temp

mani_skill2/envs/sapien_env.py

@@ -614,20 +614,26 @@ def _clear_sim_state(self):
     # -------------------------------------------------------------------------- #
 
     def step(self, action: Union[None, np.ndarray, Dict]):
+        print("gamer1")
         with sapien.profile("step_action"):
             self.step_action(action)
         self._elapsed_steps += 1
+        print("gamer2")
         # TODO (stao): I think evaluation should always occur first before generating observations
         # as evaluation is more likely to use privileged information whereas observations only sometimes should include privileged information
+        print("gamer3")
         with sapien.profile("get_obs"):
             obs = self.get_obs()
+        print("gamer4")
         info = self.get_info(obs=obs)
+        print("gamer5")
         reward = self.get_reward(obs=obs, action=action, info=info)
+        print("gamer6")
         terminated = info["success"]
         if self.num_envs == 1:
             terminated = terminated[0]
             reward = reward[0]
-
+        print("gamer7")
         if physx.is_gpu_enabled():
             return obs, reward, terminated, torch.Tensor(False), info
         else:

mani_skill2/utils/sapien_utils.py

@@ -71,12 +71,9 @@ def _unbatch(array: Union[Array, Sequence]):
         if isinstance(array, torch.Tensor):
             return array.squeeze(0)
     if isinstance(array, np.ndarray):
-        print('NDARRAY', array)
-        print(array.shape)
         if np.iterable(array) and array.shape[0] == 1:
             return array.squeeze(0)
     if isinstance(array, list):
-        print('LIST', array)
         if len(array) == 1:
             return array[0]
     return array