm experiencing a segmentation fault after I call env.step when there are multiple renderings open. I saw a closed thread on openai for the same issue, but I'm unsure if the bug still persists across some of their environments.

[maximecb]

Hi Chase. I think this is an issue related to the PyQT renderer. Can you provide a code snippet to reproduce the issue?

[chasebos91]

Hey thanks for the quick response. Here's where I've instantiated multiple envs:

import gym
from gym_minigrid.wrappers import RGBImgObsWrapper
from PEARL import PEARL

env0 = RGBImgObsWrapper(gym.make("MiniGrid-Empty-Random-5x5-v0"))
env00 = RGBImgObsWrapper(gym.make("MiniGrid-FourRooms-v0"))
env1 = RGBImgObsWrapper(gym.make("MiniGrid-MultiRoom-N2-S4-v0"))
env2 = RGBImgObsWrapper(gym.make("MiniGrid-MultiRoom-N4-S5-v0"))
train_tasks = [env00, env0, env1, env2]
test_env = RGBImgObsWrapper(gym.make("MiniGrid-MultiRoom-N6-v0"))
test_tasks = [test_env]
#train
pearl = PEARL(train_tasks[0], train_tasks, test_tasks)
pearl.train_algo()

#########################################################################

And here's where I'm rendering each environment to initialize them into replay buffers:

def get_path(self, env, policy, max_length):
           ...
	s = env.reset()
	env.render()
	for i in range(max_length):
		a = policy.get_action(s)
		a = self.return_action(env, a)
		s_p, r, t, e_info = env.step(a)
                   ... # the line above breaks it

########################################################################

Thanks again for looking into it. I'll look up the issue in ref to PyQT in the meantime.

[maximecb]

One temporary fix that could work is to run the environments in forked processes.

Otherwise though, next week I will start looking at implementing a renderer based on NumPy (see #76). I would ideally like to eliminate the PyQT dependency. You're not the only one to run into issues. If you're available to help me test next week that could he useful :)

[chasebos91]

Hey, thanks that's awesome. I'd be happy to help, though I'm working on the implementation so it will definitely not be bug-free. If it's still helpful though, that would be great!

[maximecb]

Hi @chasebos91. I have the new version of MiniGrid that uses NumPy to produce graphics on a new branch maximecb-numpy-render, and matplotlib for rendering windows. Could you test it out and let me know if it fixes the problem you encountered or if you run into any other issues?

[chasebos91]

So far so good! Will let you know as I keep working on it. Thank you!!
** EDIT: Sorry nvm I'm a GitHub noob, can you tell me how I can access the branch?

[maximecb]

You can access it by first doing a git pull, then:

git checkout maximecb-numpy-render`

Then to install the package in editable mode:

pip3 install -e .

[maximecb]

This is now pushed to the master branch as well :)