diff --git a/omnisafe/algorithms/on_policy/base/policy_gradient.py b/omnisafe/algorithms/on_policy/base/policy_gradient.py
index ae3ec1d22..cd194e35a 100644
--- a/omnisafe/algorithms/on_policy/base/policy_gradient.py
+++ b/omnisafe/algorithms/on_policy/base/policy_gradient.py
@@ -15,7 +15,7 @@
 """Implementation of the Policy Gradient algorithm."""
 
 import time
-from typing import Dict, Tuple, Union
+from typing import Any, Dict, Tuple, Union
 
 import torch
 import torch.nn as nn
@@ -98,12 +98,11 @@ def _init_log(self) -> None:
             config=self._cfgs,
         )
 
-        obs_normalizer = self._env.save()['obs_normalizer']
-        what_to_save = {
-            'pi': self._actor_critic.actor,
-            'obs_normalizer': obs_normalizer,
-        }
-
+        what_to_save: Dict[str, Any] = {}
+        what_to_save['pi'] = self._actor_critic.actor
+        if self._cfgs.algo_cfgs.obs_normalize:
+            obs_normalizer = self._env.save()['obs_normalizer']
+            what_to_save['obs_normalizer'] = obs_normalizer
         self._logger.setup_torch_saver(what_to_save)
         self._logger.torch_save()
 
diff --git a/omnisafe/common/logger.py b/omnisafe/common/logger.py
index 73a6b41a1..0a7b56892 100644
--- a/omnisafe/common/logger.py
+++ b/omnisafe/common/logger.py
@@ -15,6 +15,7 @@
 """Implementation of the Logger."""
 
 import atexit
+import csv
 import os
 import time
 from collections import deque
@@ -96,7 +97,7 @@ def __init__(  # pylint: disable=too-many-arguments,too-many-locals
         self,
         output_dir: str,
         exp_name: str,
-        output_fname: str = 'progress.txt',
+        output_fname: str = 'progress.csv',
         verbose: bool = True,
         seed: int = 0,
         use_tensorboard: bool = True,
@@ -123,6 +124,7 @@ def __init__(  # pylint: disable=too-many-arguments,too-many-locals
             )
             atexit.register(self._output_file.close)
             self.log(f'Logging data to {self._output_file.name}', 'cyan', bold=True)
+            self._csv_writer = csv.writer(self._output_file)
 
         self._epoch: int = 0
         self._first_row: bool = True
@@ -277,9 +279,9 @@ def dump_tabular(self) -> None:
                 self._proc_bar.update(1)
 
             if self._first_row:
-                self._output_file.write(' '.join(self._current_row.keys()) + '\n')
+                self._csv_writer.writerow(self._current_row.keys())
                 self._first_row = False
-            self._output_file.write(' '.join(map(str, self._current_row.values())) + '\n')
+            self._csv_writer.writerow(self._current_row.values())
             self._output_file.flush()
 
             if self._use_tensorboard:
diff --git a/omnisafe/configs/on-policy/CPO.yaml b/omnisafe/configs/on-policy/CPO.yaml
index 36054c030..9caa93101 100644
--- a/omnisafe/configs/on-policy/CPO.yaml
+++ b/omnisafe/configs/on-policy/CPO.yaml
@@ -37,7 +37,7 @@ defaults:
     # batch size for each iteration
     batch_size: 16384
     # target kl divergence
-    target_kl: 0.01
+    target_kl: 0.02
     # entropy coefficient
     entropy_coef: 0.0
     # normalize reward
diff --git a/omnisafe/envs/wrapper.py b/omnisafe/envs/wrapper.py
index 0a774a1d8..70f658c9a 100644
--- a/omnisafe/envs/wrapper.py
+++ b/omnisafe/envs/wrapper.py
@@ -260,7 +260,7 @@ def step(
         action = self._old_min_action + (self._old_max_action - self._old_min_action) * (
             action - self._min_action
         ) / (self._max_action - self._min_action)
-        return super().step(action)
+        return super().step(action.numpy())
 
 
 class Unsqueeze(Wrapper):
@@ -283,6 +283,7 @@ def __init__(self, env: CMDP) -> None:
     def step(
         self, action: torch.Tensor
     ) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor, Dict]:
+        action = action.squeeze(0)
         obs, reward, cost, terminated, truncated, info = super().step(action)
         obs, reward, cost, terminated, truncated = map(
             lambda x: x.unsqueeze(0), (obs, reward, cost, terminated, truncated)
diff --git a/omnisafe/utils/math.py b/omnisafe/utils/math.py
index b8e936f5f..0e43c467c 100644
--- a/omnisafe/utils/math.py
+++ b/omnisafe/utils/math.py
@@ -73,7 +73,7 @@ def gaussian_kl(
             \mu_q) - k + log(\frac{det(\Sigma_p)}{det(\Sigma_q)}))
 
         where :math:`\mu_p` and :math:`\mu_q` are the mean of :math:`p` and :math:`q`, respectively.
-        :math:`\Sigma_p` and :math:`\Sigma_q` are the covariance of :math:`p` and :math:`q`, respectively.
+        :math:`\Sigma_p` and :math:`\Sigma_q` are the co-variance of :math:`p` and :math:`q`, respectively.
         :math:`k` is the dimension of the distribution.
 
     For more details,
@@ -83,8 +83,8 @@ def gaussian_kl(
     Args:
         mean_p (torch.Tensor): mean of the first distribution, shape (B, n)
         mean_q (torch.Tensor): mean of the second distribution, shape (B, n)
-        var_p (torch.Tensor): covariance of the first distribution, shape (B, n, n)
-        var_q (torch.Tensor): covariance of the second distribution, shape (B, n, n)
+        var_p (torch.Tensor): co-variance of the first distribution, shape (B, n, n)
+        var_q (torch.Tensor): co-variance of the second distribution, shape (B, n, n)
     """
     len_q = var_q.size(-1)
     mean_p = mean_p.unsqueeze(-1)  # (B, n, 1)