Method forward_policy

roach/models/ppo_policy.py:175–180

(self, obs_dict: Dict[str, np.ndarray])

        return values

    def forward_policy(self, obs_dict: Dict[str, np.ndarray]) -> np.ndarray:
        with th.no_grad():
            obs_tensor_dict = dict([(k, th.as_tensor(v).to(self.device)) for k, v in obs_dict.items()])
            features, cnn_feature = self._get_features(**obs_tensor_dict)
            distribution, mu, sigma = self._get_action_dist_from_features(features)
        return mu, sigma

    def scale_action(self, action: th.Tensor, eps=1e-7) -> th.Tensor:
        # input action \in [a_low, a_high]