运行Super Mario Bros RL代码遇JoypadSpace.reset() TypeError，如何修复？

代码示例 #

from nes_py.wrappers import JoypadSpace
import gym
import gym_super_mario_bros
from gym_super_mario_bros.actions import SIMPLE_MOVEMENT
from gym.wrappers import GrayScaleObservation
from stable_baselines3.common.vec_env import VecFrameStack,DummyVecEnv
from matplotlib import pyplot as plt

env = gym_super_mario_bros.make('SuperMarioBros-v0',apply_api_compatibility=True,render_mode="human")
env = JoypadSpace(env, SIMPLE_MOVEMENT)
env = GrayScaleObservation(env,keep_dim=True)
env = DummyVecEnv([lambda:env])
env = VecFrameStack(env,4,channels_order='last')
state = env.reset()

报错信息 #

TypeError: JoypadSpace.reset() got an unexpected keyword argument 'seed'

修复方案 #

方案1：降级Gym版本到0.21.x #

Gym 0.26+版本在reset方法中新增了seed等参数，但旧版nes-py的JoypadSpace未适配该API。执行以下命令降级：

pip install gym==0.21.0

方案2：自定义兼容版JoypadSpace包装器 #

如果想保留高版本Gym，可以自定义包装器覆盖reset方法，忽略不支持的参数：

from nes_py.wrappers import JoypadSpace as OriginalJoypadSpace
import gym
import gym_super_mario_bros
from gym_super_mario_bros.actions import SIMPLE_MOVEMENT
from gym.wrappers import GrayScaleObservation
from stable_baselines3.common.vec_env import VecFrameStack,DummyVecEnv
from matplotlib import pyplot as plt

# 自定义兼容的JoypadSpace
class JoypadSpace(OriginalJoypadSpace):
    def reset(self, **kwargs):
        # 忽略seed等不支持的参数，调用原方法
        return super().reset()

# 正常初始化环境，使用自定义的JoypadSpace
env = gym_super_mario_bros.make('SuperMarioBros-v0',apply_api_compatibility=True,render_mode="human")
env = JoypadSpace(env, SIMPLE_MOVEMENT)
env = GrayScaleObservation(env,keep_dim=True)
env = DummyVecEnv([lambda:env])
env = VecFrameStack(env,4,channels_order='last')
state = env.reset()

方案3：升级nes-py到最新版本 #

检查nes-py是否已修复该兼容性问题，执行升级命令：

pip install --upgrade nes-py

内容的提问来源于stack exchange，提问作者abzd