StableBaselines3 - NotImplementedError：不支持观察空间

Question

我正在尝试在

cleanrl

环境上运行

Pendulum-v1

。我通过转到 here 并将默认值

env-id

更改为

parser.add_argument("--env-id", type=str, default="Pendulum-v1",help="the id of the environment")

来做到这一点

但是，我不断收到错误 -

NotImplementedError: Box([-1. -1. -8.], [1. 1. 8.], (3,), <class 'numpy.float32'>) observation space is not supported

因此，我将这个错误调试到从

SB3

导入的ReplayBuffer中。这是问题函数 -

def get_obs_shape(
    observation_space: spaces.Space,
) -> Union[Tuple[int, ...], Dict[str, Tuple[int, ...]]]:
    """
    Get the shape of the observation (useful for the buffers).

    :param observation_space:
    :return:
    """
    if isinstance(observation_space, spaces.Box):
        return observation_space.shape
    elif isinstance(observation_space, spaces.Discrete):
        # Observation is an int
        return (1,)
    elif isinstance(observation_space, spaces.MultiDiscrete):
        # Number of discrete features
        return (int(len(observation_space.nvec)),)
    elif isinstance(observation_space, spaces.MultiBinary):
        # Number of binary features
        return observation_space.shape
    elif isinstance(observation_space, spaces.Dict):
        return {key: get_obs_shape(subspace) for (key, subspace) in observation_space.spaces.items()}  # type: ignore[misc]

    else:
        raise NotImplementedError(f"{observation_space} observation space is not supported")

我尝试打印

observation_space

形状，这就是我得到的 -

Box([-1. -1. -8.], [1. 1. 8.], (3,), float32)

。知道为什么 SB3 不接受我的观察形状吗？

Answer 1

我觉得问题可能出在gymnasium或者gym套餐的版本上。我只是发现gymnasium==0.29.0 和gym==0.24 中的spaces.Box() 类型不相同。也许您将 sb3 与gymnasium一起使用，但您的 Pendulum-v1 环境正在使用gym。