Bug for VizdoomEnv

[caozhenxiang-kouji]

There seems to be a serious bug with your VizdoomEnv.

When I use the following code to build a doom environment:

import gymnasium
from vizdoom import gymnasium_wrapper

env = gymnasium.make("VizdoomHealthGathering-v0", frame_skip=4)

And I try to use the following code to step forward for the environment:

obs, reward, terminated, truncated, info = env.step(env.action_space.sample())

And my agent is always moving forward and never turn left or right.

It seems that if the env.step function takes in an int type, the env_action in env.step function can produce correct results like [0,0,0],[0,0,1],[0,1,0],[1,0,0]. However when the env.step function takes in a numpy.int64 type, the env_action will turn into [0,0,0],[1,1,1],[2,2,2],[3,3,3], which forces the agent to keep moving forward.

My way of solving this bug is to write another gym.Wrapper to overwrite the step function:

class ChangeAction(gym.Wrapper):
    def step(self, action):
        action = int(action)
        return self.env.step(action)

It doesn't seem to be an elegant way since every time you make a environment you should wrap it again. But it works fine.

Hopes you can fix this bug soon!

[mwydmuch]

Hi @caozhenxiang-kouji, thank you very much for the report; this is indeed a serious bug! The fix will be merged soon, and a new patch release should be available on PyPI in a few hours.

[mwydmuch]

I released new version 1.2.3 that fixes the problem, it should be deployed to PyPI in ~2-3h. Once again, thanks for reporting the problem.

[caozhenxiang-kouji]

Great, thanks