diff --git a/README.md b/README.md
index 429839f2..b258e414 100644
--- a/README.md
+++ b/README.md
@@ -10,6 +10,8 @@
 [![Donate](https://img.shields.io/badge/donate-Liberapay-yellow)](https://liberapay.com/TensorforceTeam/donate)
 
 
+**This project is not maintained any longer!**
+
 
 #### Introduction
 
diff --git a/docs/requirements.txt b/docs/requirements.txt
index 31627ba2..45ed003a 100644
--- a/docs/requirements.txt
+++ b/docs/requirements.txt
@@ -1,5 +1,5 @@
 m2r
-mistune == 0.8.4  # temporary
+mistune == 2.0.3  # temporary
 msgpack
 msgpack-numpy
 recommonmark
diff --git a/requirements-all.txt b/requirements-all.txt
index 5b74667e..bf7b0e3b 100644
--- a/requirements-all.txt
+++ b/requirements-all.txt
@@ -1,11 +1,11 @@
-gym >= 0.21.0
+gym >= 0.21.0, <0.23
 h5py >= 3.6.0
 matplotlib >= 3.5.1
 msgpack >= 1.0.3
 msgpack-numpy >= 0.4.7.1
 numpy ~= 1.21.5
 Pillow >= 9.0.0
-tensorflow == 2.8.0
+tensorflow == 2.12.1
 tqdm >= 4.62.3
 tensorflow-addons >= 0.15.0
 hpbandster >= 0.7.4
diff --git a/requirements.txt b/requirements.txt
index 5e53df18..1d9ed7f1 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -1,9 +1,9 @@
-gym >= 0.21.0
+gym >= 0.21.0, <0.23
 h5py >= 3.6.0
 matplotlib >= 3.5.1
 msgpack >= 1.0.3
 msgpack-numpy >= 0.4.7.1
 numpy ~= 1.21.5
 Pillow >= 9.0.0
-tensorflow == 2.8.0
+tensorflow == 2.12.1
 tqdm >= 4.62.3
diff --git a/tensorforce/core/optimizers/tf_optimizer.py b/tensorforce/core/optimizers/tf_optimizer.py
index ecd48d52..a0568149 100755
--- a/tensorforce/core/optimizers/tf_optimizer.py
+++ b/tensorforce/core/optimizers/tf_optimizer.py
@@ -22,14 +22,14 @@
 
 
 tensorflow_optimizers = dict(
-    adadelta=tf.keras.optimizers.Adadelta,
-    adagrad=tf.keras.optimizers.Adagrad,
-    adam=tf.keras.optimizers.Adam,
-    adamax=tf.keras.optimizers.Adamax,
-    ftrl=tf.keras.optimizers.Ftrl,
-    nadam=tf.keras.optimizers.Nadam,
-    rmsprop=tf.keras.optimizers.RMSprop,
-    sgd=tf.keras.optimizers.SGD
+    adadelta=tf.keras.optimizers.legacy.Adadelta,
+    adagrad=tf.keras.optimizers.legacy.Adagrad,
+    adam=tf.keras.optimizers.legacy.Adam,
+    adamax=tf.keras.optimizers.legacy.Adamax,
+    ftrl=tf.keras.optimizers.legacy.Ftrl,
+    nadam=tf.keras.optimizers.legacy.Nadam,
+    rmsprop=tf.keras.optimizers.legacy.RMSprop,
+    sgd=tf.keras.optimizers.legacy.SGD
 )
 
 
diff --git a/tensorforce/core/utils/dicts.py b/tensorforce/core/utils/dicts.py
index dcc95638..7dab1f84 100644
--- a/tensorforce/core/utils/dicts.py
+++ b/tensorforce/core/utils/dicts.py
@@ -32,10 +32,7 @@ def __init__(self, arg=None, *, value_type=None, overwrite=None, **kwargs):
         )
 
     def __setattr__(self, name, value):
-        if name.startswith('_self_'):
-            super(NestedDict, self).__setattr__(name, value)
-        else:
-            super().__setattr__(name, value)
+        super(NestedDict, self).__setattr__(name, value)
 
     def __setitem__(self, key, value):
         if key is None:
diff --git a/tensorforce/environments/openai_gym.py b/tensorforce/environments/openai_gym.py
index 575e64ac..898e7d3c 100644
--- a/tensorforce/environments/openai_gym.py
+++ b/tensorforce/environments/openai_gym.py
@@ -113,7 +113,7 @@ def create_level(cls, level, max_episode_steps, reward_threshold, **kwargs):
         # Modified specification
         if requires_register:
             entry_point = gym.envs.registry.env_specs[level].entry_point
-            _kwargs = dict(gym.envs.registry.env_specs[level]._kwargs)
+            _kwargs = dict(gym.envs.registry.env_specs[level].kwargs)
             nondeterministic = gym.envs.registry.env_specs[level].nondeterministic
 
             if '-v' in level and level[level.rindex('-v') + 2:].isdigit():
diff --git a/tensorforce/execution/runner.py b/tensorforce/execution/runner.py
index 261aa5d6..15f75eb9 100644
--- a/tensorforce/execution/runner.py
+++ b/tensorforce/execution/runner.py
@@ -275,6 +275,7 @@ def run(
                 additional information (averaged over number of episodes given via mean_horizon):
                 <ul>
                 <li>return &ndash; cumulative episode return</li>
+                <li>std &ndash; standard deviation of cumulative episode return</li>
                 <li>ts/ep &ndash; timesteps per episode</li>
                 <li>sec/ep &ndash; seconds per episode</li>
                 <li>ms/ts &ndash; milliseconds per timestep</li>
@@ -418,13 +419,13 @@ def boolean_callback(runner, parallel):
                 # Episode-based tqdm (default option if both num_episodes and num_timesteps set)
                 assert self.num_episodes != float('inf')
                 bar_format = (
-                    '{l_bar}{bar}| {n_fmt}/{total_fmt} [{elapsed}, return={postfix[0]:.2f}, ts/ep='
-                    '{postfix[1]}, sec/ep={postfix[2]:.2f}, ms/ts={postfix[3]:.1f}, agent='
-                    '{postfix[4]:.1f}%]'
+                    '{l_bar}{bar}| {n_fmt}/{total_fmt} [{elapsed}, return={postfix[0]:.2f}, stddev={postfix[1]:.2f}, ts/ep='
+                    '{postfix[2]}, sec/ep={postfix[3]:.2f}, ms/ts={postfix[4]:.1f}, agent='
+                    '{postfix[5]:.1f}%]'
                 )
-                postfix = [0.0, 0, 0.0, 0.0, 0.0]
+                postfix = [0.0, 0.0, 0, 0.0, 0.0, 0.0]
                 if self.is_environment_remote:
-                    bar_format = bar_format[:-1] + ', comm={postfix[5]:.1f}%]'
+                    bar_format = bar_format[:-1] + ', comm={postfix[6]:.1f}%]'
                     postfix.append(0.0)
 
                 self.tqdm = tqdm(
@@ -436,7 +437,9 @@ def boolean_callback(runner, parallel):
                 def tqdm_callback(runner, parallel):
                     if len(runner.evaluation_returns) > 0:
                         mean_return = float(np.mean(runner.evaluation_returns[-mean_horizon:]))
+                        std_return = float(np.std(runner.evaluation_returns[-mean_horizon:]))
                         runner.tqdm.postfix[0] = mean_return
+                        runner.tqdm.postfix[1] = std_return
                     if len(runner.episode_timesteps) > 0:
                         mean_ts_per_ep = int(np.mean(runner.episode_timesteps[-mean_horizon:]))
                         mean_sec_per_ep = float(np.mean(runner.episode_seconds[-mean_horizon:]))
@@ -451,15 +454,15 @@ def tqdm_callback(runner, parallel):
                             mean_rel_agent = mean_agent_sec * 100.0 / mean_sec_per_ep
                         except ZeroDivisionError:
                             mean_rel_agent = 0.0
-                        runner.tqdm.postfix[1] = mean_ts_per_ep
-                        runner.tqdm.postfix[2] = mean_sec_per_ep
-                        runner.tqdm.postfix[3] = mean_ms_per_ts
-                        runner.tqdm.postfix[4] = mean_rel_agent
+                        runner.tqdm.postfix[2] = mean_ts_per_ep
+                        runner.tqdm.postfix[3] = mean_sec_per_ep
+                        runner.tqdm.postfix[4] = mean_ms_per_ts
+                        runner.tqdm.postfix[5] = mean_rel_agent
                     if runner.is_environment_remote and len(runner.episode_env_seconds) > 0:
                         mean_env_sec = float(np.mean(runner.episode_env_seconds[-mean_horizon:]))
                         mean_rel_comm = (mean_agent_sec + mean_env_sec) * 100.0 / mean_sec_per_ep
                         mean_rel_comm = 100.0 - mean_rel_comm
-                        runner.tqdm.postfix[5] = mean_rel_comm
+                        runner.tqdm.postfix[6] = mean_rel_comm
                     runner.tqdm.update(n=(runner.episodes - runner.tqdm_last_update))
                     runner.tqdm_last_update = runner.episodes
                     return inner_callback(runner, parallel)