Update README.md

README.md

@@ -12,13 +12,123 @@ configs:
 
 This dataset was created using [LeRobot](https://github.com/huggingface/lerobot).
 
-## Dataset Description
+# 1 Dataset Description
 
+This dataset includes 62 Lightwheel-Robocasa-Tasks, collected using the g1-controller robot in environments provided by LW-BenchHub.  
+The robot configuration used during data collection is **`G1-Controller`**, using `UnitreeG1ControllerEnvCfg` inherits from `UnitreeG1EnvCfg`.
 
+---
+
+## 1.1 Robot State
+
+The robot state recorded in the environment is stored in: **observation.state**
+- **Dimension**: 31  
+- **Description**: Joint positions of the robot  
+- **Joint mapping**: The exact joint ordering and semantics can be found in the **Dataset Structure** section below.
+
+---
+
+## 1.2 Robot Action
+
+The robot actions are recorded in: **action**
+- **Dimension**: 19  
+The action is ordered as: base (3), left arm (7), right arm (7), left hand (1), right hand (1).
+
+| Component       | Dimension | Description                         |
+|:----------------|:---------:|:------------------------------------|
+| Base action     | 3         | Planar base motion in world frame   |
+| Arm action      | 14        | 7-DOF left arm + 7-DOF right arm (end-effector poses) |
+| Hand action     | 2         | Continuous gripper control          |
+
+```python
+@configclass
+class G1PinkActionsCfg(PinkActionsCfg):
+    base_action: mdp.RelativeJointPositionActionCfg = mdp.RelativeJointPositionActionCfg(
+        asset_name="robot",
+        joint_names=["base_.*"],
+        scale={
+            "base_x_joint": 0.01,
+            "base_y_joint": 0.01,
+            "base_yaw_joint": 0.02,
+        },
+        use_zero_offset=True,
+    )
+
+    arms_action: G1ActionCfg = G1ActionCfg(asset_name="robot", joint_names=[
+        "left_shoulder_pitch_joint",
+        "left_shoulder_roll_joint",
+        "left_shoulder_yaw_joint",
+        "left_elbow_joint",
+        "left_wrist_roll_joint",
+        "left_wrist_pitch_joint",
+        "left_wrist_yaw_joint",
+        "right_shoulder_pitch_joint",
+        "right_shoulder_roll_joint",
+        "right_shoulder_yaw_joint",
+        "right_elbow_joint",
+        "right_wrist_roll_joint",
+        "right_wrist_pitch_joint",
+        "right_wrist_yaw_joint"
+    ])
+```
+
+## 1.2.1 Base Action
+
+The base action is implemented using `RelativeJointPositionActionCfg`, you can refer to the `G1PinkActionsCfg` above.
 
-- **Homepage:** [More Information Needed]
-- **Paper:** [More Information Needed]
-- **License:** apache-2.0
+- **Controlled dimensions**:
+  - `base_x_joint` (index 0): Forward/backward movement in world frame
+  - `base_y_joint` (index 1): Left/right movement in world frame
+  - `base_yaw_joint` (index 2): Yaw rotation
+- **Scale**:
+  - `base_x_joint`: 0.01
+  - `base_y_joint`: 0.01
+  - `base_yaw_joint`: 0.02
+
+## 1.2.2 Arm Action
+
+The arm actions control the end-effector poses (position and orientation) in the robot root frame.  
+The 14-dimensional arm action is passed to `G1ActionCfg`, where it is interpreted as a pose vector: [left_arm_pos (3), left_arm_quat (4), right_arm_pos (3), right_arm_quat (4)]  
+This pose representation is then converted into a 14-dimensional joint-space command via inverse kinematics.
+
+- **Left arm action** (indices 3-9):
+  - `left_arm_pos_x` (index 3): X position in robot root frame
+  - `left_arm_pos_y` (index 4): Y position in robot root frame
+  - `left_arm_pos_z` (index 5): Z position in robot root frame
+  - `left_arm_quat_w` (index 6): Quaternion w component
+  - `left_arm_quat_x` (index 7): Quaternion x component
+  - `left_arm_quat_y` (index 8): Quaternion y component
+  - `left_arm_quat_z` (index 9): Quaternion z component
+
+- **Right arm action** (indices 10-16):
+  - `right_arm_pos_x` (index 10): X position in robot root frame
+  - `right_arm_pos_y` (index 11): Y position in robot root frame
+  - `right_arm_pos_z` (index 12): Z position in robot root frame
+  - `right_arm_quat_w` (index 13): Quaternion w component
+  - `right_arm_quat_x` (index 14): Quaternion x component
+  - `right_arm_quat_y` (index 15): Quaternion y component
+  - `right_arm_quat_z` (index 16): Quaternion z component
+
+- **Coordinate frame**: Robot root frame
+
+
+## 1.2.3 Gripper Action
+
+Grippers are controlled using continuous values. In the action data, `-1.0` indicates fully open, `+1.0` indicates fully closed.
+
+- **Action values**:
+  - `-1.0`: fully open gripper
+  - `+1.0`: fully closed gripper
+  - Continuous values in between represent partial opening/closing
+
+- **Left gripper** (index 17): Continuous value from -1.0 (open) to 1.0 (closed)
+- **Right gripper** (index 18): Continuous value from -1.0 (open) to 1.0 (closed)
+
+- **Source**: VR trigger mapping
+  - `trigger=0.0` (not pressed) → `-1.0` (open)
+  - `trigger=1.0` (fully pressed) → `1.0` (closed)
+
+---
 
 ## Dataset Structure
 
@@ -257,4 +367,6 @@ This dataset was created using [LeRobot](https://github.com/huggingface/lerobot)
     "data_files_size_in_mb": 100,
     "video_files_size_in_mb": 500
 }
-```
+```
+## Homepage: [[Lightwheel-BenchHub](https://docs.lightwheel.net/lw_benchhub/)]
+## License: [[Apache License 2.0](https://www.apache.org/licenses/LICENSE-2.0)]