[pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

lerobot/common/policies/sac/modeling_sac.py

@@ -369,8 +369,17 @@ class SACPolicy(
         ).sum()
         return critics_loss
 
-    def compute_loss_grasp_critic(self, observations, actions, rewards, next_observations, done, observation_features=None, next_observation_features=None, complementary_info=None):
+    def compute_loss_grasp_critic(
-
+        self,
+        observations,
+        actions,
+        rewards,
+        next_observations,
+        done,
+        observation_features=None,
+        next_observation_features=None,
+        complementary_info=None,
+    ):
         batch_size = rewards.shape[0]
         grasp_actions = torch.clip(actions[:, -1].long() + 1, 0, 2)  # Map [-1, 0, 1] -> [0, 1, 2]
 
@@ -632,9 +641,7 @@ class GraspCritic(nn.Module):
         self.parameters_to_optimize += list(self.output_layer.parameters())
 
     def forward(
-        self,
+        self, observations: torch.Tensor, observation_features: torch.Tensor | None = None
-        observations: torch.Tensor,
-        observation_features: torch.Tensor | None = None
     ) -> torch.Tensor:
         device = get_device_from_parameters(self)
         # Move each tensor in observations to device

lerobot/scripts/server/buffer.py

@@ -291,9 +291,7 @@ class ReplayBuffer:
                     if isinstance(value, torch.Tensor):
                         shape = value.shape if value.ndim > 0 else (1,)
                         self.complementary_info_storage[key] = torch.zeros(
-                            (self.capacity, *shape), 
+                            (self.capacity, *shape), dtype=value.dtype, device=self.storage_device
-                            dtype=value.dtype, 
-                            device=self.storage_device
                         )
 
                 # Store the value
@@ -304,7 +302,7 @@ class ReplayBuffer:
                         # For non-tensor values (like grasp_penalty)
                         self.complementary_info_storage[key][self.position] = torch.tensor(
                             value, device=self.storage_device
-                    )
+                        )
 
         self.position = (self.position + 1) % self.capacity
         self.size = min(self.size + 1, self.capacity)
@@ -366,7 +364,7 @@ class ReplayBuffer:
 
         # Add complementary_info to batch if it exists
         batch_complementary_info = {}
-        if hasattr(self, 'complementary_info_keys') and self.complementary_info_keys:
+        if hasattr(self, "complementary_info_keys") and self.complementary_info_keys:
             for key in self.complementary_info_keys:
                 if key in self.complementary_info_storage:
                     batch_complementary_info[key] = self.complementary_info_storage[key][idx].to(self.device)

lerobot/scripts/server/end_effector_control_utils.py

@@ -334,7 +334,7 @@ class GamepadController(InputController):
                 return 0.0  # No change
 
         except pygame.error:
-            logging.error(f"Error reading gamepad. Is it still connected?")
+            logging.error("Error reading gamepad. Is it still connected?")
             return 0.0
 
 

lerobot/scripts/server/gym_manipulator.py

@@ -1077,18 +1077,20 @@ class GripperPenaltyWrapper(gym.Wrapper):
 
     def reset(self, **kwargs):
         obs, info = self.env.reset(**kwargs)
-        self.last_gripper_pos = obs["observation.state"][0, 0] # first idx for the gripper
+        self.last_gripper_pos = obs["observation.state"][0, 0]  # first idx for the gripper
         return obs, info
 
     def step(self, action):
         observation, reward, terminated, truncated, info = self.env.step(action)
 
-        if (action[-1] < -0.5 and self.last_gripper_pos > 0.9) or (action[-1] > 0.5 and self.last_gripper_pos < 0.9):
+        if (action[-1] < -0.5 and self.last_gripper_pos > 0.9) or (
+            action[-1] > 0.5 and self.last_gripper_pos < 0.9
+        ):
             info["grasp_penalty"] = self.penalty
         else:
             info["grasp_penalty"] = 0.0
 
-        self.last_gripper_pos = observation["observation.state"][0, 0] # first idx for the gripper
+        self.last_gripper_pos = observation["observation.state"][0, 0]  # first idx for the gripper
         return observation, reward, terminated, truncated, info
 
 
@@ -1167,7 +1169,7 @@ def make_robot_env(cfg) -> gym.vector.VectorEnv:
     if cfg.wrapper.ee_action_space_params is None and cfg.wrapper.joint_masking_action_space is not None:
         env = JointMaskingActionSpace(env=env, mask=cfg.wrapper.joint_masking_action_space)
     env = BatchCompitableWrapper(env=env)
-    env= GripperPenaltyWrapper(env=env)
+    env = GripperPenaltyWrapper(env=env)
 
     return env
 

lerobot/scripts/server/learner_server.py

@@ -762,7 +762,9 @@ def make_optimizers_and_scheduler(cfg, policy: nn.Module):
         lr=cfg.policy.actor_lr,
     )
     optimizer_critic = torch.optim.Adam(params=policy.critic_ensemble.parameters(), lr=cfg.policy.critic_lr)
-    optimizer_grasp_critic = torch.optim.Adam(params=policy.grasp_critic.parameters(), lr=policy.config.grasp_critic_lr)
+    optimizer_grasp_critic = torch.optim.Adam(
+        params=policy.grasp_critic.parameters(), lr=policy.config.grasp_critic_lr
+    )
     optimizer_temperature = torch.optim.Adam(params=[policy.log_alpha], lr=cfg.policy.critic_lr)
     lr_scheduler = None
     optimizers = {