diff --git a/mani_skill2/envs/scenes/tasks/pick.py b/mani_skill2/envs/scenes/tasks/pick.py
index 12a412043..68f5b1832 100644
--- a/mani_skill2/envs/scenes/tasks/pick.py
+++ b/mani_skill2/envs/scenes/tasks/pick.py
@@ -382,12 +382,13 @@ def compute_dense_reward(self, obs: Any, action: torch.Tensor, info: Dict):
 
             new_info = dict()
             for k in ["elapsed_steps", "success", "fail", "is_grasped", "robot_force", "robot_cumulative_force"]:
-                if isinstance(new_info[k], torch.Tensor):
-                    new_info[k] = info[k].clone()
-                elif isinstance(new_info[k], np.ndarray):
-                    new_info[k] = info[k].copy()
-                else:
-                    new_info[k] = info[k]
+                if k in info:
+                    if isinstance(new_info[k], torch.Tensor):
+                        new_info[k] = info[k].clone()
+                    elif isinstance(new_info[k], np.ndarray):
+                        new_info[k] = info[k].copy()
+                    else:
+                        new_info[k] = info[k]
 
             if torch.any(robot_too_far):
                 # prevent torso and arm moving too much