Add CUHK Endoscopy README files

Upload the shared CUHK OpenH_Dataset_full README.md to each Endoscopy CUHK dataset root.

Endoscopy/cuhk/openh_dataset_full/find_greater_curvature/README.md

@@ -0,0 +1,173 @@
+
+# Stomach Phantom Navigation (2-Motor) - README
+
+---
+
+## 📋 At a Glance
+
+Teleoperated demonstrations of a custom 2-motor soft robotic endoscope performing navigation and scanning tasks inside a stomach phantom. The dataset captures the coupling between cable-driven motor control and visual feedback under varying illumination conditions.
+
+---
+
+## 📖 Dataset Overview
+
+This dataset contains **2158** trajectories of human operators using a gamepad controller to navigate a soft continuum robot inside a stomach phantom. The task involves maneuvering the endoscope tip to locate, approach, and center specific anatomical targets (e.g., lower body, angularis) while the system records synchronized visual, kinematic, and electromagnetic tracker data.
+
+It is designed to train **2-DoF** continuous motor control policies with multimodal state feedback.
+
+| | |
+| :--- | :--- |
+| **Total Trajectories** | 2158 |
+| **Total Hours** | 8.07 hs |
+| **Data Type** | [x] Table-Top Phantom |
+| **License** | CC BY 4.0 |
+| **Version** | 2.1 |
+
+---
+
+## 🎯 Tasks & Domain
+
+### Domain
+
+- [x] **Surgical Robotics**
+- [ ] **Ultrasound Robotics**
+- [ ] **Other Healthcare Robotics**
+
+### Demonstrated Skills
+
+- **Soft Robot Navigation**: Controlling a continuum manipulator with non-linear dynamics.
+- **Visual Servoing**: Keeping anatomical targets centered in the camera view.
+- **3D Spatial Awareness**: Navigating a cavity using electromagnetic tracker feedback.
+
+### Tasks
+
+Data covers the following scanning tasks:
+1.  **Greater Curvature**: "Scan the greater curvature of stomach to find white oval suspicious region." with 462 episodes.
+2.  **Less Curvature**: "Scan the lesser curvature of stomach to find orange suspicious region." with 200 episodes
+3.  **Pyloric Antrum**: "Scan the pyloric antrum to find small round nodule." with 204 episodes
+4.  **Tracking in 1**: "Tracking white oval suspicious region and centering it." with 414 episodes
+5.  **Tracking in 2**: "Tracking orange suspicious region and centering it." with 156 episodes
+6.  **Tracking in 3**: "Tracking small round nodule and centering it." with 722 episodes
+
+---
+
+## 🔬 Data Collection Details
+
+### Collection Method
+
+- [x] **Human Teleoperation** 
+- [ ] **Programmatic/State-Machine**
+- [ ] **AI Policy / Autonomous**
+- [ ] **Other**
+
+### Operator Details
+
+| | Description |
+| :--- | :--- |
+| **Operator Count** | 2 |
+| **Operator Skill Level** | [x] Endoscopic Surgeon Expert for the tasks of find and track <br> [x] Intermediate (Trained Researcher) for the task of tracking suspicious region |
+| **Collection Period** | From 2025-10-01 to 2026-02-12 |
+
+### Recovery Demonstrations
+
+- [ ] **Yes**
+- [x] **No**
+
+---
+
+## Diversity Dimensions
+
+- [x] **Spatial Layout**: Robot initializes in different relative positions to the target.
+- [x] **Lighting Conditions**: Recorded state includes variable light intensity levels.
+- [x] **Target Object**: Multiple anatomical regions within the phantom.
+- [x] **Task Execution**: Different approach angles and scanning speeds.
+
+---
+
+## 🛠️ Equipment & Setup
+
+### Robotic Platform(s)
+
+- **Robot 1:** Custom Soft Robotic Endoscope (Cable-driven active bending section).
+- Gao, H., Yang, X., Xiao, X., Zhu, X., Zhang, T., Hou, C., ... & Ren, H. (2024). Transendoscopic flexible parallel continuum robotic mechanism for bimanual endoscopic submucosal dissection. The International Journal of Robotics Research, 43(3), 281-304.
+
+
+### Sensors & Cameras
+
+| Type | Model/Details |
+| :--- | :--- |
+| **Primary Camera** | Endoscopic Camera, 640x480 (RGB) @ 20fps |
+| **Tracker** |  (EM Tracker) | 6-DoF Pose (Position + Orientation) |
+| **Actuators** | DC Motors with Encoders | Absolute position feedback |
+
+
+---
+
+## 🎯 Action & State Space Representation
+
+### Action Space Representation
+
+**Primary Action Representation:**
+- [ ] **Absolute Cartesian**
+- [ ] **Relative Cartesian**
+- [ ] **Joint Space**
+- [x] **Other** (Motor Velocity Command)
+
+**Orientation Representation:**
+- [x] **Quaternions**
+- [ ] **Euler Angles**
+- [ ] **Other**  (Please specify: `Actions are scalar motor velocities`)
+
+**Reference Frame:**
+- [x] **Robot Base Frame** 
+- [ ] **Tool/End-Effector Frame**
+- [ ] **World/Global Frame**
+- [ ] **Camera Frame**
+
+**Action Dimensions:**
+The action space consists of **2 continuous dimensions** controlling the antagonistic cable-driven motors.
+
+```text
+action: [m1_spd, m2_spd]
+- m1_spd: Speed command for Motor 1 (controlling Left/Right bending)
+- m2_spd: Speed command for Motor 2 (controlling Up/Down bending)
+```
+
+### State Space Representation
+
+**State Information Included:**
+- [x] **Joint Positions** (Motor Encoders)
+- [ ] **Joint Velocities**
+- [x] **End-Effector Pose** (Cartesian position/orientation from EM Tracker)
+- [ ] **Force/Torque Readings**
+- [ ] **Gripper State**
+
+
+**State Dimensions:**
+The state observation is a **13-dimensional** vector fusing electromagnetic tracker data, motor feedback, and system parameters.
+
+```text
+observation.state: [rel_x, rel_y, rel_z, abs_x, abs_y, abs_z, qw, qx, qy, qz, m1_pos, m2_pos, light_val]
+- rel_x, rel_y, rel_z: Relative Cartesian position of the endoscope tip (relative to start/target)
+- abs_x, abs_y, abs_z: Absolute Cartesian position from NDI Trakstar (Global Frame)
+- qw, qx, qy, qz: End-effector orientation (Quaternion)
+- m1_pos, m2_pos: Absolute motor encoder positions (Cable length proxy)
+- light_val: Current LED illumination intensity (PWM duty cycle, 0-100)
+```
+---
+
+## ⏱️ Data Synchronization Approach
+
+We utilized a ROS Noetic backend to synchronize high-frequency sensor data with video frames, using the system clock (ROS Time) as the ground truth time source. Endoscopic camera frames are captured at 20 Hz using OpenCV, with each frame's timestamp strictly corresponding to the moment of capture. For sensor fusion, we integrate high-frequency (~100Hz+) pose data from an EM Tracker (NDI Trakstar) and asynchronous motor/light states polled from the microcontroller. Our alignment strategy employs a custom DataRecorder class that buffers incoming sensor messages; at each video frame trigger (20Hz), we perform time-based interpolation to align sensor data to the exact frame timestamp. Specifically, position is aligned using linear interpolation between the two nearest tracker packets, orientation is handled via Spherical Linear Interpolation (SLERP) for quaternions, and motor/light states use zero-order hold (nearest neighbor) as they change at control frequency. The resulting timestamp field in the dataset reflects this synchronized capture time, ensuring precise temporal alignment between pixel changes and physical state. 
+
+
+## 👥 Attribution & Contact
+
+*Please provide attribution for the dataset creators and a point of contact.*
+
+| | |
+| :--- | :--- |
+| **Dataset Lead** | `[Chi Kit Ng, Yidong Zhang, Yang Tao, Zhiqing Tang, Tianchun Wu, Chim Ho Yin, Lihang, Hongliang Ren]` |
+| **Institution** | `The Chinese University of Hong Kong` |
+| **Contact Email** | `[chikitng@link.cuhk.edu.hk]` |
+| **Citation (BibTeX)** | <pre><code>@misc{[Ex-Endovla VLA Dataset],<br>  author = {[Chi Kit Ng, Yidong Zhang, Yang Tao, Zhiqing Tang, Tianchun Wu, Chim Ho Yin, Hongliang Ren]},<br>  title = {[Endoscopic Find and Track VLA Dataset]},<br>  year = {2025},<br>  publisher = {Open-H-Embodiment},<br>  note = {https://hrpp.research.virginia.edu/teams/irb-sbs/researcher-guide-irb-sbs/identifiers}<br>}</code></pre> |

Endoscopy/cuhk/openh_dataset_full/find_lesser_curvature/README.md

@@ -0,0 +1,173 @@
+
+# Stomach Phantom Navigation (2-Motor) - README
+
+---
+
+## 📋 At a Glance
+
+Teleoperated demonstrations of a custom 2-motor soft robotic endoscope performing navigation and scanning tasks inside a stomach phantom. The dataset captures the coupling between cable-driven motor control and visual feedback under varying illumination conditions.
+
+---
+
+## 📖 Dataset Overview
+
+This dataset contains **2158** trajectories of human operators using a gamepad controller to navigate a soft continuum robot inside a stomach phantom. The task involves maneuvering the endoscope tip to locate, approach, and center specific anatomical targets (e.g., lower body, angularis) while the system records synchronized visual, kinematic, and electromagnetic tracker data.
+
+It is designed to train **2-DoF** continuous motor control policies with multimodal state feedback.
+
+| | |
+| :--- | :--- |
+| **Total Trajectories** | 2158 |
+| **Total Hours** | 8.07 hs |
+| **Data Type** | [x] Table-Top Phantom |
+| **License** | CC BY 4.0 |
+| **Version** | 2.1 |
+
+---
+
+## 🎯 Tasks & Domain
+
+### Domain
+
+- [x] **Surgical Robotics**
+- [ ] **Ultrasound Robotics**
+- [ ] **Other Healthcare Robotics**
+
+### Demonstrated Skills
+
+- **Soft Robot Navigation**: Controlling a continuum manipulator with non-linear dynamics.
+- **Visual Servoing**: Keeping anatomical targets centered in the camera view.
+- **3D Spatial Awareness**: Navigating a cavity using electromagnetic tracker feedback.
+
+### Tasks
+
+Data covers the following scanning tasks:
+1.  **Greater Curvature**: "Scan the greater curvature of stomach to find white oval suspicious region." with 462 episodes.
+2.  **Less Curvature**: "Scan the lesser curvature of stomach to find orange suspicious region." with 200 episodes
+3.  **Pyloric Antrum**: "Scan the pyloric antrum to find small round nodule." with 204 episodes
+4.  **Tracking in 1**: "Tracking white oval suspicious region and centering it." with 414 episodes
+5.  **Tracking in 2**: "Tracking orange suspicious region and centering it." with 156 episodes
+6.  **Tracking in 3**: "Tracking small round nodule and centering it." with 722 episodes
+
+---
+
+## 🔬 Data Collection Details
+
+### Collection Method
+
+- [x] **Human Teleoperation** 
+- [ ] **Programmatic/State-Machine**
+- [ ] **AI Policy / Autonomous**
+- [ ] **Other**
+
+### Operator Details
+
+| | Description |
+| :--- | :--- |
+| **Operator Count** | 2 |
+| **Operator Skill Level** | [x] Endoscopic Surgeon Expert for the tasks of find and track <br> [x] Intermediate (Trained Researcher) for the task of tracking suspicious region |
+| **Collection Period** | From 2025-10-01 to 2026-02-12 |
+
+### Recovery Demonstrations
+
+- [ ] **Yes**
+- [x] **No**
+
+---
+
+## Diversity Dimensions
+
+- [x] **Spatial Layout**: Robot initializes in different relative positions to the target.
+- [x] **Lighting Conditions**: Recorded state includes variable light intensity levels.
+- [x] **Target Object**: Multiple anatomical regions within the phantom.
+- [x] **Task Execution**: Different approach angles and scanning speeds.
+
+---
+
+## 🛠️ Equipment & Setup
+
+### Robotic Platform(s)
+
+- **Robot 1:** Custom Soft Robotic Endoscope (Cable-driven active bending section).
+- Gao, H., Yang, X., Xiao, X., Zhu, X., Zhang, T., Hou, C., ... & Ren, H. (2024). Transendoscopic flexible parallel continuum robotic mechanism for bimanual endoscopic submucosal dissection. The International Journal of Robotics Research, 43(3), 281-304.
+
+
+### Sensors & Cameras
+
+| Type | Model/Details |
+| :--- | :--- |
+| **Primary Camera** | Endoscopic Camera, 640x480 (RGB) @ 20fps |
+| **Tracker** |  (EM Tracker) | 6-DoF Pose (Position + Orientation) |
+| **Actuators** | DC Motors with Encoders | Absolute position feedback |
+
+
+---
+
+## 🎯 Action & State Space Representation
+
+### Action Space Representation
+
+**Primary Action Representation:**
+- [ ] **Absolute Cartesian**
+- [ ] **Relative Cartesian**
+- [ ] **Joint Space**
+- [x] **Other** (Motor Velocity Command)
+
+**Orientation Representation:**
+- [x] **Quaternions**
+- [ ] **Euler Angles**
+- [ ] **Other**  (Please specify: `Actions are scalar motor velocities`)
+
+**Reference Frame:**
+- [x] **Robot Base Frame** 
+- [ ] **Tool/End-Effector Frame**
+- [ ] **World/Global Frame**
+- [ ] **Camera Frame**
+
+**Action Dimensions:**
+The action space consists of **2 continuous dimensions** controlling the antagonistic cable-driven motors.
+
+```text
+action: [m1_spd, m2_spd]
+- m1_spd: Speed command for Motor 1 (controlling Left/Right bending)
+- m2_spd: Speed command for Motor 2 (controlling Up/Down bending)
+```
+
+### State Space Representation
+
+**State Information Included:**
+- [x] **Joint Positions** (Motor Encoders)
+- [ ] **Joint Velocities**
+- [x] **End-Effector Pose** (Cartesian position/orientation from EM Tracker)
+- [ ] **Force/Torque Readings**
+- [ ] **Gripper State**
+
+
+**State Dimensions:**
+The state observation is a **13-dimensional** vector fusing electromagnetic tracker data, motor feedback, and system parameters.
+
+```text
+observation.state: [rel_x, rel_y, rel_z, abs_x, abs_y, abs_z, qw, qx, qy, qz, m1_pos, m2_pos, light_val]
+- rel_x, rel_y, rel_z: Relative Cartesian position of the endoscope tip (relative to start/target)
+- abs_x, abs_y, abs_z: Absolute Cartesian position from NDI Trakstar (Global Frame)
+- qw, qx, qy, qz: End-effector orientation (Quaternion)
+- m1_pos, m2_pos: Absolute motor encoder positions (Cable length proxy)
+- light_val: Current LED illumination intensity (PWM duty cycle, 0-100)
+```
+---
+
+## ⏱️ Data Synchronization Approach
+
+We utilized a ROS Noetic backend to synchronize high-frequency sensor data with video frames, using the system clock (ROS Time) as the ground truth time source. Endoscopic camera frames are captured at 20 Hz using OpenCV, with each frame's timestamp strictly corresponding to the moment of capture. For sensor fusion, we integrate high-frequency (~100Hz+) pose data from an EM Tracker (NDI Trakstar) and asynchronous motor/light states polled from the microcontroller. Our alignment strategy employs a custom DataRecorder class that buffers incoming sensor messages; at each video frame trigger (20Hz), we perform time-based interpolation to align sensor data to the exact frame timestamp. Specifically, position is aligned using linear interpolation between the two nearest tracker packets, orientation is handled via Spherical Linear Interpolation (SLERP) for quaternions, and motor/light states use zero-order hold (nearest neighbor) as they change at control frequency. The resulting timestamp field in the dataset reflects this synchronized capture time, ensuring precise temporal alignment between pixel changes and physical state. 
+
+
+## 👥 Attribution & Contact
+
+*Please provide attribution for the dataset creators and a point of contact.*
+
+| | |
+| :--- | :--- |
+| **Dataset Lead** | `[Chi Kit Ng, Yidong Zhang, Yang Tao, Zhiqing Tang, Tianchun Wu, Chim Ho Yin, Lihang, Hongliang Ren]` |
+| **Institution** | `The Chinese University of Hong Kong` |
+| **Contact Email** | `[chikitng@link.cuhk.edu.hk]` |
+| **Citation (BibTeX)** | <pre><code>@misc{[Ex-Endovla VLA Dataset],<br>  author = {[Chi Kit Ng, Yidong Zhang, Yang Tao, Zhiqing Tang, Tianchun Wu, Chim Ho Yin, Hongliang Ren]},<br>  title = {[Endoscopic Find and Track VLA Dataset]},<br>  year = {2025},<br>  publisher = {Open-H-Embodiment},<br>  note = {https://hrpp.research.virginia.edu/teams/irb-sbs/researcher-guide-irb-sbs/identifiers}<br>}</code></pre> |

Endoscopy/cuhk/openh_dataset_full/find_pyloric_antrum/README.md

@@ -0,0 +1,173 @@
+
+# Stomach Phantom Navigation (2-Motor) - README
+
+---
+
+## 📋 At a Glance
+
+Teleoperated demonstrations of a custom 2-motor soft robotic endoscope performing navigation and scanning tasks inside a stomach phantom. The dataset captures the coupling between cable-driven motor control and visual feedback under varying illumination conditions.
+
+---
+
+## 📖 Dataset Overview
+
+This dataset contains **2158** trajectories of human operators using a gamepad controller to navigate a soft continuum robot inside a stomach phantom. The task involves maneuvering the endoscope tip to locate, approach, and center specific anatomical targets (e.g., lower body, angularis) while the system records synchronized visual, kinematic, and electromagnetic tracker data.
+
+It is designed to train **2-DoF** continuous motor control policies with multimodal state feedback.
+
+| | |
+| :--- | :--- |
+| **Total Trajectories** | 2158 |
+| **Total Hours** | 8.07 hs |
+| **Data Type** | [x] Table-Top Phantom |
+| **License** | CC BY 4.0 |
+| **Version** | 2.1 |
+
+---
+
+## 🎯 Tasks & Domain
+
+### Domain
+
+- [x] **Surgical Robotics**
+- [ ] **Ultrasound Robotics**
+- [ ] **Other Healthcare Robotics**
+
+### Demonstrated Skills
+
+- **Soft Robot Navigation**: Controlling a continuum manipulator with non-linear dynamics.
+- **Visual Servoing**: Keeping anatomical targets centered in the camera view.
+- **3D Spatial Awareness**: Navigating a cavity using electromagnetic tracker feedback.
+
+### Tasks
+
+Data covers the following scanning tasks:
+1.  **Greater Curvature**: "Scan the greater curvature of stomach to find white oval suspicious region." with 462 episodes.
+2.  **Less Curvature**: "Scan the lesser curvature of stomach to find orange suspicious region." with 200 episodes
+3.  **Pyloric Antrum**: "Scan the pyloric antrum to find small round nodule." with 204 episodes
+4.  **Tracking in 1**: "Tracking white oval suspicious region and centering it." with 414 episodes
+5.  **Tracking in 2**: "Tracking orange suspicious region and centering it." with 156 episodes
+6.  **Tracking in 3**: "Tracking small round nodule and centering it." with 722 episodes
+
+---
+
+## 🔬 Data Collection Details
+
+### Collection Method
+
+- [x] **Human Teleoperation** 
+- [ ] **Programmatic/State-Machine**
+- [ ] **AI Policy / Autonomous**
+- [ ] **Other**
+
+### Operator Details
+
+| | Description |
+| :--- | :--- |
+| **Operator Count** | 2 |
+| **Operator Skill Level** | [x] Endoscopic Surgeon Expert for the tasks of find and track <br> [x] Intermediate (Trained Researcher) for the task of tracking suspicious region |
+| **Collection Period** | From 2025-10-01 to 2026-02-12 |
+
+### Recovery Demonstrations
+
+- [ ] **Yes**
+- [x] **No**
+
+---
+
+## Diversity Dimensions
+
+- [x] **Spatial Layout**: Robot initializes in different relative positions to the target.
+- [x] **Lighting Conditions**: Recorded state includes variable light intensity levels.
+- [x] **Target Object**: Multiple anatomical regions within the phantom.
+- [x] **Task Execution**: Different approach angles and scanning speeds.
+
+---
+
+## 🛠️ Equipment & Setup
+
+### Robotic Platform(s)
+
+- **Robot 1:** Custom Soft Robotic Endoscope (Cable-driven active bending section).
+- Gao, H., Yang, X., Xiao, X., Zhu, X., Zhang, T., Hou, C., ... & Ren, H. (2024). Transendoscopic flexible parallel continuum robotic mechanism for bimanual endoscopic submucosal dissection. The International Journal of Robotics Research, 43(3), 281-304.
+
+
+### Sensors & Cameras
+
+| Type | Model/Details |
+| :--- | :--- |
+| **Primary Camera** | Endoscopic Camera, 640x480 (RGB) @ 20fps |
+| **Tracker** |  (EM Tracker) | 6-DoF Pose (Position + Orientation) |
+| **Actuators** | DC Motors with Encoders | Absolute position feedback |
+
+
+---
+
+## 🎯 Action & State Space Representation
+
+### Action Space Representation
+
+**Primary Action Representation:**
+- [ ] **Absolute Cartesian**
+- [ ] **Relative Cartesian**
+- [ ] **Joint Space**
+- [x] **Other** (Motor Velocity Command)
+
+**Orientation Representation:**
+- [x] **Quaternions**
+- [ ] **Euler Angles**
+- [ ] **Other**  (Please specify: `Actions are scalar motor velocities`)
+
+**Reference Frame:**
+- [x] **Robot Base Frame** 
+- [ ] **Tool/End-Effector Frame**
+- [ ] **World/Global Frame**
+- [ ] **Camera Frame**
+
+**Action Dimensions:**
+The action space consists of **2 continuous dimensions** controlling the antagonistic cable-driven motors.
+
+```text
+action: [m1_spd, m2_spd]
+- m1_spd: Speed command for Motor 1 (controlling Left/Right bending)
+- m2_spd: Speed command for Motor 2 (controlling Up/Down bending)
+```
+
+### State Space Representation
+
+**State Information Included:**
+- [x] **Joint Positions** (Motor Encoders)
+- [ ] **Joint Velocities**
+- [x] **End-Effector Pose** (Cartesian position/orientation from EM Tracker)
+- [ ] **Force/Torque Readings**
+- [ ] **Gripper State**
+
+
+**State Dimensions:**
+The state observation is a **13-dimensional** vector fusing electromagnetic tracker data, motor feedback, and system parameters.
+
+```text
+observation.state: [rel_x, rel_y, rel_z, abs_x, abs_y, abs_z, qw, qx, qy, qz, m1_pos, m2_pos, light_val]
+- rel_x, rel_y, rel_z: Relative Cartesian position of the endoscope tip (relative to start/target)
+- abs_x, abs_y, abs_z: Absolute Cartesian position from NDI Trakstar (Global Frame)
+- qw, qx, qy, qz: End-effector orientation (Quaternion)
+- m1_pos, m2_pos: Absolute motor encoder positions (Cable length proxy)
+- light_val: Current LED illumination intensity (PWM duty cycle, 0-100)
+```
+---
+
+## ⏱️ Data Synchronization Approach
+
+We utilized a ROS Noetic backend to synchronize high-frequency sensor data with video frames, using the system clock (ROS Time) as the ground truth time source. Endoscopic camera frames are captured at 20 Hz using OpenCV, with each frame's timestamp strictly corresponding to the moment of capture. For sensor fusion, we integrate high-frequency (~100Hz+) pose data from an EM Tracker (NDI Trakstar) and asynchronous motor/light states polled from the microcontroller. Our alignment strategy employs a custom DataRecorder class that buffers incoming sensor messages; at each video frame trigger (20Hz), we perform time-based interpolation to align sensor data to the exact frame timestamp. Specifically, position is aligned using linear interpolation between the two nearest tracker packets, orientation is handled via Spherical Linear Interpolation (SLERP) for quaternions, and motor/light states use zero-order hold (nearest neighbor) as they change at control frequency. The resulting timestamp field in the dataset reflects this synchronized capture time, ensuring precise temporal alignment between pixel changes and physical state. 
+
+
+## 👥 Attribution & Contact
+
+*Please provide attribution for the dataset creators and a point of contact.*
+
+| | |
+| :--- | :--- |
+| **Dataset Lead** | `[Chi Kit Ng, Yidong Zhang, Yang Tao, Zhiqing Tang, Tianchun Wu, Chim Ho Yin, Lihang, Hongliang Ren]` |
+| **Institution** | `The Chinese University of Hong Kong` |
+| **Contact Email** | `[chikitng@link.cuhk.edu.hk]` |
+| **Citation (BibTeX)** | <pre><code>@misc{[Ex-Endovla VLA Dataset],<br>  author = {[Chi Kit Ng, Yidong Zhang, Yang Tao, Zhiqing Tang, Tianchun Wu, Chim Ho Yin, Hongliang Ren]},<br>  title = {[Endoscopic Find and Track VLA Dataset]},<br>  year = {2025},<br>  publisher = {Open-H-Embodiment},<br>  note = {https://hrpp.research.virginia.edu/teams/irb-sbs/researcher-guide-irb-sbs/identifiers}<br>}</code></pre> |

Endoscopy/cuhk/openh_dataset_full/track_orange_lesion/README.md

@@ -0,0 +1,173 @@
+
+# Stomach Phantom Navigation (2-Motor) - README
+
+---
+
+## 📋 At a Glance
+
+Teleoperated demonstrations of a custom 2-motor soft robotic endoscope performing navigation and scanning tasks inside a stomach phantom. The dataset captures the coupling between cable-driven motor control and visual feedback under varying illumination conditions.
+
+---
+
+## 📖 Dataset Overview
+
+This dataset contains **2158** trajectories of human operators using a gamepad controller to navigate a soft continuum robot inside a stomach phantom. The task involves maneuvering the endoscope tip to locate, approach, and center specific anatomical targets (e.g., lower body, angularis) while the system records synchronized visual, kinematic, and electromagnetic tracker data.
+
+It is designed to train **2-DoF** continuous motor control policies with multimodal state feedback.
+
+| | |
+| :--- | :--- |
+| **Total Trajectories** | 2158 |
+| **Total Hours** | 8.07 hs |
+| **Data Type** | [x] Table-Top Phantom |
+| **License** | CC BY 4.0 |
+| **Version** | 2.1 |
+
+---
+
+## 🎯 Tasks & Domain
+
+### Domain
+
+- [x] **Surgical Robotics**
+- [ ] **Ultrasound Robotics**
+- [ ] **Other Healthcare Robotics**
+
+### Demonstrated Skills
+
+- **Soft Robot Navigation**: Controlling a continuum manipulator with non-linear dynamics.
+- **Visual Servoing**: Keeping anatomical targets centered in the camera view.
+- **3D Spatial Awareness**: Navigating a cavity using electromagnetic tracker feedback.
+
+### Tasks
+
+Data covers the following scanning tasks:
+1.  **Greater Curvature**: "Scan the greater curvature of stomach to find white oval suspicious region." with 462 episodes.
+2.  **Less Curvature**: "Scan the lesser curvature of stomach to find orange suspicious region." with 200 episodes
+3.  **Pyloric Antrum**: "Scan the pyloric antrum to find small round nodule." with 204 episodes
+4.  **Tracking in 1**: "Tracking white oval suspicious region and centering it." with 414 episodes
+5.  **Tracking in 2**: "Tracking orange suspicious region and centering it." with 156 episodes
+6.  **Tracking in 3**: "Tracking small round nodule and centering it." with 722 episodes
+
+---
+
+## 🔬 Data Collection Details
+
+### Collection Method
+
+- [x] **Human Teleoperation** 
+- [ ] **Programmatic/State-Machine**
+- [ ] **AI Policy / Autonomous**
+- [ ] **Other**
+
+### Operator Details
+
+| | Description |
+| :--- | :--- |
+| **Operator Count** | 2 |
+| **Operator Skill Level** | [x] Endoscopic Surgeon Expert for the tasks of find and track <br> [x] Intermediate (Trained Researcher) for the task of tracking suspicious region |
+| **Collection Period** | From 2025-10-01 to 2026-02-12 |
+
+### Recovery Demonstrations
+
+- [ ] **Yes**
+- [x] **No**
+
+---
+
+## Diversity Dimensions
+
+- [x] **Spatial Layout**: Robot initializes in different relative positions to the target.
+- [x] **Lighting Conditions**: Recorded state includes variable light intensity levels.
+- [x] **Target Object**: Multiple anatomical regions within the phantom.
+- [x] **Task Execution**: Different approach angles and scanning speeds.
+
+---
+
+## 🛠️ Equipment & Setup
+
+### Robotic Platform(s)
+
+- **Robot 1:** Custom Soft Robotic Endoscope (Cable-driven active bending section).
+- Gao, H., Yang, X., Xiao, X., Zhu, X., Zhang, T., Hou, C., ... & Ren, H. (2024). Transendoscopic flexible parallel continuum robotic mechanism for bimanual endoscopic submucosal dissection. The International Journal of Robotics Research, 43(3), 281-304.
+
+
+### Sensors & Cameras
+
+| Type | Model/Details |
+| :--- | :--- |
+| **Primary Camera** | Endoscopic Camera, 640x480 (RGB) @ 20fps |
+| **Tracker** |  (EM Tracker) | 6-DoF Pose (Position + Orientation) |
+| **Actuators** | DC Motors with Encoders | Absolute position feedback |
+
+
+---
+
+## 🎯 Action & State Space Representation
+
+### Action Space Representation
+
+**Primary Action Representation:**
+- [ ] **Absolute Cartesian**
+- [ ] **Relative Cartesian**
+- [ ] **Joint Space**
+- [x] **Other** (Motor Velocity Command)
+
+**Orientation Representation:**
+- [x] **Quaternions**
+- [ ] **Euler Angles**
+- [ ] **Other**  (Please specify: `Actions are scalar motor velocities`)
+
+**Reference Frame:**
+- [x] **Robot Base Frame** 
+- [ ] **Tool/End-Effector Frame**
+- [ ] **World/Global Frame**
+- [ ] **Camera Frame**
+
+**Action Dimensions:**
+The action space consists of **2 continuous dimensions** controlling the antagonistic cable-driven motors.
+
+```text
+action: [m1_spd, m2_spd]
+- m1_spd: Speed command for Motor 1 (controlling Left/Right bending)
+- m2_spd: Speed command for Motor 2 (controlling Up/Down bending)
+```
+
+### State Space Representation
+
+**State Information Included:**
+- [x] **Joint Positions** (Motor Encoders)
+- [ ] **Joint Velocities**
+- [x] **End-Effector Pose** (Cartesian position/orientation from EM Tracker)
+- [ ] **Force/Torque Readings**
+- [ ] **Gripper State**
+
+
+**State Dimensions:**
+The state observation is a **13-dimensional** vector fusing electromagnetic tracker data, motor feedback, and system parameters.
+
+```text
+observation.state: [rel_x, rel_y, rel_z, abs_x, abs_y, abs_z, qw, qx, qy, qz, m1_pos, m2_pos, light_val]
+- rel_x, rel_y, rel_z: Relative Cartesian position of the endoscope tip (relative to start/target)
+- abs_x, abs_y, abs_z: Absolute Cartesian position from NDI Trakstar (Global Frame)
+- qw, qx, qy, qz: End-effector orientation (Quaternion)
+- m1_pos, m2_pos: Absolute motor encoder positions (Cable length proxy)
+- light_val: Current LED illumination intensity (PWM duty cycle, 0-100)
+```
+---
+
+## ⏱️ Data Synchronization Approach
+
+We utilized a ROS Noetic backend to synchronize high-frequency sensor data with video frames, using the system clock (ROS Time) as the ground truth time source. Endoscopic camera frames are captured at 20 Hz using OpenCV, with each frame's timestamp strictly corresponding to the moment of capture. For sensor fusion, we integrate high-frequency (~100Hz+) pose data from an EM Tracker (NDI Trakstar) and asynchronous motor/light states polled from the microcontroller. Our alignment strategy employs a custom DataRecorder class that buffers incoming sensor messages; at each video frame trigger (20Hz), we perform time-based interpolation to align sensor data to the exact frame timestamp. Specifically, position is aligned using linear interpolation between the two nearest tracker packets, orientation is handled via Spherical Linear Interpolation (SLERP) for quaternions, and motor/light states use zero-order hold (nearest neighbor) as they change at control frequency. The resulting timestamp field in the dataset reflects this synchronized capture time, ensuring precise temporal alignment between pixel changes and physical state. 
+
+
+## 👥 Attribution & Contact
+
+*Please provide attribution for the dataset creators and a point of contact.*
+
+| | |
+| :--- | :--- |
+| **Dataset Lead** | `[Chi Kit Ng, Yidong Zhang, Yang Tao, Zhiqing Tang, Tianchun Wu, Chim Ho Yin, Lihang, Hongliang Ren]` |
+| **Institution** | `The Chinese University of Hong Kong` |
+| **Contact Email** | `[chikitng@link.cuhk.edu.hk]` |
+| **Citation (BibTeX)** | <pre><code>@misc{[Ex-Endovla VLA Dataset],<br>  author = {[Chi Kit Ng, Yidong Zhang, Yang Tao, Zhiqing Tang, Tianchun Wu, Chim Ho Yin, Hongliang Ren]},<br>  title = {[Endoscopic Find and Track VLA Dataset]},<br>  year = {2025},<br>  publisher = {Open-H-Embodiment},<br>  note = {https://hrpp.research.virginia.edu/teams/irb-sbs/researcher-guide-irb-sbs/identifiers}<br>}</code></pre> |

Endoscopy/cuhk/openh_dataset_full/track_small_round_nodule/README.md

@@ -0,0 +1,173 @@
+
+# Stomach Phantom Navigation (2-Motor) - README
+
+---
+
+## 📋 At a Glance
+
+Teleoperated demonstrations of a custom 2-motor soft robotic endoscope performing navigation and scanning tasks inside a stomach phantom. The dataset captures the coupling between cable-driven motor control and visual feedback under varying illumination conditions.
+
+---
+
+## 📖 Dataset Overview
+
+This dataset contains **2158** trajectories of human operators using a gamepad controller to navigate a soft continuum robot inside a stomach phantom. The task involves maneuvering the endoscope tip to locate, approach, and center specific anatomical targets (e.g., lower body, angularis) while the system records synchronized visual, kinematic, and electromagnetic tracker data.
+
+It is designed to train **2-DoF** continuous motor control policies with multimodal state feedback.
+
+| | |
+| :--- | :--- |
+| **Total Trajectories** | 2158 |
+| **Total Hours** | 8.07 hs |
+| **Data Type** | [x] Table-Top Phantom |
+| **License** | CC BY 4.0 |
+| **Version** | 2.1 |
+
+---
+
+## 🎯 Tasks & Domain
+
+### Domain
+
+- [x] **Surgical Robotics**
+- [ ] **Ultrasound Robotics**
+- [ ] **Other Healthcare Robotics**
+
+### Demonstrated Skills
+
+- **Soft Robot Navigation**: Controlling a continuum manipulator with non-linear dynamics.
+- **Visual Servoing**: Keeping anatomical targets centered in the camera view.
+- **3D Spatial Awareness**: Navigating a cavity using electromagnetic tracker feedback.
+
+### Tasks
+
+Data covers the following scanning tasks:
+1.  **Greater Curvature**: "Scan the greater curvature of stomach to find white oval suspicious region." with 462 episodes.
+2.  **Less Curvature**: "Scan the lesser curvature of stomach to find orange suspicious region." with 200 episodes
+3.  **Pyloric Antrum**: "Scan the pyloric antrum to find small round nodule." with 204 episodes
+4.  **Tracking in 1**: "Tracking white oval suspicious region and centering it." with 414 episodes
+5.  **Tracking in 2**: "Tracking orange suspicious region and centering it." with 156 episodes
+6.  **Tracking in 3**: "Tracking small round nodule and centering it." with 722 episodes
+
+---
+
+## 🔬 Data Collection Details
+
+### Collection Method
+
+- [x] **Human Teleoperation** 
+- [ ] **Programmatic/State-Machine**
+- [ ] **AI Policy / Autonomous**
+- [ ] **Other**
+
+### Operator Details
+
+| | Description |
+| :--- | :--- |
+| **Operator Count** | 2 |
+| **Operator Skill Level** | [x] Endoscopic Surgeon Expert for the tasks of find and track <br> [x] Intermediate (Trained Researcher) for the task of tracking suspicious region |
+| **Collection Period** | From 2025-10-01 to 2026-02-12 |
+
+### Recovery Demonstrations
+
+- [ ] **Yes**
+- [x] **No**
+
+---
+
+## Diversity Dimensions
+
+- [x] **Spatial Layout**: Robot initializes in different relative positions to the target.
+- [x] **Lighting Conditions**: Recorded state includes variable light intensity levels.
+- [x] **Target Object**: Multiple anatomical regions within the phantom.
+- [x] **Task Execution**: Different approach angles and scanning speeds.
+
+---
+
+## 🛠️ Equipment & Setup
+
+### Robotic Platform(s)
+
+- **Robot 1:** Custom Soft Robotic Endoscope (Cable-driven active bending section).
+- Gao, H., Yang, X., Xiao, X., Zhu, X., Zhang, T., Hou, C., ... & Ren, H. (2024). Transendoscopic flexible parallel continuum robotic mechanism for bimanual endoscopic submucosal dissection. The International Journal of Robotics Research, 43(3), 281-304.
+
+
+### Sensors & Cameras
+
+| Type | Model/Details |
+| :--- | :--- |
+| **Primary Camera** | Endoscopic Camera, 640x480 (RGB) @ 20fps |
+| **Tracker** |  (EM Tracker) | 6-DoF Pose (Position + Orientation) |
+| **Actuators** | DC Motors with Encoders | Absolute position feedback |
+
+
+---
+
+## 🎯 Action & State Space Representation
+
+### Action Space Representation
+
+**Primary Action Representation:**
+- [ ] **Absolute Cartesian**
+- [ ] **Relative Cartesian**
+- [ ] **Joint Space**
+- [x] **Other** (Motor Velocity Command)
+
+**Orientation Representation:**
+- [x] **Quaternions**
+- [ ] **Euler Angles**
+- [ ] **Other**  (Please specify: `Actions are scalar motor velocities`)
+
+**Reference Frame:**
+- [x] **Robot Base Frame** 
+- [ ] **Tool/End-Effector Frame**
+- [ ] **World/Global Frame**
+- [ ] **Camera Frame**
+
+**Action Dimensions:**
+The action space consists of **2 continuous dimensions** controlling the antagonistic cable-driven motors.
+
+```text
+action: [m1_spd, m2_spd]
+- m1_spd: Speed command for Motor 1 (controlling Left/Right bending)
+- m2_spd: Speed command for Motor 2 (controlling Up/Down bending)
+```
+
+### State Space Representation
+
+**State Information Included:**
+- [x] **Joint Positions** (Motor Encoders)
+- [ ] **Joint Velocities**
+- [x] **End-Effector Pose** (Cartesian position/orientation from EM Tracker)
+- [ ] **Force/Torque Readings**
+- [ ] **Gripper State**
+
+
+**State Dimensions:**
+The state observation is a **13-dimensional** vector fusing electromagnetic tracker data, motor feedback, and system parameters.
+
+```text
+observation.state: [rel_x, rel_y, rel_z, abs_x, abs_y, abs_z, qw, qx, qy, qz, m1_pos, m2_pos, light_val]
+- rel_x, rel_y, rel_z: Relative Cartesian position of the endoscope tip (relative to start/target)
+- abs_x, abs_y, abs_z: Absolute Cartesian position from NDI Trakstar (Global Frame)
+- qw, qx, qy, qz: End-effector orientation (Quaternion)
+- m1_pos, m2_pos: Absolute motor encoder positions (Cable length proxy)
+- light_val: Current LED illumination intensity (PWM duty cycle, 0-100)
+```
+---
+
+## ⏱️ Data Synchronization Approach
+
+We utilized a ROS Noetic backend to synchronize high-frequency sensor data with video frames, using the system clock (ROS Time) as the ground truth time source. Endoscopic camera frames are captured at 20 Hz using OpenCV, with each frame's timestamp strictly corresponding to the moment of capture. For sensor fusion, we integrate high-frequency (~100Hz+) pose data from an EM Tracker (NDI Trakstar) and asynchronous motor/light states polled from the microcontroller. Our alignment strategy employs a custom DataRecorder class that buffers incoming sensor messages; at each video frame trigger (20Hz), we perform time-based interpolation to align sensor data to the exact frame timestamp. Specifically, position is aligned using linear interpolation between the two nearest tracker packets, orientation is handled via Spherical Linear Interpolation (SLERP) for quaternions, and motor/light states use zero-order hold (nearest neighbor) as they change at control frequency. The resulting timestamp field in the dataset reflects this synchronized capture time, ensuring precise temporal alignment between pixel changes and physical state. 
+
+
+## 👥 Attribution & Contact
+
+*Please provide attribution for the dataset creators and a point of contact.*
+
+| | |
+| :--- | :--- |
+| **Dataset Lead** | `[Chi Kit Ng, Yidong Zhang, Yang Tao, Zhiqing Tang, Tianchun Wu, Chim Ho Yin, Lihang, Hongliang Ren]` |
+| **Institution** | `The Chinese University of Hong Kong` |
+| **Contact Email** | `[chikitng@link.cuhk.edu.hk]` |
+| **Citation (BibTeX)** | <pre><code>@misc{[Ex-Endovla VLA Dataset],<br>  author = {[Chi Kit Ng, Yidong Zhang, Yang Tao, Zhiqing Tang, Tianchun Wu, Chim Ho Yin, Hongliang Ren]},<br>  title = {[Endoscopic Find and Track VLA Dataset]},<br>  year = {2025},<br>  publisher = {Open-H-Embodiment},<br>  note = {https://hrpp.research.virginia.edu/teams/irb-sbs/researcher-guide-irb-sbs/identifiers}<br>}</code></pre> |

Endoscopy/cuhk/openh_dataset_full/track_white_oval_roi/README.md

@@ -0,0 +1,173 @@
+
+# Stomach Phantom Navigation (2-Motor) - README
+
+---
+
+## 📋 At a Glance
+
+Teleoperated demonstrations of a custom 2-motor soft robotic endoscope performing navigation and scanning tasks inside a stomach phantom. The dataset captures the coupling between cable-driven motor control and visual feedback under varying illumination conditions.
+
+---
+
+## 📖 Dataset Overview
+
+This dataset contains **2158** trajectories of human operators using a gamepad controller to navigate a soft continuum robot inside a stomach phantom. The task involves maneuvering the endoscope tip to locate, approach, and center specific anatomical targets (e.g., lower body, angularis) while the system records synchronized visual, kinematic, and electromagnetic tracker data.
+
+It is designed to train **2-DoF** continuous motor control policies with multimodal state feedback.
+
+| | |
+| :--- | :--- |
+| **Total Trajectories** | 2158 |
+| **Total Hours** | 8.07 hs |
+| **Data Type** | [x] Table-Top Phantom |
+| **License** | CC BY 4.0 |
+| **Version** | 2.1 |
+
+---
+
+## 🎯 Tasks & Domain
+
+### Domain
+
+- [x] **Surgical Robotics**
+- [ ] **Ultrasound Robotics**
+- [ ] **Other Healthcare Robotics**
+
+### Demonstrated Skills
+
+- **Soft Robot Navigation**: Controlling a continuum manipulator with non-linear dynamics.
+- **Visual Servoing**: Keeping anatomical targets centered in the camera view.
+- **3D Spatial Awareness**: Navigating a cavity using electromagnetic tracker feedback.
+
+### Tasks
+
+Data covers the following scanning tasks:
+1.  **Greater Curvature**: "Scan the greater curvature of stomach to find white oval suspicious region." with 462 episodes.
+2.  **Less Curvature**: "Scan the lesser curvature of stomach to find orange suspicious region." with 200 episodes
+3.  **Pyloric Antrum**: "Scan the pyloric antrum to find small round nodule." with 204 episodes
+4.  **Tracking in 1**: "Tracking white oval suspicious region and centering it." with 414 episodes
+5.  **Tracking in 2**: "Tracking orange suspicious region and centering it." with 156 episodes
+6.  **Tracking in 3**: "Tracking small round nodule and centering it." with 722 episodes
+
+---
+
+## 🔬 Data Collection Details
+
+### Collection Method
+
+- [x] **Human Teleoperation** 
+- [ ] **Programmatic/State-Machine**
+- [ ] **AI Policy / Autonomous**
+- [ ] **Other**
+
+### Operator Details
+
+| | Description |
+| :--- | :--- |
+| **Operator Count** | 2 |
+| **Operator Skill Level** | [x] Endoscopic Surgeon Expert for the tasks of find and track <br> [x] Intermediate (Trained Researcher) for the task of tracking suspicious region |
+| **Collection Period** | From 2025-10-01 to 2026-02-12 |
+
+### Recovery Demonstrations
+
+- [ ] **Yes**
+- [x] **No**
+
+---
+
+## Diversity Dimensions
+
+- [x] **Spatial Layout**: Robot initializes in different relative positions to the target.
+- [x] **Lighting Conditions**: Recorded state includes variable light intensity levels.
+- [x] **Target Object**: Multiple anatomical regions within the phantom.
+- [x] **Task Execution**: Different approach angles and scanning speeds.
+
+---
+
+## 🛠️ Equipment & Setup
+
+### Robotic Platform(s)
+
+- **Robot 1:** Custom Soft Robotic Endoscope (Cable-driven active bending section).
+- Gao, H., Yang, X., Xiao, X., Zhu, X., Zhang, T., Hou, C., ... & Ren, H. (2024). Transendoscopic flexible parallel continuum robotic mechanism for bimanual endoscopic submucosal dissection. The International Journal of Robotics Research, 43(3), 281-304.
+
+
+### Sensors & Cameras
+
+| Type | Model/Details |
+| :--- | :--- |
+| **Primary Camera** | Endoscopic Camera, 640x480 (RGB) @ 20fps |
+| **Tracker** |  (EM Tracker) | 6-DoF Pose (Position + Orientation) |
+| **Actuators** | DC Motors with Encoders | Absolute position feedback |
+
+
+---
+
+## 🎯 Action & State Space Representation
+
+### Action Space Representation
+
+**Primary Action Representation:**
+- [ ] **Absolute Cartesian**
+- [ ] **Relative Cartesian**
+- [ ] **Joint Space**
+- [x] **Other** (Motor Velocity Command)
+
+**Orientation Representation:**
+- [x] **Quaternions**
+- [ ] **Euler Angles**
+- [ ] **Other**  (Please specify: `Actions are scalar motor velocities`)
+
+**Reference Frame:**
+- [x] **Robot Base Frame** 
+- [ ] **Tool/End-Effector Frame**
+- [ ] **World/Global Frame**
+- [ ] **Camera Frame**
+
+**Action Dimensions:**
+The action space consists of **2 continuous dimensions** controlling the antagonistic cable-driven motors.
+
+```text
+action: [m1_spd, m2_spd]
+- m1_spd: Speed command for Motor 1 (controlling Left/Right bending)
+- m2_spd: Speed command for Motor 2 (controlling Up/Down bending)
+```
+
+### State Space Representation
+
+**State Information Included:**
+- [x] **Joint Positions** (Motor Encoders)
+- [ ] **Joint Velocities**
+- [x] **End-Effector Pose** (Cartesian position/orientation from EM Tracker)
+- [ ] **Force/Torque Readings**
+- [ ] **Gripper State**
+
+
+**State Dimensions:**
+The state observation is a **13-dimensional** vector fusing electromagnetic tracker data, motor feedback, and system parameters.
+
+```text
+observation.state: [rel_x, rel_y, rel_z, abs_x, abs_y, abs_z, qw, qx, qy, qz, m1_pos, m2_pos, light_val]
+- rel_x, rel_y, rel_z: Relative Cartesian position of the endoscope tip (relative to start/target)
+- abs_x, abs_y, abs_z: Absolute Cartesian position from NDI Trakstar (Global Frame)
+- qw, qx, qy, qz: End-effector orientation (Quaternion)
+- m1_pos, m2_pos: Absolute motor encoder positions (Cable length proxy)
+- light_val: Current LED illumination intensity (PWM duty cycle, 0-100)
+```
+---
+
+## ⏱️ Data Synchronization Approach
+
+We utilized a ROS Noetic backend to synchronize high-frequency sensor data with video frames, using the system clock (ROS Time) as the ground truth time source. Endoscopic camera frames are captured at 20 Hz using OpenCV, with each frame's timestamp strictly corresponding to the moment of capture. For sensor fusion, we integrate high-frequency (~100Hz+) pose data from an EM Tracker (NDI Trakstar) and asynchronous motor/light states polled from the microcontroller. Our alignment strategy employs a custom DataRecorder class that buffers incoming sensor messages; at each video frame trigger (20Hz), we perform time-based interpolation to align sensor data to the exact frame timestamp. Specifically, position is aligned using linear interpolation between the two nearest tracker packets, orientation is handled via Spherical Linear Interpolation (SLERP) for quaternions, and motor/light states use zero-order hold (nearest neighbor) as they change at control frequency. The resulting timestamp field in the dataset reflects this synchronized capture time, ensuring precise temporal alignment between pixel changes and physical state. 
+
+
+## 👥 Attribution & Contact
+
+*Please provide attribution for the dataset creators and a point of contact.*
+
+| | |
+| :--- | :--- |
+| **Dataset Lead** | `[Chi Kit Ng, Yidong Zhang, Yang Tao, Zhiqing Tang, Tianchun Wu, Chim Ho Yin, Lihang, Hongliang Ren]` |
+| **Institution** | `The Chinese University of Hong Kong` |
+| **Contact Email** | `[chikitng@link.cuhk.edu.hk]` |
+| **Citation (BibTeX)** | <pre><code>@misc{[Ex-Endovla VLA Dataset],<br>  author = {[Chi Kit Ng, Yidong Zhang, Yang Tao, Zhiqing Tang, Tianchun Wu, Chim Ho Yin, Hongliang Ren]},<br>  title = {[Endoscopic Find and Track VLA Dataset]},<br>  year = {2025},<br>  publisher = {Open-H-Embodiment},<br>  note = {https://hrpp.research.virginia.edu/teams/irb-sbs/researcher-guide-irb-sbs/identifiers}<br>}</code></pre> |