diff --git a/mani_skill/agents/controllers/__init__.py b/mani_skill/agents/controllers/__init__.py
index a0fdc446f..d2867b85e 100644
--- a/mani_skill/agents/controllers/__init__.py
+++ b/mani_skill/agents/controllers/__init__.py
@@ -15,6 +15,7 @@
 from .pd_joint_pos_vel import PDJointPosVelController, PDJointPosVelControllerConfig
 from .passive_controller import PassiveController, PassiveControllerConfig
 from .pd_base_vel import PDBaseVelController, PDBaseVelControllerConfig
+from .pd_base_vel import PDBaseForwardVelController, PDBaseForwardVelControllerConfig
 
 
 def deepcopy_dict(configs: dict):
diff --git a/mani_skill/agents/controllers/pd_base_vel.py b/mani_skill/agents/controllers/pd_base_vel.py
index 264d89cc9..7374c9004 100644
--- a/mani_skill/agents/controllers/pd_base_vel.py
+++ b/mani_skill/agents/controllers/pd_base_vel.py
@@ -1,7 +1,7 @@
 import numpy as np
 import torch
 
-from mani_skill.utils.geometry import rotate_2d_vec_by_angle
+from gymnasium import spaces
 from mani_skill.utils.structs.types import Array
 
 from .pd_joint_vel import PDJointVelController, PDJointVelControllerConfig
@@ -34,3 +34,41 @@ def set_action(self, action: Array):
 
 class PDBaseVelControllerConfig(PDJointVelControllerConfig):
     controller_cls = PDBaseVelController
+
+
+class PDBaseForwardVelController(PDJointVelController):
+    """PDJointVelController for forward-only ego-centric base movement."""
+
+    def _initialize_action_space(self):
+        assert len(self.joints) >= 3, len(self.joints)
+        low = np.float32(np.broadcast_to(self.config.lower, 2))
+        high = np.float32(np.broadcast_to(self.config.upper, 2))
+        self.single_action_space = spaces.Box(low, high, dtype=np.float32)
+
+    def set_action(self, action: Array):
+        action = self._preprocess_action(action)
+        # action[:, 0] should correspond to forward vel
+        # action[:, 1] should correspond to rotation vel
+
+        # Convert to ego-centric action
+        # Assume the 3rd DoF stands for orientation
+        ori = self.qpos[:, 2]
+        rot_mat = torch.zeros(ori.shape[0], 2, 2, device=action.device)
+        rot_mat[:, 0, 0] = torch.cos(ori)
+        rot_mat[:, 0, 1] = -torch.sin(ori)
+        rot_mat[:, 1, 0] = torch.sin(ori)
+        rot_mat[:, 1, 1] = torch.cos(ori)
+
+        # Assume the 1st DoF stands for forward movement
+        # make action with 0 y vel
+        move_action = action.clone()
+        move_action[:, 1] = 0
+        vel = (rot_mat @ move_action.float().unsqueeze(-1)).squeeze(-1)
+        new_action = torch.hstack([vel, action[:, 1:]])
+        self.articulation.set_joint_drive_velocity_targets(
+            new_action, self.joints, self.active_joint_indices
+        )
+
+
+class PDBaseForwardVelControllerConfig(PDJointVelControllerConfig):
+    controller_cls = PDBaseForwardVelController
diff --git a/mani_skill/agents/robots/fetch/fetch.py b/mani_skill/agents/robots/fetch/fetch.py
index dfd781d2a..efd6746d2 100644
--- a/mani_skill/agents/robots/fetch/fetch.py
+++ b/mani_skill/agents/robots/fetch/fetch.py
@@ -250,10 +250,10 @@ def _controller_configs(self):
         # -------------------------------------------------------------------------- #
         # Base
         # -------------------------------------------------------------------------- #
-        base_pd_joint_vel = PDBaseVelControllerConfig(
+        base_pd_joint_vel = PDBaseForwardVelControllerConfig(
             self.base_joint_names,
-            lower=[-0.5, -0.5, -3.14],
-            upper=[0.5, 0.5, 3.14],
+            lower=[-0.5, -3.14],
+            upper=[0.5, 3.14],
             damping=1000,
             force_limit=500,
         )