diff --git a/test/envs/test_auto_reset.py b/test/envs/test_auto_reset.py
index edc56efbb81..50fb56d8a07 100644
--- a/test/envs/test_auto_reset.py
+++ b/test/envs/test_auto_reset.py
@@ -99,7 +99,7 @@ def test_native_auto_reset_step_and_maybe_reset(self):
             )
 
         assert tensordict["next", "done"].all()
-        assert (tensordict["next", "observation"] == 0).all()
+        assert (tensordict["next", "observation"] == tensordict["observation"]).all()
         assert not tensordict_["done"].any()
         assert tensordict_["is_init"].all()
 
diff --git a/torchrl/envs/common.py b/torchrl/envs/common.py
index 117953ac1de..7a0cc8362d1 100644
--- a/torchrl/envs/common.py
+++ b/torchrl/envs/common.py
@@ -3693,8 +3693,9 @@ def _native_autoreset_set_invalid_next_observation(
             if obs is None or not isinstance(obs, torch.Tensor):
                 continue
             reset_observations[obs_key] = obs.clone()
+            current_obs = tensordict.get(obs_key, None)
             if obs.is_floating_point():
-                obs[done] = torch.nan
+                obs[done] = current_obs[done]
             else:
                 obs[done] = 0
         return reset_observations
diff --git a/torchrl/envs/libs/gym.py b/torchrl/envs/libs/gym.py
index b4ab20f40d4..bcf29404211 100644
--- a/torchrl/envs/libs/gym.py
+++ b/torchrl/envs/libs/gym.py
@@ -955,8 +955,8 @@ class GymWrapper(GymLikeEnv, metaclass=_GymAsyncMeta):
             observation returned by the environment as the next root observation
             instead of calling reset from
             :meth:`~torchrl.envs.EnvBase.step_and_maybe_reset`. The terminal
-            ``"next"`` observation is still invalid and filled with ``NaN`` for
-            floating point observations. Defaults to ``False``.
+            ``"next"`` observation is replaced with the current (root)
+            observation. Defaults to ``False``.
 
     Attributes:
         available_envs (List[str]): a list of environments to build.
diff --git a/torchrl/envs/transforms/_misc.py b/torchrl/envs/transforms/_misc.py
index 296a8b35ad1..b0588dd9a50 100644
--- a/torchrl/envs/transforms/_misc.py
+++ b/torchrl/envs/transforms/_misc.py
@@ -408,8 +408,8 @@ class VecGymEnvTransform(Transform):
         native_autoreset (bool, optional): if ``True``, leaves the native
             auto-reset observation available to the environment wrapper so it
             can be cloned into the next root observation, while the terminal
-            floating point ``"next"`` observation is marked with ``NaN``.
-            Defaults to ``False``.
+            ``"next"`` observation is replaced with the current (root)
+            observation. Defaults to ``False``.
 
     .. note:: In general, this class should not be handled directly. It is
         created whenever a vectorized environment is placed within a :class:`GymWrapper`.