inverted_pendulum.py 文件源码

def step(self, action):
        action = self._discrete_actions[action[0]]
        action += np.random.uniform(-10., 10.)
        sa = np.append(self._state, action)
        new_state = odeint(self._dpds, sa, [0, self._dt])

        self._state = new_state[-1, :-1]
        self._state[0] = self._range_pi(self._state[0])

        if np.abs(self._state[0]) > np.pi / 2.:
            reward = -1
            absorbing = True
        else:
            reward = 0
            absorbing = False

        return self._state, reward, absorbing, {}