diff --git a/benchmarks/acc.py b/benchmarks/acc.py
index f1118a3f..bf47244f 100644
--- a/benchmarks/acc.py
+++ b/benchmarks/acc.py
@@ -50,12 +50,12 @@ def step(self, action: np.ndarray) -> \
                 self.observation_space.low,
                 self.observation_space.high)
         reward = 2.0 + x if x < 0 else -10
-        done = bool(x >= 0) or self.steps > self._max_episode_steps
+        done = self.steps >= self._max_episode_steps
         self.steps += 1
         return self.state, reward, done, {}
 
     def predict_done(self, state: np.ndarray) -> bool:
-        return state[0] >= 0
+        return False
 
     def seed(self, seed: int):
         self.action_space.seed(seed)
diff --git a/benchmarks/car_racing.py b/benchmarks/car_racing.py
index ddd29a17..4a359744 100644
--- a/benchmarks/car_racing.py
+++ b/benchmarks/car_racing.py
@@ -57,8 +57,7 @@ def step(self, action: np.ndarray) -> \
             reward = -(6.0 + abs(x - 3.0) + abs(y - 3.0))
 
         done = self.corner and x <= 0.0 and y <= 0.0
-        done = done or self.steps >= self._max_episode_steps or \
-            self.unsafe(self.state)
+        done = done or self.steps >= self._max_episode_steps
         self.steps += 1
 
         return self.state, reward, done, {}
@@ -72,5 +71,5 @@ def seed(self, seed: int):
         self.init_space.seed(seed)
 
     def unsafe(self, state: np.ndarray) -> bool:
-        return self.state[0] >= 1.0 and self.state[0] <= 2.0 and \
-            self.state[1] >= 1.0 and self.state[1] <= 2.0
+        return state[0] >= 1.0 and state[0] <= 2.0 and \
+            state[1] >= 1.0 and state[1] <= 2.0
diff --git a/benchmarks/mid_obstacle.py b/benchmarks/mid_obstacle.py
index a6888615..0bd6ce0a 100644
--- a/benchmarks/mid_obstacle.py
+++ b/benchmarks/mid_obstacle.py
@@ -49,7 +49,7 @@ def step(self, action: np.ndarray) -> \
         reward = -(abs(x - 3.0) + abs(y - 3.0))
 
         done = x >= 3.0 and y >= 3.0
-        done = done or self.steps > self._max_episode_steps
+        done = done or self.steps >= self._max_episode_steps
         self.steps += 1
 
         return self.state, reward, done, {}
@@ -63,5 +63,5 @@ def seed(self, seed: int):
         self.init_space.seed(seed)
 
     def unsafe(self, state: np.ndarray) -> bool:
-        return self.state[0] >= 1.0 and self.state[0] <= 2.0 and \
-            self.state[1] >= 1.0 and self.state[1] <= 2.0
+        return state[0] >= 1.0 and state[0] <= 2.0 and \
+            state[1] >= 1.0 and state[1] <= 2.0
diff --git a/benchmarks/mountain_car.py b/benchmarks/mountain_car.py
index 5641bcc1..55cb3272 100644
--- a/benchmarks/mountain_car.py
+++ b/benchmarks/mountain_car.py
@@ -54,7 +54,7 @@ def step(self, action: np.ndarray) -> \
         return self.state, reward, done, {}
 
     def predict_done(self, state: np.ndarray) -> bool:
-        return state[0] >= 0.6 or state[0] < -np.pi / 3
+        return state[0] >= 0.6
 
     def seed(self, seed: int):
         self.action_space.seed(seed)
@@ -62,4 +62,4 @@ def seed(self, seed: int):
         self.init_space.seed(seed)
 
     def unsafe(self, state: np.ndarray) -> bool:
-        return self.state[0] < self.safe_limit
+        return state[0] < self.safe_limit
diff --git a/benchmarks/noisy_road.py b/benchmarks/noisy_road.py
index 650dff2f..586aa503 100644
--- a/benchmarks/noisy_road.py
+++ b/benchmarks/noisy_road.py
@@ -46,8 +46,7 @@ def step(self, action: np.ndarray) -> \
                              self.observation_space.high)
 
         reward = -abs(x - 3.0)
-        done = x >= 3.0 or self.steps >= self._max_episode_steps or \
-            self.unsafe(self.state)
+        done = x >= 3.0 or self.steps >= self._max_episode_steps
         self.steps += 1
 
         return self.state, reward, done, {}
diff --git a/benchmarks/noisy_road_2d.py b/benchmarks/noisy_road_2d.py
index ba03b18f..8a2d8b92 100644
--- a/benchmarks/noisy_road_2d.py
+++ b/benchmarks/noisy_road_2d.py
@@ -86,8 +86,7 @@ def step(self, action: np.ndarray) -> \
 
         reward = -(abs(x - 3.0) + abs(y - 3.0))
         done = x >= 3.0 and y >= 3.0
-        done = done or self.steps >= self._max_episode_steps or \
-            self.unsafe(self.state)
+        done = done or self.steps >= self._max_episode_steps
         self.steps += 1
 
         return self.state, reward, done, {}
@@ -102,4 +101,4 @@ def seed(self, seed: int):
         self.rng = np.random.default_rng(np.random.PCG64(seed))
 
     def unsafe(self, state: np.ndarray) -> bool:
-        return state[0]**2 + state[1]**2 >= self.max_speed**2
+        return state[2]**2 + state[3]**2 >= self.max_speed**2
diff --git a/benchmarks/obstacle.py b/benchmarks/obstacle.py
index 7be9481e..bb648486 100644
--- a/benchmarks/obstacle.py
+++ b/benchmarks/obstacle.py
@@ -49,8 +49,7 @@ def step(self, action: np.ndarray) -> \
         reward = -(abs(x - 3.0) + abs(y - 3.0))
 
         done = x >= 3.0 and y >= 3.0
-        done = done or self.steps >= self._max_episode_steps or \
-            self.unsafe(self.state)
+        done = done or self.steps >= self._max_episode_steps
         self.steps += 1
 
         return self.state, reward, done, {}
@@ -64,5 +63,5 @@ def seed(self, seed: int):
         self.init_space.seed(seed)
 
     def unsafe(self, state: np.ndarray) -> bool:
-        return self.state[0] >= 0.0 and self.state[0] <= 1.0 and \
-            self.state[1] >= 2.0 and self.state[1] <= 3.0
+        return state[0] >= 0.0 and state[0] <= 1.0 and \
+            state[1] >= 2.0 and state[1] <= 3.0
diff --git a/benchmarks/pendulum.py b/benchmarks/pendulum.py
index f0bba866..1a8f7f05 100644
--- a/benchmarks/pendulum.py
+++ b/benchmarks/pendulum.py
@@ -52,13 +52,12 @@ def step(self, action: np.ndarray) -> \
                 self.observation_space.low,
                 self.observation_space.high)
         reward = -abs(theta)
-        done = bool(abs(theta) >= 0.4) or \
-            self.steps >= self._max_episode_steps or self.unsafe(self.state)
+        done = self.steps >= self._max_episode_steps
         self.steps += 1
         return self.state, reward, done, {}
 
     def predict_done(self, state: np.ndarray) -> bool:
-        return abs(state[0]) >= 0.4
+        return False
 
     def seed(self, seed: int):
         self.action_space.seed(seed)
@@ -66,4 +65,4 @@ def seed(self, seed: int):
         self.init_space.seed(seed)
 
     def unsafe(self, state: np.ndarray) -> bool:
-        return abs(self.state[0]) >= 0.4
+        return abs(state[0]) >= 0.4
diff --git a/benchmarks/road.py b/benchmarks/road.py
index 7fb3b8dc..89536f5c 100644
--- a/benchmarks/road.py
+++ b/benchmarks/road.py
@@ -43,8 +43,7 @@ def step(self, action: np.ndarray) -> \
                              self.observation_space.high)
 
         reward = -abs(x - 3.0)
-        done = x >= 3.0 or self.steps >= self._max_episode_steps or \
-            self.unsafe(self.state)
+        done = x >= 3.0 or self.steps >= self._max_episode_steps
         self.steps += 1
 
         return self.state, reward, done, {}
diff --git a/benchmarks/road_2d.py b/benchmarks/road_2d.py
index 14924033..d1200132 100644
--- a/benchmarks/road_2d.py
+++ b/benchmarks/road_2d.py
@@ -83,8 +83,7 @@ def step(self, action: np.ndarray) -> \
 
         reward = -(abs(x - 3.0) + abs(y - 3.0))
         done = x >= 3.0 and y >= 3.0
-        done = done or self.steps >= self._max_episode_steps or \
-            self.unsafe(self.state)
+        done = done or self.steps >= self._max_episode_steps
         self.steps += 1
 
         return self.state, reward, done, {}
@@ -98,4 +97,4 @@ def seed(self, seed: int):
         self.init_space.seed(seed)
 
     def unsafe(self, state: np.ndarray) -> bool:
-        return state[0]**2 + state[1]**2 >= self.max_speed**2
+        return state[2]**2 + state[3]**2 >= self.max_speed**2