diff --git a/isaaclab_experiments/isaaclab_fpo/isaaclab_fpo/runners/on_policy_runner.py b/isaaclab_experiments/isaaclab_fpo/isaaclab_fpo/runners/on_policy_runner.py
index 9fa63bb..b86f40b 100644
--- a/isaaclab_experiments/isaaclab_fpo/isaaclab_fpo/runners/on_policy_runner.py
+++ b/isaaclab_experiments/isaaclab_fpo/isaaclab_fpo/runners/on_policy_runner.py
@@ -169,6 +169,12 @@ def learn(self, num_learning_iterations: int, init_at_random_ep_len: bool = Fals
         obs, extras = self.env.get_observations()
         privileged_obs = extras["observations"].get(self.privileged_obs_type, obs)
         obs, privileged_obs = obs.to(self.device), privileged_obs.to(self.device)
+        # Normalize before the first act() call (fixes #6).
+        obs = self.obs_normalizer(obs)
+        if self.privileged_obs_type is not None:
+            privileged_obs = self.privileged_obs_normalizer(privileged_obs)
+        else:
+            privileged_obs = obs
         self.train_mode()  # switch to train mode (for dropout for example)
 
         # Book keeping