multiprocessing_env.py 文件源码

def _step(self, action_n):
        observation_n, reward_n, done_n, info = self.env.step(action_n)
        # Pass along ID of potentially-done episode
        for i, info_i in enumerate(info['n']):
            info_i['vectorized.episode_id'] = self.episode_ids[i]

        done_i = np.argwhere(done_n).reshape(-1)
        if len(done_i):
            for i in done_i:
                self.extra_done.add(self.episode_ids[i])
                # Episode completed, so we bump its value
                self.episode_ids[i] += self.n
                if self.episode_limit is not None and self.episode_ids[i] > self.episode_limit:
                    logger.debug('Masking: index=%s episode_id=%s', i, self.episode_ids[i])
                    self.env.mask(i)
            self._set_done_to()

        # Pass along the number of contiguous episodes that are now done
        info['vectorized.done_to'] = self.done_to
        return observation_n, reward_n, done_n, info