diff --git a/rllib/policy/policy.py b/rllib/policy/policy.py
index 214db1fb45ba..592a9bc7ba44 100644
--- a/rllib/policy/policy.py
+++ b/rllib/policy/policy.py
@@ -289,9 +289,12 @@ def from_state(state: PolicyState) -> "Policy":
 
         # Create the new policy.
         new_policy = pol_spec.policy_class(
-            observation_space=pol_spec.observation_space,
-            action_space=pol_spec.action_space,
-            config=pol_spec.config,
+            # Note(jungong) : we are intentionally not using keyward arguments here
+            # because some policies name the observation space parameter obs_space,
+            # and some others name it observation_space.
+            pol_spec.observation_space,
+            pol_spec.action_space,
+            pol_spec.config,
         )
 
         # Set the new policy's state (weights, optimizer vars, exploration state,