wrappers.py 文件源码

def _step(self, action):
        s, r, t, i = super()._step(action)
        assert isinstance(self.env, Env)
        assert isinstance(self.env.action_space, Box)
        l = self.env.action_space.low
        h = self.env.action_space.high
        m = h - l
        dif = (action - np.clip(action, l - self.slack * m, h + self.slack * m))
        i.setdefault('unwrapped_reward', r)
        r -= self.alpha * np.mean(np.square(dif / m))
        return s, r, t, i