diff --git a/src/lerobot/envs/vlabench.py b/src/lerobot/envs/vlabench.py
index 548edf846..ea72b77cb 100644
--- a/src/lerobot/envs/vlabench.py
+++ b/src/lerobot/envs/vlabench.py
@@ -43,8 +43,8 @@ from .utils import _LazyAsyncVectorEnv
 logger = logging.getLogger(__name__)
 
 ACTION_DIM = 7  # pos(3) + euler(3) + gripper(1)
-ACTION_LOW = -1.0
-ACTION_HIGH = 1.0
+ACTION_LOW = np.array([-1.0, -1.0, -1.0, -1.0, -1.0, -1.0, 0.0], dtype=np.float32)
+ACTION_HIGH = np.array([1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], dtype=np.float32)
 
 # Default max episode steps per task type
 DEFAULT_MAX_EPISODE_STEPS = 500
@@ -177,9 +177,7 @@ class VLABenchEnv(gym.Env):
         else:
             raise ValueError(f"Unsupported obs_type: {self.obs_type}")
 
-        self.action_space = spaces.Box(
-            low=ACTION_LOW, high=ACTION_HIGH, shape=(ACTION_DIM,), dtype=np.float32
-        )
+        self.action_space = spaces.Box(low=ACTION_LOW, high=ACTION_HIGH, dtype=np.float32)
 
     # Max attempts to rebuild the underlying env when MuJoCo throws
     # `PhysicsError` (e.g. mjWARN_BADQACC) during VLABench's 20-step
@@ -346,6 +344,19 @@ class VLABenchEnv(gym.Env):
             dtype=np.float64,
         )
 
+    @staticmethod
+    def _normalize_gripper_action(gripper: float) -> float:
+        """Normalize the scalar gripper command to VLABench's [0, 1] convention.
+
+        The native VLABench collector uses 0=open and 1=closed. Older LeRobot
+        integrations often assumed a symmetric [-1, 1] gripper channel, so we
+        preserve backward compatibility by remapping negative values from
+        [-1, 1] -> [0, 1] before clipping.
+        """
+        if gripper < 0.0:
+            gripper = 0.5 * (float(np.clip(gripper, -1.0, 1.0)) + 1.0)
+        return float(np.clip(gripper, 0.0, 1.0))
+
     def _build_ctrl_from_action(self, action: np.ndarray, ctrl_dim: int) -> np.ndarray:
         """Convert a 7D EEF action into the `ctrl_dim`-sized joint command vector.
 
@@ -367,7 +378,7 @@ class VLABenchEnv(gym.Env):
 
         pos = np.asarray(action[:3], dtype=np.float64)
         rx, ry, rz = float(action[3]), float(action[4]), float(action[5])
-        gripper = float(action[6])
+        gripper = self._normalize_gripper_action(float(action[6]))
         quat = self._euler_xyz_to_quat_wxyz(rx, ry, rz)
 
         assert self._env is not None
@@ -390,7 +401,7 @@ class VLABenchEnv(gym.Env):
 
         # Gripper: action scalar in [0, 1] (0=open, 1=closed). Map linearly to
         # finger qpos in [CLOSED, OPEN]. Franka has 2 mirrored fingers.
-        g = float(np.clip(gripper, 0.0, 1.0))
+        g = gripper
         finger_qpos = self._FRANKA_FINGER_OPEN + g * (self._FRANKA_FINGER_CLOSED - self._FRANKA_FINGER_OPEN)
 
         ctrl = np.zeros(ctrl_dim, dtype=np.float64)
diff --git a/tests/envs/test_vlabench_env.py b/tests/envs/test_vlabench_env.py
new file mode 100644
index 000000000..97456cc57
--- /dev/null
+++ b/tests/envs/test_vlabench_env.py
@@ -0,0 +1,40 @@
+#!/usr/bin/env python
+
+# Copyright 2025 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import numpy as np
+
+from lerobot.envs.vlabench import ACTION_HIGH, ACTION_LOW, VLABenchEnv
+
+
+def test_vlabench_action_space_uses_zero_to_one_gripper_channel():
+    env = VLABenchEnv()
+
+    np.testing.assert_array_equal(env.action_space.low, ACTION_LOW)
+    np.testing.assert_array_equal(env.action_space.high, ACTION_HIGH)
+    assert env.action_space.shape == (7,)
+    assert env.action_space.low[-1] == 0.0
+    assert env.action_space.high[-1] == 1.0
+    np.testing.assert_array_equal(env.action_space.low[:6], np.full(6, -1.0, dtype=np.float32))
+    np.testing.assert_array_equal(env.action_space.high[:6], np.full(6, 1.0, dtype=np.float32))
+
+
+def test_vlabench_gripper_action_normalization_keeps_backward_compatibility():
+    assert VLABenchEnv._normalize_gripper_action(-1.0) == 0.0
+    assert VLABenchEnv._normalize_gripper_action(-0.5) == 0.25
+    assert VLABenchEnv._normalize_gripper_action(0.0) == 0.0
+    assert VLABenchEnv._normalize_gripper_action(0.5) == 0.5
+    assert VLABenchEnv._normalize_gripper_action(1.0) == 1.0
+    assert VLABenchEnv._normalize_gripper_action(1.5) == 1.0