[IBR-2091] Convet action type to numpy array in select_action function

Author: isk03276

rl_algorithms/gail/agent.py

@@ -168,7 +168,10 @@ def train(self):
                 # gail reward (imitation reward)
                 gail_reward = compute_gail_reward(
                     self.learner.discriminator(
-                        (numpy2floattensor(state, self.learner.device), action)
+                        (
+                            numpy2floattensor(state, self.learner.device),
+                            numpy2floattensor(action, self.learner.device),
+                        )
                     )
                 )
 

rl_algorithms/ppo/agent.py

@@ -135,6 +135,7 @@ def select_action(self, state: np.ndarray) -> torch.Tensor:
         with torch.no_grad():
             state = numpy2floattensor(state, self.learner.device)
             selected_action, dist = self.learner.actor(state)
+            selected_action = selected_action.detach()
             log_prob = dist.log_prob(selected_action)
             value = self.learner.critic(state)
 
@@ -155,7 +156,7 @@ def select_action(self, state: np.ndarray) -> torch.Tensor:
                 self.values.append(value)
                 self.log_probs.append(_log_prob)
 
-        return selected_action
+        return selected_action.detach().cpu().numpy()
 
     def step(
         self, action: Union[np.ndarray, torch.Tensor]