Refresh canonical JHU SMARTS README files

Replaces canonical SMARTS leaf meta README placeholders with the updated SMARTS README and patches stale canonical parent documentation.

Surgical/jhu/lcsr/smarts/README.md

@@ -5,7 +5,7 @@ This is the canonical JHU SMARTS namespace in Open-H.
 - Canonical SMARTS root: `Surgical/jhu/lcsr/smarts/`
 - Legacy offline SMARTS leaves remain temporarily under `Surgical/jhu/lscr/smarts/offline_recorder_extracted/...` during the deprecation window.
 - Canonical offline mapping: `SurgSync-stitch-coldcut/P1..P3` maps to legacy `offline_data_part1..3`.
-- Canonical online mapping: `SurgSync-multitask/P1..P4` will be published from `online_data_part1..4` source archives.
+- Canonical online mapping: `SurgSync-multitask/P1..P4` maps to `online_data_part1..4` and is published on Hugging Face.
 
 This dataset includes two sub-datasets: **offline_recorder** and **online_recorder**.
 

Surgical/jhu/lcsr/smarts/SurgSync-multitask/P1/meta/README.md

@@ -1,2 +1,229 @@
-Placeholder
-Please refer to the README.md uploaded in the parent folder, thanks!
+# SurgSync-multitask P1
+
+Canonical SMARTS leaf metadata README.
+
+- Canonical path: `Surgical/jhu/lcsr/smarts/SurgSync-multitask/P1/`
+- Source archive mapping: `online_data_part1.zip`.
+- This leaf is one canonical part of the broader JHU SMARTS dataset.
+
+The broader SMARTS dataset includes two sub-datasets: **offline_recorder** and **online_recorder**.
+
+---
+
+## 📋 At a Glance
+
+*Provide a one-sentence summary of your dataset.*
+
+Teleoperated demonstrations of a da Vinci robot (dVRK-Si) performing multiple canonical tasks on ex-vivo tissue or table-top phantoms; ~50% data collected within the Intuitive abdominal dome model. All data collection uses a modern stereo chip-on-tip endoscope. Both endoscope camera calibration and robot hand-eye calibration will be provided.
+
+---
+
+## File Structure
+
+For the dataset, it should
+
+```text
+./offline_recorder or online_recorder
+├── calibration/
+│   ├── case-*...
+│   │   ├── camera calibration
+│   │   │   ├── left.yaml
+│   │   │   ├── right.yaml
+│   │   │   └── stereo_calib_params.json
+│   │   └── hand_eye_calibration
+│   │   │   ├── PSM1/2-registration-dVRK.json
+│   │   │   └── PSM1/2-registration-open-cv.json
+├── data/
+│   └── case-*...
+├── videos/
+│   └── case-*...
+├── meta/
+│   ├── episodes.jsonl
+│   ├── episodes_stats.jsonl
+│   ├── tasks.jsonl
+│   ├── info.json
+│   └── README.md
+└── total_time.json
+```
+
+---
+
+## 📖 Dataset Overview
+
+*Briefly describe the purpose and content of this dataset. What key skills or scenarios does it demonstrate?*
+
+This dataset comprises 2500+ teleoperated expert demonstrations of multiple canonical tasks on ex-vivo tissue or table-top phantoms using a surgical robotic system. The tasks include simple interrupted stitch, cold-cut dissection, peg transfer and tissue manipulation. It includes stereo endoscope calibration and robot hand–eye calibration (use dVRK camera registration format), along with high-fidelity, sharp stereo endoscopic video captured via a modern chip-on-tip endoscope. In addition to full task executions, the dataset provides consecutive subtask-level trajectories, enabling analysis of skill composition and procedural structure. Overall, it supports research in imitation learning and skill learning, surgical gesture/subtask recognition, and advanced perception (e.g., tool/tissue interaction understanding and visual tracking) in realistic surgical scenarios
+
+| |                                                                                                                                                  |
+| :--- |:-------------------------------------------------------------------------------------------------------------------------------------------------|
+| **Total Trajectories** | `1087 ex-vivo (offline) + 1061 ex-vivo (online) + 361 phantom (online)`                                                                          |
+| **Total Hours** | `2.83 ex-vivo (offline) + 3.31 ex-vivo (online) + 0.35 phantom (online)`                                                                         |
+| **Data Type** | `[ ] Clinical` `[X] Ex-Vivo` `[X] Table-Top Phantom` `[ ] Digital Simulation` `[ ] Physical Simulation` `[ ] Other (If checked, update "Other")` |
+| **License** | CC BY 4.0                                                                                                                                        |
+| **Version** | `[1.0]`                                                                                                                                          |
+
+**Note:** The user study experiments have been conducted under HIRB00000701 at Johns Hopkins University.
+
+---
+
+## 🎯 Tasks & Domain
+
+### Domain
+
+*Select the primary domain for this dataset.*
+
+- [X] **Surgical Robotics**
+- [ ] **Ultrasound Robotics**
+- [ ] **Other Healthcare Robotics** (Please specify: `[Your Domain]`)
+
+### Demonstrated Skills
+
+*List the primary skills or procedures demonstrated in this dataset.*
+
+The primary skills or procedures demonstrated in this dataset include but not limited to:
+
+- simple interrupted stitching and its subtasks
+- cold cut dissection and its subtasks
+- peg transfer and its subtasks
+- tissue manipulation and its subtasks
+- ...
+
+---
+
+## 🔬 Data Collection Details
+
+### Collection Method
+
+*How was the data collected?*
+
+- [X] **Human Teleoperation**
+- [ ] **Programmatic/State-Machine**
+- [ ] **AI Policy / Autonomous**
+- [ ] **Other** (Please specify: `[Your Method]`)
+
+### Operator Details
+
+| | Description                                                                                                                                                                 |
+| :--- |:----------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
+| **Operator Count** | `[13]`                                                                                                                                                                      |
+| **Operator Skill Level** | `[4] Expert (e.g., Surgeon, Sonographer)` <br> `[5] Intermediate (e.g., Trained Researcher)` <br> `[4] Novice (e.g., ML Researcher with minimal experience)` <br> `[ ] N/A` |
+| **Collection Period** | From `[2025-08-01]` to `[2026-01-08]`                                                                                                                                       |
+
+### Recovery Demonstrations
+
+*Does this dataset include examples of recovering from failure?*
+
+- [ ] **Yes**
+- [X] **No**
+
+**If yes, please briefly describe the recovery process:**
+
+**Note:** For your reference, this dataset includes recovery and failure demonstrations, but they are not labeled.
+
+---
+
+## 💡 Diversity Dimensions
+
+*Check all dimensions that were intentionally varied during data collection.*
+
+- [X] **Camera Position / Angle**
+- [X] **Lighting Conditions**
+- [X] **Target Object** (e.g., different phantom models, suture types)
+- [X] **Spatial Layout** (e.g., placing the target suture needle in various locations)
+- [ ] **Robot Embodiment** (if multiple robots were used)
+- [X] **Task Execution** (e.g., different techniques for the same task)
+- [X] **Background / Scene**
+- [ ] **Other** (Please specify: `[Your Dimension]`)
+
+*If you checked any of the above please briefly elaborate below.*
+
+The camera (endoscope) can move with dVRK-Si ECM (endoscopy camera manipulator) as needed. The lighting conditions will be changed due to the camera movements. For the stitch, it has both ex-vivo tissue (chicken breast) and silicon phantom cases. The orientation of the wounds could vary from cases to cases. For cold-cut dissection, the target tissue can be different. The options include chicken drumsticks, thin-slide beef/pork, pork belly. The target suture needle could be in various locations. The background can be ex-vivo tissue or the surgical drapes. The detailed subtask approach can vary from cases to cases. For example, the knot-typing could be double throw or single throw.
+
+
+---
+
+## 🛠️ Equipment & Setup
+
+### Robotic Platform(s)
+
+*List the primary robot(s) used.*
+
+- **Robot 1:** `dVRK-Si (the next generation da Vinci Reseach Kit)`
+
+
+### Sensors & Cameras
+
+*List the sensors and cameras used. Specify model names where possible. (Add and remove rows as needed)*
+
+| Type | Model/Details                                                                                                           |
+| :--- |:------------------------------------------------------------------------------------------------------------------------|
+| **Primary Camera** | `Endoscopic Camera from Cornerstone Robotics Limited, 1920x1080 @ 60fps, recoreded in 10 FPS`                           |
+| **Room/3rd Person Camera** | `Intel RealSense RGBD camera, only using RGB channel as a mono side-view camera, 1920x1080 @ 30fps, recorded in 10 FPS` |
+| **Force/Torque Sensor** | `N/A`                                                                                                                   |
+| **Medical Imager** | `N/A`                                                                                                                   |
+| **Other** | `[Specify]`                                                                                                             |
+
+**Note** The camera calibration files and hand-eye calibration matrices are just for the primary camera.
+
+---
+
+## 🎯 Action & State Space Representation (will update if needed)
+
+*Describe how actions and robot states are represented in your dataset. This is crucial for understanding data compatibility and enabling effective policy learning.*
+
+**Please refer to the subfolder README.md for more details.**
+
+---
+
+## ⏱️ Data Synchronization Approach
+
+*Describe how you achieved proper data synchronization across different sensors, cameras, and robotic systems during data collection. This is crucial for ensuring temporal alignment of all modalities in your dataset.*
+
+We use a time-synchronized multi-modal data collection framework from an accepted ICRA 2026 paper:
+```
+@inproceedings{zhou2026surgsync,
+    title={SurgSync: Time-Synchronized Multi-modal Data Collection Framework and Dataset for Surgical Robotics},
+    author={Zhou, Haoying and ... and Kazanzides, Peter},
+    booktitle={IEEE Intl. Conf. on Robotics and Automation (ICRA)},
+    year={2026}
+}
+```
+We have two desktop: one for dVRK running and one for all the other pipelines. They are connected using ROS network and its NTP. 
+
+We have two modes when data collection, and the performance is highly dependent on the hardware.
+
+**Online(-matching) Recorder**: (not uploaded yet)
+
+The design enforces strict time synchronization using multi-threading: only samples that fall within a user-defined time tolerance are admitted. This yields slightly uneven inter-sample intervals (irregular dt), 
+but it retains the natural continuity of smooth teleoperation segments and avoids label/feature drift. In our study, a time tolerance of 10 ms is selected to ensure both high time 
+alignment tightness and consecutive recorder output. 
+
+As a result, the recorder time latency is 6.36 (+/- 4.72) ms with a median of 5.58ms. The recording frequency is 4.04 (+\- 1.69) Hz. We will assemble everything assume 10 FPS.
+
+**Offline(-matching) Recorder**: (already fully uploaded)
+
+Our offline-matching approach decouples recording from time alignments to maximize
+the recording system efficiency. Therefore, this recorder produces synchronized datasets in two stages: (i) a lightweight
+recorder logs camera streams to videos (pre-synced) and raw kinematic streams to binary files to disk with minimal processing;
+(ii) an offline post-processing pipeline reconstructs a fixed-rate frame sequence and, for each frame, gathers the five
+closest samples using nearest-timestamp lookup for subsequent interpolation. Compared to the online-matching recorder (which
+pairs visual and kinematic data in real time), this two-stage design avoids tolerance-based dropping of data during capture
+yielding a higher throughput and uniform intervals between synchronized data packets at the cost of requiring more storage
+and substantial time for post-collection time-matching and interpolation.
+
+As a result, the recorder time latency is 1.35 (+/- 0.81) ms with a median of 1.33ms. And the recording frequency is a solid 10.00 FPS.
+
+**Note:** *Both recorders may have few outliners due to our suboptimal workstation hardware.*
+
+---
+
+## 👥 Attribution & Contact
+
+*Please provide attribution for the dataset creators and a point of contact.*
+
+| |                                                                                                                                                                                                                                                                                                                                                                                                           |
+| :--- |:----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
+| **Dataset Lead** | `[Haoying Zhou (student lead), Peter Kazanzides (faculty PI)]`                                                                                                                                                                                                                                                                                               |
+| **Institution** | `[Johns Hopkins University, Worcester Polytechnic Institute (Haoying Zhou)]`                                                                                                                                                                                                                                                                                                                              |
+| **Contact Email** | `[hzhou6@wpi.edu, hzhou62@jh.edu, pkaz@jhu.edu]`                                                                                                                                                                                                                                                                                                                                                          |
+| **Citation (BibTeX)** | <pre><code>@misc{[SurgSyncExt],<br>  author = {[Haoying Zhou, Peter Kazanzides, Chang Liu, Junlin Wu, Yimeng Wu, Hongjun Wu]},<br>  title = {[SurgSyncExt: Time-Synchronized Multi-Modal Dataset for Surgical Robotics]},<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> |

Surgical/jhu/lcsr/smarts/SurgSync-multitask/P2/meta/README.md

@@ -1,2 +1,229 @@
-Placeholder
-Please refer to the README.md uploaded in the parent folder, thanks!
+# SurgSync-multitask P2
+
+Canonical SMARTS leaf metadata README.
+
+- Canonical path: `Surgical/jhu/lcsr/smarts/SurgSync-multitask/P2/`
+- Source archive mapping: `online_data_part2.zip`.
+- This leaf is one canonical part of the broader JHU SMARTS dataset.
+
+The broader SMARTS dataset includes two sub-datasets: **offline_recorder** and **online_recorder**.
+
+---
+
+## 📋 At a Glance
+
+*Provide a one-sentence summary of your dataset.*
+
+Teleoperated demonstrations of a da Vinci robot (dVRK-Si) performing multiple canonical tasks on ex-vivo tissue or table-top phantoms; ~50% data collected within the Intuitive abdominal dome model. All data collection uses a modern stereo chip-on-tip endoscope. Both endoscope camera calibration and robot hand-eye calibration will be provided.
+
+---
+
+## File Structure
+
+For the dataset, it should
+
+```text
+./offline_recorder or online_recorder
+├── calibration/
+│   ├── case-*...
+│   │   ├── camera calibration
+│   │   │   ├── left.yaml
+│   │   │   ├── right.yaml
+│   │   │   └── stereo_calib_params.json
+│   │   └── hand_eye_calibration
+│   │   │   ├── PSM1/2-registration-dVRK.json
+│   │   │   └── PSM1/2-registration-open-cv.json
+├── data/
+│   └── case-*...
+├── videos/
+│   └── case-*...
+├── meta/
+│   ├── episodes.jsonl
+│   ├── episodes_stats.jsonl
+│   ├── tasks.jsonl
+│   ├── info.json
+│   └── README.md
+└── total_time.json
+```
+
+---
+
+## 📖 Dataset Overview
+
+*Briefly describe the purpose and content of this dataset. What key skills or scenarios does it demonstrate?*
+
+This dataset comprises 2500+ teleoperated expert demonstrations of multiple canonical tasks on ex-vivo tissue or table-top phantoms using a surgical robotic system. The tasks include simple interrupted stitch, cold-cut dissection, peg transfer and tissue manipulation. It includes stereo endoscope calibration and robot hand–eye calibration (use dVRK camera registration format), along with high-fidelity, sharp stereo endoscopic video captured via a modern chip-on-tip endoscope. In addition to full task executions, the dataset provides consecutive subtask-level trajectories, enabling analysis of skill composition and procedural structure. Overall, it supports research in imitation learning and skill learning, surgical gesture/subtask recognition, and advanced perception (e.g., tool/tissue interaction understanding and visual tracking) in realistic surgical scenarios
+
+| |                                                                                                                                                  |
+| :--- |:-------------------------------------------------------------------------------------------------------------------------------------------------|
+| **Total Trajectories** | `1087 ex-vivo (offline) + 1061 ex-vivo (online) + 361 phantom (online)`                                                                          |
+| **Total Hours** | `2.83 ex-vivo (offline) + 3.31 ex-vivo (online) + 0.35 phantom (online)`                                                                         |
+| **Data Type** | `[ ] Clinical` `[X] Ex-Vivo` `[X] Table-Top Phantom` `[ ] Digital Simulation` `[ ] Physical Simulation` `[ ] Other (If checked, update "Other")` |
+| **License** | CC BY 4.0                                                                                                                                        |
+| **Version** | `[1.0]`                                                                                                                                          |
+
+**Note:** The user study experiments have been conducted under HIRB00000701 at Johns Hopkins University.
+
+---
+
+## 🎯 Tasks & Domain
+
+### Domain
+
+*Select the primary domain for this dataset.*
+
+- [X] **Surgical Robotics**
+- [ ] **Ultrasound Robotics**
+- [ ] **Other Healthcare Robotics** (Please specify: `[Your Domain]`)
+
+### Demonstrated Skills
+
+*List the primary skills or procedures demonstrated in this dataset.*
+
+The primary skills or procedures demonstrated in this dataset include but not limited to:
+
+- simple interrupted stitching and its subtasks
+- cold cut dissection and its subtasks
+- peg transfer and its subtasks
+- tissue manipulation and its subtasks
+- ...
+
+---
+
+## 🔬 Data Collection Details
+
+### Collection Method
+
+*How was the data collected?*
+
+- [X] **Human Teleoperation**
+- [ ] **Programmatic/State-Machine**
+- [ ] **AI Policy / Autonomous**
+- [ ] **Other** (Please specify: `[Your Method]`)
+
+### Operator Details
+
+| | Description                                                                                                                                                                 |
+| :--- |:----------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
+| **Operator Count** | `[13]`                                                                                                                                                                      |
+| **Operator Skill Level** | `[4] Expert (e.g., Surgeon, Sonographer)` <br> `[5] Intermediate (e.g., Trained Researcher)` <br> `[4] Novice (e.g., ML Researcher with minimal experience)` <br> `[ ] N/A` |
+| **Collection Period** | From `[2025-08-01]` to `[2026-01-08]`                                                                                                                                       |
+
+### Recovery Demonstrations
+
+*Does this dataset include examples of recovering from failure?*
+
+- [ ] **Yes**
+- [X] **No**
+
+**If yes, please briefly describe the recovery process:**
+
+**Note:** For your reference, this dataset includes recovery and failure demonstrations, but they are not labeled.
+
+---
+
+## 💡 Diversity Dimensions
+
+*Check all dimensions that were intentionally varied during data collection.*
+
+- [X] **Camera Position / Angle**
+- [X] **Lighting Conditions**
+- [X] **Target Object** (e.g., different phantom models, suture types)
+- [X] **Spatial Layout** (e.g., placing the target suture needle in various locations)
+- [ ] **Robot Embodiment** (if multiple robots were used)
+- [X] **Task Execution** (e.g., different techniques for the same task)
+- [X] **Background / Scene**
+- [ ] **Other** (Please specify: `[Your Dimension]`)
+
+*If you checked any of the above please briefly elaborate below.*
+
+The camera (endoscope) can move with dVRK-Si ECM (endoscopy camera manipulator) as needed. The lighting conditions will be changed due to the camera movements. For the stitch, it has both ex-vivo tissue (chicken breast) and silicon phantom cases. The orientation of the wounds could vary from cases to cases. For cold-cut dissection, the target tissue can be different. The options include chicken drumsticks, thin-slide beef/pork, pork belly. The target suture needle could be in various locations. The background can be ex-vivo tissue or the surgical drapes. The detailed subtask approach can vary from cases to cases. For example, the knot-typing could be double throw or single throw.
+
+
+---
+
+## 🛠️ Equipment & Setup
+
+### Robotic Platform(s)
+
+*List the primary robot(s) used.*
+
+- **Robot 1:** `dVRK-Si (the next generation da Vinci Reseach Kit)`
+
+
+### Sensors & Cameras
+
+*List the sensors and cameras used. Specify model names where possible. (Add and remove rows as needed)*
+
+| Type | Model/Details                                                                                                           |
+| :--- |:------------------------------------------------------------------------------------------------------------------------|
+| **Primary Camera** | `Endoscopic Camera from Cornerstone Robotics Limited, 1920x1080 @ 60fps, recoreded in 10 FPS`                           |
+| **Room/3rd Person Camera** | `Intel RealSense RGBD camera, only using RGB channel as a mono side-view camera, 1920x1080 @ 30fps, recorded in 10 FPS` |
+| **Force/Torque Sensor** | `N/A`                                                                                                                   |
+| **Medical Imager** | `N/A`                                                                                                                   |
+| **Other** | `[Specify]`                                                                                                             |
+
+**Note** The camera calibration files and hand-eye calibration matrices are just for the primary camera.
+
+---
+
+## 🎯 Action & State Space Representation (will update if needed)
+
+*Describe how actions and robot states are represented in your dataset. This is crucial for understanding data compatibility and enabling effective policy learning.*
+
+**Please refer to the subfolder README.md for more details.**
+
+---
+
+## ⏱️ Data Synchronization Approach
+
+*Describe how you achieved proper data synchronization across different sensors, cameras, and robotic systems during data collection. This is crucial for ensuring temporal alignment of all modalities in your dataset.*
+
+We use a time-synchronized multi-modal data collection framework from an accepted ICRA 2026 paper:
+```
+@inproceedings{zhou2026surgsync,
+    title={SurgSync: Time-Synchronized Multi-modal Data Collection Framework and Dataset for Surgical Robotics},
+    author={Zhou, Haoying and ... and Kazanzides, Peter},
+    booktitle={IEEE Intl. Conf. on Robotics and Automation (ICRA)},
+    year={2026}
+}
+```
+We have two desktop: one for dVRK running and one for all the other pipelines. They are connected using ROS network and its NTP. 
+
+We have two modes when data collection, and the performance is highly dependent on the hardware.
+
+**Online(-matching) Recorder**: (not uploaded yet)
+
+The design enforces strict time synchronization using multi-threading: only samples that fall within a user-defined time tolerance are admitted. This yields slightly uneven inter-sample intervals (irregular dt), 
+but it retains the natural continuity of smooth teleoperation segments and avoids label/feature drift. In our study, a time tolerance of 10 ms is selected to ensure both high time 
+alignment tightness and consecutive recorder output. 
+
+As a result, the recorder time latency is 6.36 (+/- 4.72) ms with a median of 5.58ms. The recording frequency is 4.04 (+\- 1.69) Hz. We will assemble everything assume 10 FPS.
+
+**Offline(-matching) Recorder**: (already fully uploaded)
+
+Our offline-matching approach decouples recording from time alignments to maximize
+the recording system efficiency. Therefore, this recorder produces synchronized datasets in two stages: (i) a lightweight
+recorder logs camera streams to videos (pre-synced) and raw kinematic streams to binary files to disk with minimal processing;
+(ii) an offline post-processing pipeline reconstructs a fixed-rate frame sequence and, for each frame, gathers the five
+closest samples using nearest-timestamp lookup for subsequent interpolation. Compared to the online-matching recorder (which
+pairs visual and kinematic data in real time), this two-stage design avoids tolerance-based dropping of data during capture
+yielding a higher throughput and uniform intervals between synchronized data packets at the cost of requiring more storage
+and substantial time for post-collection time-matching and interpolation.
+
+As a result, the recorder time latency is 1.35 (+/- 0.81) ms with a median of 1.33ms. And the recording frequency is a solid 10.00 FPS.
+
+**Note:** *Both recorders may have few outliners due to our suboptimal workstation hardware.*
+
+---
+
+## 👥 Attribution & Contact
+
+*Please provide attribution for the dataset creators and a point of contact.*
+
+| |                                                                                                                                                                                                                                                                                                                                                                                                           |
+| :--- |:----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
+| **Dataset Lead** | `[Haoying Zhou (student lead), Peter Kazanzides (faculty PI)]`                                                                                                                                                                                                                                                                                               |
+| **Institution** | `[Johns Hopkins University, Worcester Polytechnic Institute (Haoying Zhou)]`                                                                                                                                                                                                                                                                                                                              |
+| **Contact Email** | `[hzhou6@wpi.edu, hzhou62@jh.edu, pkaz@jhu.edu]`                                                                                                                                                                                                                                                                                                                                                          |
+| **Citation (BibTeX)** | <pre><code>@misc{[SurgSyncExt],<br>  author = {[Haoying Zhou, Peter Kazanzides, Chang Liu, Junlin Wu, Yimeng Wu, Hongjun Wu]},<br>  title = {[SurgSyncExt: Time-Synchronized Multi-Modal Dataset for Surgical Robotics]},<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> |

Surgical/jhu/lcsr/smarts/SurgSync-multitask/P3/meta/README.md

@@ -1,2 +1,229 @@
-Placeholder
-Please refer to the README.md uploaded in the parent folder, thanks!
+# SurgSync-multitask P3
+
+Canonical SMARTS leaf metadata README.
+
+- Canonical path: `Surgical/jhu/lcsr/smarts/SurgSync-multitask/P3/`
+- Source archive mapping: `online_data_part3.zip`.
+- This leaf is one canonical part of the broader JHU SMARTS dataset.
+
+The broader SMARTS dataset includes two sub-datasets: **offline_recorder** and **online_recorder**.
+
+---
+
+## 📋 At a Glance
+
+*Provide a one-sentence summary of your dataset.*
+
+Teleoperated demonstrations of a da Vinci robot (dVRK-Si) performing multiple canonical tasks on ex-vivo tissue or table-top phantoms; ~50% data collected within the Intuitive abdominal dome model. All data collection uses a modern stereo chip-on-tip endoscope. Both endoscope camera calibration and robot hand-eye calibration will be provided.
+
+---
+
+## File Structure
+
+For the dataset, it should
+
+```text
+./offline_recorder or online_recorder
+├── calibration/
+│   ├── case-*...
+│   │   ├── camera calibration
+│   │   │   ├── left.yaml
+│   │   │   ├── right.yaml
+│   │   │   └── stereo_calib_params.json
+│   │   └── hand_eye_calibration
+│   │   │   ├── PSM1/2-registration-dVRK.json
+│   │   │   └── PSM1/2-registration-open-cv.json
+├── data/
+│   └── case-*...
+├── videos/
+│   └── case-*...
+├── meta/
+│   ├── episodes.jsonl
+│   ├── episodes_stats.jsonl
+│   ├── tasks.jsonl
+│   ├── info.json
+│   └── README.md
+└── total_time.json
+```
+
+---
+
+## 📖 Dataset Overview
+
+*Briefly describe the purpose and content of this dataset. What key skills or scenarios does it demonstrate?*
+
+This dataset comprises 2500+ teleoperated expert demonstrations of multiple canonical tasks on ex-vivo tissue or table-top phantoms using a surgical robotic system. The tasks include simple interrupted stitch, cold-cut dissection, peg transfer and tissue manipulation. It includes stereo endoscope calibration and robot hand–eye calibration (use dVRK camera registration format), along with high-fidelity, sharp stereo endoscopic video captured via a modern chip-on-tip endoscope. In addition to full task executions, the dataset provides consecutive subtask-level trajectories, enabling analysis of skill composition and procedural structure. Overall, it supports research in imitation learning and skill learning, surgical gesture/subtask recognition, and advanced perception (e.g., tool/tissue interaction understanding and visual tracking) in realistic surgical scenarios
+
+| |                                                                                                                                                  |
+| :--- |:-------------------------------------------------------------------------------------------------------------------------------------------------|
+| **Total Trajectories** | `1087 ex-vivo (offline) + 1061 ex-vivo (online) + 361 phantom (online)`                                                                          |
+| **Total Hours** | `2.83 ex-vivo (offline) + 3.31 ex-vivo (online) + 0.35 phantom (online)`                                                                         |
+| **Data Type** | `[ ] Clinical` `[X] Ex-Vivo` `[X] Table-Top Phantom` `[ ] Digital Simulation` `[ ] Physical Simulation` `[ ] Other (If checked, update "Other")` |
+| **License** | CC BY 4.0                                                                                                                                        |
+| **Version** | `[1.0]`                                                                                                                                          |
+
+**Note:** The user study experiments have been conducted under HIRB00000701 at Johns Hopkins University.
+
+---
+
+## 🎯 Tasks & Domain
+
+### Domain
+
+*Select the primary domain for this dataset.*
+
+- [X] **Surgical Robotics**
+- [ ] **Ultrasound Robotics**
+- [ ] **Other Healthcare Robotics** (Please specify: `[Your Domain]`)
+
+### Demonstrated Skills
+
+*List the primary skills or procedures demonstrated in this dataset.*
+
+The primary skills or procedures demonstrated in this dataset include but not limited to:
+
+- simple interrupted stitching and its subtasks
+- cold cut dissection and its subtasks
+- peg transfer and its subtasks
+- tissue manipulation and its subtasks
+- ...
+
+---
+
+## 🔬 Data Collection Details
+
+### Collection Method
+
+*How was the data collected?*
+
+- [X] **Human Teleoperation**
+- [ ] **Programmatic/State-Machine**
+- [ ] **AI Policy / Autonomous**
+- [ ] **Other** (Please specify: `[Your Method]`)
+
+### Operator Details
+
+| | Description                                                                                                                                                                 |
+| :--- |:----------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
+| **Operator Count** | `[13]`                                                                                                                                                                      |
+| **Operator Skill Level** | `[4] Expert (e.g., Surgeon, Sonographer)` <br> `[5] Intermediate (e.g., Trained Researcher)` <br> `[4] Novice (e.g., ML Researcher with minimal experience)` <br> `[ ] N/A` |
+| **Collection Period** | From `[2025-08-01]` to `[2026-01-08]`                                                                                                                                       |
+
+### Recovery Demonstrations
+
+*Does this dataset include examples of recovering from failure?*
+
+- [ ] **Yes**
+- [X] **No**
+
+**If yes, please briefly describe the recovery process:**
+
+**Note:** For your reference, this dataset includes recovery and failure demonstrations, but they are not labeled.
+
+---
+
+## 💡 Diversity Dimensions
+
+*Check all dimensions that were intentionally varied during data collection.*
+
+- [X] **Camera Position / Angle**
+- [X] **Lighting Conditions**
+- [X] **Target Object** (e.g., different phantom models, suture types)
+- [X] **Spatial Layout** (e.g., placing the target suture needle in various locations)
+- [ ] **Robot Embodiment** (if multiple robots were used)
+- [X] **Task Execution** (e.g., different techniques for the same task)
+- [X] **Background / Scene**
+- [ ] **Other** (Please specify: `[Your Dimension]`)
+
+*If you checked any of the above please briefly elaborate below.*
+
+The camera (endoscope) can move with dVRK-Si ECM (endoscopy camera manipulator) as needed. The lighting conditions will be changed due to the camera movements. For the stitch, it has both ex-vivo tissue (chicken breast) and silicon phantom cases. The orientation of the wounds could vary from cases to cases. For cold-cut dissection, the target tissue can be different. The options include chicken drumsticks, thin-slide beef/pork, pork belly. The target suture needle could be in various locations. The background can be ex-vivo tissue or the surgical drapes. The detailed subtask approach can vary from cases to cases. For example, the knot-typing could be double throw or single throw.
+
+
+---
+
+## 🛠️ Equipment & Setup
+
+### Robotic Platform(s)
+
+*List the primary robot(s) used.*
+
+- **Robot 1:** `dVRK-Si (the next generation da Vinci Reseach Kit)`
+
+
+### Sensors & Cameras
+
+*List the sensors and cameras used. Specify model names where possible. (Add and remove rows as needed)*
+
+| Type | Model/Details                                                                                                           |
+| :--- |:------------------------------------------------------------------------------------------------------------------------|
+| **Primary Camera** | `Endoscopic Camera from Cornerstone Robotics Limited, 1920x1080 @ 60fps, recoreded in 10 FPS`                           |
+| **Room/3rd Person Camera** | `Intel RealSense RGBD camera, only using RGB channel as a mono side-view camera, 1920x1080 @ 30fps, recorded in 10 FPS` |
+| **Force/Torque Sensor** | `N/A`                                                                                                                   |
+| **Medical Imager** | `N/A`                                                                                                                   |
+| **Other** | `[Specify]`                                                                                                             |
+
+**Note** The camera calibration files and hand-eye calibration matrices are just for the primary camera.
+
+---
+
+## 🎯 Action & State Space Representation (will update if needed)
+
+*Describe how actions and robot states are represented in your dataset. This is crucial for understanding data compatibility and enabling effective policy learning.*
+
+**Please refer to the subfolder README.md for more details.**
+
+---
+
+## ⏱️ Data Synchronization Approach
+
+*Describe how you achieved proper data synchronization across different sensors, cameras, and robotic systems during data collection. This is crucial for ensuring temporal alignment of all modalities in your dataset.*
+
+We use a time-synchronized multi-modal data collection framework from an accepted ICRA 2026 paper:
+```
+@inproceedings{zhou2026surgsync,
+    title={SurgSync: Time-Synchronized Multi-modal Data Collection Framework and Dataset for Surgical Robotics},
+    author={Zhou, Haoying and ... and Kazanzides, Peter},
+    booktitle={IEEE Intl. Conf. on Robotics and Automation (ICRA)},
+    year={2026}
+}
+```
+We have two desktop: one for dVRK running and one for all the other pipelines. They are connected using ROS network and its NTP. 
+
+We have two modes when data collection, and the performance is highly dependent on the hardware.
+
+**Online(-matching) Recorder**: (not uploaded yet)
+
+The design enforces strict time synchronization using multi-threading: only samples that fall within a user-defined time tolerance are admitted. This yields slightly uneven inter-sample intervals (irregular dt), 
+but it retains the natural continuity of smooth teleoperation segments and avoids label/feature drift. In our study, a time tolerance of 10 ms is selected to ensure both high time 
+alignment tightness and consecutive recorder output. 
+
+As a result, the recorder time latency is 6.36 (+/- 4.72) ms with a median of 5.58ms. The recording frequency is 4.04 (+\- 1.69) Hz. We will assemble everything assume 10 FPS.
+
+**Offline(-matching) Recorder**: (already fully uploaded)
+
+Our offline-matching approach decouples recording from time alignments to maximize
+the recording system efficiency. Therefore, this recorder produces synchronized datasets in two stages: (i) a lightweight
+recorder logs camera streams to videos (pre-synced) and raw kinematic streams to binary files to disk with minimal processing;
+(ii) an offline post-processing pipeline reconstructs a fixed-rate frame sequence and, for each frame, gathers the five
+closest samples using nearest-timestamp lookup for subsequent interpolation. Compared to the online-matching recorder (which
+pairs visual and kinematic data in real time), this two-stage design avoids tolerance-based dropping of data during capture
+yielding a higher throughput and uniform intervals between synchronized data packets at the cost of requiring more storage
+and substantial time for post-collection time-matching and interpolation.
+
+As a result, the recorder time latency is 1.35 (+/- 0.81) ms with a median of 1.33ms. And the recording frequency is a solid 10.00 FPS.
+
+**Note:** *Both recorders may have few outliners due to our suboptimal workstation hardware.*
+
+---
+
+## 👥 Attribution & Contact
+
+*Please provide attribution for the dataset creators and a point of contact.*
+
+| |                                                                                                                                                                                                                                                                                                                                                                                                           |
+| :--- |:----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
+| **Dataset Lead** | `[Haoying Zhou (student lead), Peter Kazanzides (faculty PI)]`                                                                                                                                                                                                                                                                                               |
+| **Institution** | `[Johns Hopkins University, Worcester Polytechnic Institute (Haoying Zhou)]`                                                                                                                                                                                                                                                                                                                              |
+| **Contact Email** | `[hzhou6@wpi.edu, hzhou62@jh.edu, pkaz@jhu.edu]`                                                                                                                                                                                                                                                                                                                                                          |
+| **Citation (BibTeX)** | <pre><code>@misc{[SurgSyncExt],<br>  author = {[Haoying Zhou, Peter Kazanzides, Chang Liu, Junlin Wu, Yimeng Wu, Hongjun Wu]},<br>  title = {[SurgSyncExt: Time-Synchronized Multi-Modal Dataset for Surgical Robotics]},<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> |

Surgical/jhu/lcsr/smarts/SurgSync-multitask/P4/meta/README.md

@@ -1,2 +1,229 @@
-Placeholder
-Please refer to the README.md uploaded in the parent folder, thanks!
+# SurgSync-multitask P4
+
+Canonical SMARTS leaf metadata README.
+
+- Canonical path: `Surgical/jhu/lcsr/smarts/SurgSync-multitask/P4/`
+- Source archive mapping: `online_data_part4.zip`.
+- This leaf is one canonical part of the broader JHU SMARTS dataset.
+
+The broader SMARTS dataset includes two sub-datasets: **offline_recorder** and **online_recorder**.
+
+---
+
+## 📋 At a Glance
+
+*Provide a one-sentence summary of your dataset.*
+
+Teleoperated demonstrations of a da Vinci robot (dVRK-Si) performing multiple canonical tasks on ex-vivo tissue or table-top phantoms; ~50% data collected within the Intuitive abdominal dome model. All data collection uses a modern stereo chip-on-tip endoscope. Both endoscope camera calibration and robot hand-eye calibration will be provided.
+
+---
+
+## File Structure
+
+For the dataset, it should
+
+```text
+./offline_recorder or online_recorder
+├── calibration/
+│   ├── case-*...
+│   │   ├── camera calibration
+│   │   │   ├── left.yaml
+│   │   │   ├── right.yaml
+│   │   │   └── stereo_calib_params.json
+│   │   └── hand_eye_calibration
+│   │   │   ├── PSM1/2-registration-dVRK.json
+│   │   │   └── PSM1/2-registration-open-cv.json
+├── data/
+│   └── case-*...
+├── videos/
+│   └── case-*...
+├── meta/
+│   ├── episodes.jsonl
+│   ├── episodes_stats.jsonl
+│   ├── tasks.jsonl
+│   ├── info.json
+│   └── README.md
+└── total_time.json
+```
+
+---
+
+## 📖 Dataset Overview
+
+*Briefly describe the purpose and content of this dataset. What key skills or scenarios does it demonstrate?*
+
+This dataset comprises 2500+ teleoperated expert demonstrations of multiple canonical tasks on ex-vivo tissue or table-top phantoms using a surgical robotic system. The tasks include simple interrupted stitch, cold-cut dissection, peg transfer and tissue manipulation. It includes stereo endoscope calibration and robot hand–eye calibration (use dVRK camera registration format), along with high-fidelity, sharp stereo endoscopic video captured via a modern chip-on-tip endoscope. In addition to full task executions, the dataset provides consecutive subtask-level trajectories, enabling analysis of skill composition and procedural structure. Overall, it supports research in imitation learning and skill learning, surgical gesture/subtask recognition, and advanced perception (e.g., tool/tissue interaction understanding and visual tracking) in realistic surgical scenarios
+
+| |                                                                                                                                                  |
+| :--- |:-------------------------------------------------------------------------------------------------------------------------------------------------|
+| **Total Trajectories** | `1087 ex-vivo (offline) + 1061 ex-vivo (online) + 361 phantom (online)`                                                                          |
+| **Total Hours** | `2.83 ex-vivo (offline) + 3.31 ex-vivo (online) + 0.35 phantom (online)`                                                                         |
+| **Data Type** | `[ ] Clinical` `[X] Ex-Vivo` `[X] Table-Top Phantom` `[ ] Digital Simulation` `[ ] Physical Simulation` `[ ] Other (If checked, update "Other")` |
+| **License** | CC BY 4.0                                                                                                                                        |
+| **Version** | `[1.0]`                                                                                                                                          |
+
+**Note:** The user study experiments have been conducted under HIRB00000701 at Johns Hopkins University.
+
+---
+
+## 🎯 Tasks & Domain
+
+### Domain
+
+*Select the primary domain for this dataset.*
+
+- [X] **Surgical Robotics**
+- [ ] **Ultrasound Robotics**
+- [ ] **Other Healthcare Robotics** (Please specify: `[Your Domain]`)
+
+### Demonstrated Skills
+
+*List the primary skills or procedures demonstrated in this dataset.*
+
+The primary skills or procedures demonstrated in this dataset include but not limited to:
+
+- simple interrupted stitching and its subtasks
+- cold cut dissection and its subtasks
+- peg transfer and its subtasks
+- tissue manipulation and its subtasks
+- ...
+
+---
+
+## 🔬 Data Collection Details
+
+### Collection Method
+
+*How was the data collected?*
+
+- [X] **Human Teleoperation**
+- [ ] **Programmatic/State-Machine**
+- [ ] **AI Policy / Autonomous**
+- [ ] **Other** (Please specify: `[Your Method]`)
+
+### Operator Details
+
+| | Description                                                                                                                                                                 |
+| :--- |:----------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
+| **Operator Count** | `[13]`                                                                                                                                                                      |
+| **Operator Skill Level** | `[4] Expert (e.g., Surgeon, Sonographer)` <br> `[5] Intermediate (e.g., Trained Researcher)` <br> `[4] Novice (e.g., ML Researcher with minimal experience)` <br> `[ ] N/A` |
+| **Collection Period** | From `[2025-08-01]` to `[2026-01-08]`                                                                                                                                       |
+
+### Recovery Demonstrations
+
+*Does this dataset include examples of recovering from failure?*
+
+- [ ] **Yes**
+- [X] **No**
+
+**If yes, please briefly describe the recovery process:**
+
+**Note:** For your reference, this dataset includes recovery and failure demonstrations, but they are not labeled.
+
+---
+
+## 💡 Diversity Dimensions
+
+*Check all dimensions that were intentionally varied during data collection.*
+
+- [X] **Camera Position / Angle**
+- [X] **Lighting Conditions**
+- [X] **Target Object** (e.g., different phantom models, suture types)
+- [X] **Spatial Layout** (e.g., placing the target suture needle in various locations)
+- [ ] **Robot Embodiment** (if multiple robots were used)
+- [X] **Task Execution** (e.g., different techniques for the same task)
+- [X] **Background / Scene**
+- [ ] **Other** (Please specify: `[Your Dimension]`)
+
+*If you checked any of the above please briefly elaborate below.*
+
+The camera (endoscope) can move with dVRK-Si ECM (endoscopy camera manipulator) as needed. The lighting conditions will be changed due to the camera movements. For the stitch, it has both ex-vivo tissue (chicken breast) and silicon phantom cases. The orientation of the wounds could vary from cases to cases. For cold-cut dissection, the target tissue can be different. The options include chicken drumsticks, thin-slide beef/pork, pork belly. The target suture needle could be in various locations. The background can be ex-vivo tissue or the surgical drapes. The detailed subtask approach can vary from cases to cases. For example, the knot-typing could be double throw or single throw.
+
+
+---
+
+## 🛠️ Equipment & Setup
+
+### Robotic Platform(s)
+
+*List the primary robot(s) used.*
+
+- **Robot 1:** `dVRK-Si (the next generation da Vinci Reseach Kit)`
+
+
+### Sensors & Cameras
+
+*List the sensors and cameras used. Specify model names where possible. (Add and remove rows as needed)*
+
+| Type | Model/Details                                                                                                           |
+| :--- |:------------------------------------------------------------------------------------------------------------------------|
+| **Primary Camera** | `Endoscopic Camera from Cornerstone Robotics Limited, 1920x1080 @ 60fps, recoreded in 10 FPS`                           |
+| **Room/3rd Person Camera** | `Intel RealSense RGBD camera, only using RGB channel as a mono side-view camera, 1920x1080 @ 30fps, recorded in 10 FPS` |
+| **Force/Torque Sensor** | `N/A`                                                                                                                   |
+| **Medical Imager** | `N/A`                                                                                                                   |
+| **Other** | `[Specify]`                                                                                                             |
+
+**Note** The camera calibration files and hand-eye calibration matrices are just for the primary camera.
+
+---
+
+## 🎯 Action & State Space Representation (will update if needed)
+
+*Describe how actions and robot states are represented in your dataset. This is crucial for understanding data compatibility and enabling effective policy learning.*
+
+**Please refer to the subfolder README.md for more details.**
+
+---
+
+## ⏱️ Data Synchronization Approach
+
+*Describe how you achieved proper data synchronization across different sensors, cameras, and robotic systems during data collection. This is crucial for ensuring temporal alignment of all modalities in your dataset.*
+
+We use a time-synchronized multi-modal data collection framework from an accepted ICRA 2026 paper:
+```
+@inproceedings{zhou2026surgsync,
+    title={SurgSync: Time-Synchronized Multi-modal Data Collection Framework and Dataset for Surgical Robotics},
+    author={Zhou, Haoying and ... and Kazanzides, Peter},
+    booktitle={IEEE Intl. Conf. on Robotics and Automation (ICRA)},
+    year={2026}
+}
+```
+We have two desktop: one for dVRK running and one for all the other pipelines. They are connected using ROS network and its NTP. 
+
+We have two modes when data collection, and the performance is highly dependent on the hardware.
+
+**Online(-matching) Recorder**: (not uploaded yet)
+
+The design enforces strict time synchronization using multi-threading: only samples that fall within a user-defined time tolerance are admitted. This yields slightly uneven inter-sample intervals (irregular dt), 
+but it retains the natural continuity of smooth teleoperation segments and avoids label/feature drift. In our study, a time tolerance of 10 ms is selected to ensure both high time 
+alignment tightness and consecutive recorder output. 
+
+As a result, the recorder time latency is 6.36 (+/- 4.72) ms with a median of 5.58ms. The recording frequency is 4.04 (+\- 1.69) Hz. We will assemble everything assume 10 FPS.
+
+**Offline(-matching) Recorder**: (already fully uploaded)
+
+Our offline-matching approach decouples recording from time alignments to maximize
+the recording system efficiency. Therefore, this recorder produces synchronized datasets in two stages: (i) a lightweight
+recorder logs camera streams to videos (pre-synced) and raw kinematic streams to binary files to disk with minimal processing;
+(ii) an offline post-processing pipeline reconstructs a fixed-rate frame sequence and, for each frame, gathers the five
+closest samples using nearest-timestamp lookup for subsequent interpolation. Compared to the online-matching recorder (which
+pairs visual and kinematic data in real time), this two-stage design avoids tolerance-based dropping of data during capture
+yielding a higher throughput and uniform intervals between synchronized data packets at the cost of requiring more storage
+and substantial time for post-collection time-matching and interpolation.
+
+As a result, the recorder time latency is 1.35 (+/- 0.81) ms with a median of 1.33ms. And the recording frequency is a solid 10.00 FPS.
+
+**Note:** *Both recorders may have few outliners due to our suboptimal workstation hardware.*
+
+---
+
+## 👥 Attribution & Contact
+
+*Please provide attribution for the dataset creators and a point of contact.*
+
+| |                                                                                                                                                                                                                                                                                                                                                                                                           |
+| :--- |:----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
+| **Dataset Lead** | `[Haoying Zhou (student lead), Peter Kazanzides (faculty PI)]`                                                                                                                                                                                                                                                                                               |
+| **Institution** | `[Johns Hopkins University, Worcester Polytechnic Institute (Haoying Zhou)]`                                                                                                                                                                                                                                                                                                                              |
+| **Contact Email** | `[hzhou6@wpi.edu, hzhou62@jh.edu, pkaz@jhu.edu]`                                                                                                                                                                                                                                                                                                                                                          |
+| **Citation (BibTeX)** | <pre><code>@misc{[SurgSyncExt],<br>  author = {[Haoying Zhou, Peter Kazanzides, Chang Liu, Junlin Wu, Yimeng Wu, Hongjun Wu]},<br>  title = {[SurgSyncExt: Time-Synchronized Multi-Modal Dataset for Surgical Robotics]},<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> |

Surgical/jhu/lcsr/smarts/SurgSync-multitask/README.md

@@ -4,7 +4,7 @@ Canonical parent documentation for the online JHU SMARTS subsets.
 
 - Canonical parent path: `Surgical/jhu/lcsr/smarts/SurgSync-multitask/`
 - Part mapping: `P1 -> online_data_part1`, `P2 -> online_data_part2`, `P3 -> online_data_part3`, `P4 -> online_data_part4`.
-- The payload leaves for `P1..P4` are being prepared from the Draco source archives and are not yet all published on Hugging Face.
+- The payload leaves for `P1..P4` are published on Hugging Face.
 
 ---
 

Surgical/jhu/lcsr/smarts/SurgSync-stitch-coldcut/P1/meta/README.md

@@ -1,2 +1,229 @@
-Placeholder
-Please refer to the README.md uploaded in the parent folder, thanks!
+# SurgSync-stitch-coldcut P1
+
+Canonical SMARTS leaf metadata README.
+
+- Canonical path: `Surgical/jhu/lcsr/smarts/SurgSync-stitch-coldcut/P1/`
+- Legacy source mapping: `Surgical/jhu/lscr/smarts/offline_recorder_extracted/offline_data_part1`.
+- This leaf is one canonical part of the broader JHU SMARTS dataset.
+
+The broader SMARTS dataset includes two sub-datasets: **offline_recorder** and **online_recorder**.
+
+---
+
+## 📋 At a Glance
+
+*Provide a one-sentence summary of your dataset.*
+
+Teleoperated demonstrations of a da Vinci robot (dVRK-Si) performing multiple canonical tasks on ex-vivo tissue or table-top phantoms; ~50% data collected within the Intuitive abdominal dome model. All data collection uses a modern stereo chip-on-tip endoscope. Both endoscope camera calibration and robot hand-eye calibration will be provided.
+
+---
+
+## File Structure
+
+For the dataset, it should
+
+```text
+./offline_recorder or online_recorder
+├── calibration/
+│   ├── case-*...
+│   │   ├── camera calibration
+│   │   │   ├── left.yaml
+│   │   │   ├── right.yaml
+│   │   │   └── stereo_calib_params.json
+│   │   └── hand_eye_calibration
+│   │   │   ├── PSM1/2-registration-dVRK.json
+│   │   │   └── PSM1/2-registration-open-cv.json
+├── data/
+│   └── case-*...
+├── videos/
+│   └── case-*...
+├── meta/
+│   ├── episodes.jsonl
+│   ├── episodes_stats.jsonl
+│   ├── tasks.jsonl
+│   ├── info.json
+│   └── README.md
+└── total_time.json
+```
+
+---
+
+## 📖 Dataset Overview
+
+*Briefly describe the purpose and content of this dataset. What key skills or scenarios does it demonstrate?*
+
+This dataset comprises 2500+ teleoperated expert demonstrations of multiple canonical tasks on ex-vivo tissue or table-top phantoms using a surgical robotic system. The tasks include simple interrupted stitch, cold-cut dissection, peg transfer and tissue manipulation. It includes stereo endoscope calibration and robot hand–eye calibration (use dVRK camera registration format), along with high-fidelity, sharp stereo endoscopic video captured via a modern chip-on-tip endoscope. In addition to full task executions, the dataset provides consecutive subtask-level trajectories, enabling analysis of skill composition and procedural structure. Overall, it supports research in imitation learning and skill learning, surgical gesture/subtask recognition, and advanced perception (e.g., tool/tissue interaction understanding and visual tracking) in realistic surgical scenarios
+
+| |                                                                                                                                                  |
+| :--- |:-------------------------------------------------------------------------------------------------------------------------------------------------|
+| **Total Trajectories** | `1087 ex-vivo (offline) + 1061 ex-vivo (online) + 361 phantom (online)`                                                                          |
+| **Total Hours** | `2.83 ex-vivo (offline) + 3.31 ex-vivo (online) + 0.35 phantom (online)`                                                                         |
+| **Data Type** | `[ ] Clinical` `[X] Ex-Vivo` `[X] Table-Top Phantom` `[ ] Digital Simulation` `[ ] Physical Simulation` `[ ] Other (If checked, update "Other")` |
+| **License** | CC BY 4.0                                                                                                                                        |
+| **Version** | `[1.0]`                                                                                                                                          |
+
+**Note:** The user study experiments have been conducted under HIRB00000701 at Johns Hopkins University.
+
+---
+
+## 🎯 Tasks & Domain
+
+### Domain
+
+*Select the primary domain for this dataset.*
+
+- [X] **Surgical Robotics**
+- [ ] **Ultrasound Robotics**
+- [ ] **Other Healthcare Robotics** (Please specify: `[Your Domain]`)
+
+### Demonstrated Skills
+
+*List the primary skills or procedures demonstrated in this dataset.*
+
+The primary skills or procedures demonstrated in this dataset include but not limited to:
+
+- simple interrupted stitching and its subtasks
+- cold cut dissection and its subtasks
+- peg transfer and its subtasks
+- tissue manipulation and its subtasks
+- ...
+
+---
+
+## 🔬 Data Collection Details
+
+### Collection Method
+
+*How was the data collected?*
+
+- [X] **Human Teleoperation**
+- [ ] **Programmatic/State-Machine**
+- [ ] **AI Policy / Autonomous**
+- [ ] **Other** (Please specify: `[Your Method]`)
+
+### Operator Details
+
+| | Description                                                                                                                                                                 |
+| :--- |:----------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
+| **Operator Count** | `[13]`                                                                                                                                                                      |
+| **Operator Skill Level** | `[4] Expert (e.g., Surgeon, Sonographer)` <br> `[5] Intermediate (e.g., Trained Researcher)` <br> `[4] Novice (e.g., ML Researcher with minimal experience)` <br> `[ ] N/A` |
+| **Collection Period** | From `[2025-08-01]` to `[2026-01-08]`                                                                                                                                       |
+
+### Recovery Demonstrations
+
+*Does this dataset include examples of recovering from failure?*
+
+- [ ] **Yes**
+- [X] **No**
+
+**If yes, please briefly describe the recovery process:**
+
+**Note:** For your reference, this dataset includes recovery and failure demonstrations, but they are not labeled.
+
+---
+
+## 💡 Diversity Dimensions
+
+*Check all dimensions that were intentionally varied during data collection.*
+
+- [X] **Camera Position / Angle**
+- [X] **Lighting Conditions**
+- [X] **Target Object** (e.g., different phantom models, suture types)
+- [X] **Spatial Layout** (e.g., placing the target suture needle in various locations)
+- [ ] **Robot Embodiment** (if multiple robots were used)
+- [X] **Task Execution** (e.g., different techniques for the same task)
+- [X] **Background / Scene**
+- [ ] **Other** (Please specify: `[Your Dimension]`)
+
+*If you checked any of the above please briefly elaborate below.*
+
+The camera (endoscope) can move with dVRK-Si ECM (endoscopy camera manipulator) as needed. The lighting conditions will be changed due to the camera movements. For the stitch, it has both ex-vivo tissue (chicken breast) and silicon phantom cases. The orientation of the wounds could vary from cases to cases. For cold-cut dissection, the target tissue can be different. The options include chicken drumsticks, thin-slide beef/pork, pork belly. The target suture needle could be in various locations. The background can be ex-vivo tissue or the surgical drapes. The detailed subtask approach can vary from cases to cases. For example, the knot-typing could be double throw or single throw.
+
+
+---
+
+## 🛠️ Equipment & Setup
+
+### Robotic Platform(s)
+
+*List the primary robot(s) used.*
+
+- **Robot 1:** `dVRK-Si (the next generation da Vinci Reseach Kit)`
+
+
+### Sensors & Cameras
+
+*List the sensors and cameras used. Specify model names where possible. (Add and remove rows as needed)*
+
+| Type | Model/Details                                                                                                           |
+| :--- |:------------------------------------------------------------------------------------------------------------------------|
+| **Primary Camera** | `Endoscopic Camera from Cornerstone Robotics Limited, 1920x1080 @ 60fps, recoreded in 10 FPS`                           |
+| **Room/3rd Person Camera** | `Intel RealSense RGBD camera, only using RGB channel as a mono side-view camera, 1920x1080 @ 30fps, recorded in 10 FPS` |
+| **Force/Torque Sensor** | `N/A`                                                                                                                   |
+| **Medical Imager** | `N/A`                                                                                                                   |
+| **Other** | `[Specify]`                                                                                                             |
+
+**Note** The camera calibration files and hand-eye calibration matrices are just for the primary camera.
+
+---
+
+## 🎯 Action & State Space Representation (will update if needed)
+
+*Describe how actions and robot states are represented in your dataset. This is crucial for understanding data compatibility and enabling effective policy learning.*
+
+**Please refer to the subfolder README.md for more details.**
+
+---
+
+## ⏱️ Data Synchronization Approach
+
+*Describe how you achieved proper data synchronization across different sensors, cameras, and robotic systems during data collection. This is crucial for ensuring temporal alignment of all modalities in your dataset.*
+
+We use a time-synchronized multi-modal data collection framework from an accepted ICRA 2026 paper:
+```
+@inproceedings{zhou2026surgsync,
+    title={SurgSync: Time-Synchronized Multi-modal Data Collection Framework and Dataset for Surgical Robotics},
+    author={Zhou, Haoying and ... and Kazanzides, Peter},
+    booktitle={IEEE Intl. Conf. on Robotics and Automation (ICRA)},
+    year={2026}
+}
+```
+We have two desktop: one for dVRK running and one for all the other pipelines. They are connected using ROS network and its NTP. 
+
+We have two modes when data collection, and the performance is highly dependent on the hardware.
+
+**Online(-matching) Recorder**: (not uploaded yet)
+
+The design enforces strict time synchronization using multi-threading: only samples that fall within a user-defined time tolerance are admitted. This yields slightly uneven inter-sample intervals (irregular dt), 
+but it retains the natural continuity of smooth teleoperation segments and avoids label/feature drift. In our study, a time tolerance of 10 ms is selected to ensure both high time 
+alignment tightness and consecutive recorder output. 
+
+As a result, the recorder time latency is 6.36 (+/- 4.72) ms with a median of 5.58ms. The recording frequency is 4.04 (+\- 1.69) Hz. We will assemble everything assume 10 FPS.
+
+**Offline(-matching) Recorder**: (already fully uploaded)
+
+Our offline-matching approach decouples recording from time alignments to maximize
+the recording system efficiency. Therefore, this recorder produces synchronized datasets in two stages: (i) a lightweight
+recorder logs camera streams to videos (pre-synced) and raw kinematic streams to binary files to disk with minimal processing;
+(ii) an offline post-processing pipeline reconstructs a fixed-rate frame sequence and, for each frame, gathers the five
+closest samples using nearest-timestamp lookup for subsequent interpolation. Compared to the online-matching recorder (which
+pairs visual and kinematic data in real time), this two-stage design avoids tolerance-based dropping of data during capture
+yielding a higher throughput and uniform intervals between synchronized data packets at the cost of requiring more storage
+and substantial time for post-collection time-matching and interpolation.
+
+As a result, the recorder time latency is 1.35 (+/- 0.81) ms with a median of 1.33ms. And the recording frequency is a solid 10.00 FPS.
+
+**Note:** *Both recorders may have few outliners due to our suboptimal workstation hardware.*
+
+---
+
+## 👥 Attribution & Contact
+
+*Please provide attribution for the dataset creators and a point of contact.*
+
+| |                                                                                                                                                                                                                                                                                                                                                                                                           |
+| :--- |:----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
+| **Dataset Lead** | `[Haoying Zhou (student lead), Peter Kazanzides (faculty PI)]`                                                                                                                                                                                                                                                                                               |
+| **Institution** | `[Johns Hopkins University, Worcester Polytechnic Institute (Haoying Zhou)]`                                                                                                                                                                                                                                                                                                                              |
+| **Contact Email** | `[hzhou6@wpi.edu, hzhou62@jh.edu, pkaz@jhu.edu]`                                                                                                                                                                                                                                                                                                                                                          |
+| **Citation (BibTeX)** | <pre><code>@misc{[SurgSyncExt],<br>  author = {[Haoying Zhou, Peter Kazanzides, Chang Liu, Junlin Wu, Yimeng Wu, Hongjun Wu]},<br>  title = {[SurgSyncExt: Time-Synchronized Multi-Modal Dataset for Surgical Robotics]},<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> |

Surgical/jhu/lcsr/smarts/SurgSync-stitch-coldcut/P2/meta/README.md

@@ -1,2 +1,229 @@
-Placeholder
-Please refer to the README.md uploaded in the parent folder, thanks!
+# SurgSync-stitch-coldcut P2
+
+Canonical SMARTS leaf metadata README.
+
+- Canonical path: `Surgical/jhu/lcsr/smarts/SurgSync-stitch-coldcut/P2/`
+- Legacy source mapping: `Surgical/jhu/lscr/smarts/offline_recorder_extracted/offline_data_part2`.
+- This leaf is one canonical part of the broader JHU SMARTS dataset.
+
+The broader SMARTS dataset includes two sub-datasets: **offline_recorder** and **online_recorder**.
+
+---
+
+## 📋 At a Glance
+
+*Provide a one-sentence summary of your dataset.*
+
+Teleoperated demonstrations of a da Vinci robot (dVRK-Si) performing multiple canonical tasks on ex-vivo tissue or table-top phantoms; ~50% data collected within the Intuitive abdominal dome model. All data collection uses a modern stereo chip-on-tip endoscope. Both endoscope camera calibration and robot hand-eye calibration will be provided.
+
+---
+
+## File Structure
+
+For the dataset, it should
+
+```text
+./offline_recorder or online_recorder
+├── calibration/
+│   ├── case-*...
+│   │   ├── camera calibration
+│   │   │   ├── left.yaml
+│   │   │   ├── right.yaml
+│   │   │   └── stereo_calib_params.json
+│   │   └── hand_eye_calibration
+│   │   │   ├── PSM1/2-registration-dVRK.json
+│   │   │   └── PSM1/2-registration-open-cv.json
+├── data/
+│   └── case-*...
+├── videos/
+│   └── case-*...
+├── meta/
+│   ├── episodes.jsonl
+│   ├── episodes_stats.jsonl
+│   ├── tasks.jsonl
+│   ├── info.json
+│   └── README.md
+└── total_time.json
+```
+
+---
+
+## 📖 Dataset Overview
+
+*Briefly describe the purpose and content of this dataset. What key skills or scenarios does it demonstrate?*
+
+This dataset comprises 2500+ teleoperated expert demonstrations of multiple canonical tasks on ex-vivo tissue or table-top phantoms using a surgical robotic system. The tasks include simple interrupted stitch, cold-cut dissection, peg transfer and tissue manipulation. It includes stereo endoscope calibration and robot hand–eye calibration (use dVRK camera registration format), along with high-fidelity, sharp stereo endoscopic video captured via a modern chip-on-tip endoscope. In addition to full task executions, the dataset provides consecutive subtask-level trajectories, enabling analysis of skill composition and procedural structure. Overall, it supports research in imitation learning and skill learning, surgical gesture/subtask recognition, and advanced perception (e.g., tool/tissue interaction understanding and visual tracking) in realistic surgical scenarios
+
+| |                                                                                                                                                  |
+| :--- |:-------------------------------------------------------------------------------------------------------------------------------------------------|
+| **Total Trajectories** | `1087 ex-vivo (offline) + 1061 ex-vivo (online) + 361 phantom (online)`                                                                          |
+| **Total Hours** | `2.83 ex-vivo (offline) + 3.31 ex-vivo (online) + 0.35 phantom (online)`                                                                         |
+| **Data Type** | `[ ] Clinical` `[X] Ex-Vivo` `[X] Table-Top Phantom` `[ ] Digital Simulation` `[ ] Physical Simulation` `[ ] Other (If checked, update "Other")` |
+| **License** | CC BY 4.0                                                                                                                                        |
+| **Version** | `[1.0]`                                                                                                                                          |
+
+**Note:** The user study experiments have been conducted under HIRB00000701 at Johns Hopkins University.
+
+---
+
+## 🎯 Tasks & Domain
+
+### Domain
+
+*Select the primary domain for this dataset.*
+
+- [X] **Surgical Robotics**
+- [ ] **Ultrasound Robotics**
+- [ ] **Other Healthcare Robotics** (Please specify: `[Your Domain]`)
+
+### Demonstrated Skills
+
+*List the primary skills or procedures demonstrated in this dataset.*
+
+The primary skills or procedures demonstrated in this dataset include but not limited to:
+
+- simple interrupted stitching and its subtasks
+- cold cut dissection and its subtasks
+- peg transfer and its subtasks
+- tissue manipulation and its subtasks
+- ...
+
+---
+
+## 🔬 Data Collection Details
+
+### Collection Method
+
+*How was the data collected?*
+
+- [X] **Human Teleoperation**
+- [ ] **Programmatic/State-Machine**
+- [ ] **AI Policy / Autonomous**
+- [ ] **Other** (Please specify: `[Your Method]`)
+
+### Operator Details
+
+| | Description                                                                                                                                                                 |
+| :--- |:----------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
+| **Operator Count** | `[13]`                                                                                                                                                                      |
+| **Operator Skill Level** | `[4] Expert (e.g., Surgeon, Sonographer)` <br> `[5] Intermediate (e.g., Trained Researcher)` <br> `[4] Novice (e.g., ML Researcher with minimal experience)` <br> `[ ] N/A` |
+| **Collection Period** | From `[2025-08-01]` to `[2026-01-08]`                                                                                                                                       |
+
+### Recovery Demonstrations
+
+*Does this dataset include examples of recovering from failure?*
+
+- [ ] **Yes**
+- [X] **No**
+
+**If yes, please briefly describe the recovery process:**
+
+**Note:** For your reference, this dataset includes recovery and failure demonstrations, but they are not labeled.
+
+---
+
+## 💡 Diversity Dimensions
+
+*Check all dimensions that were intentionally varied during data collection.*
+
+- [X] **Camera Position / Angle**
+- [X] **Lighting Conditions**
+- [X] **Target Object** (e.g., different phantom models, suture types)
+- [X] **Spatial Layout** (e.g., placing the target suture needle in various locations)
+- [ ] **Robot Embodiment** (if multiple robots were used)
+- [X] **Task Execution** (e.g., different techniques for the same task)
+- [X] **Background / Scene**
+- [ ] **Other** (Please specify: `[Your Dimension]`)
+
+*If you checked any of the above please briefly elaborate below.*
+
+The camera (endoscope) can move with dVRK-Si ECM (endoscopy camera manipulator) as needed. The lighting conditions will be changed due to the camera movements. For the stitch, it has both ex-vivo tissue (chicken breast) and silicon phantom cases. The orientation of the wounds could vary from cases to cases. For cold-cut dissection, the target tissue can be different. The options include chicken drumsticks, thin-slide beef/pork, pork belly. The target suture needle could be in various locations. The background can be ex-vivo tissue or the surgical drapes. The detailed subtask approach can vary from cases to cases. For example, the knot-typing could be double throw or single throw.
+
+
+---
+
+## 🛠️ Equipment & Setup
+
+### Robotic Platform(s)
+
+*List the primary robot(s) used.*
+
+- **Robot 1:** `dVRK-Si (the next generation da Vinci Reseach Kit)`
+
+
+### Sensors & Cameras
+
+*List the sensors and cameras used. Specify model names where possible. (Add and remove rows as needed)*
+
+| Type | Model/Details                                                                                                           |
+| :--- |:------------------------------------------------------------------------------------------------------------------------|
+| **Primary Camera** | `Endoscopic Camera from Cornerstone Robotics Limited, 1920x1080 @ 60fps, recoreded in 10 FPS`                           |
+| **Room/3rd Person Camera** | `Intel RealSense RGBD camera, only using RGB channel as a mono side-view camera, 1920x1080 @ 30fps, recorded in 10 FPS` |
+| **Force/Torque Sensor** | `N/A`                                                                                                                   |
+| **Medical Imager** | `N/A`                                                                                                                   |
+| **Other** | `[Specify]`                                                                                                             |
+
+**Note** The camera calibration files and hand-eye calibration matrices are just for the primary camera.
+
+---
+
+## 🎯 Action & State Space Representation (will update if needed)
+
+*Describe how actions and robot states are represented in your dataset. This is crucial for understanding data compatibility and enabling effective policy learning.*
+
+**Please refer to the subfolder README.md for more details.**
+
+---
+
+## ⏱️ Data Synchronization Approach
+
+*Describe how you achieved proper data synchronization across different sensors, cameras, and robotic systems during data collection. This is crucial for ensuring temporal alignment of all modalities in your dataset.*
+
+We use a time-synchronized multi-modal data collection framework from an accepted ICRA 2026 paper:
+```
+@inproceedings{zhou2026surgsync,
+    title={SurgSync: Time-Synchronized Multi-modal Data Collection Framework and Dataset for Surgical Robotics},
+    author={Zhou, Haoying and ... and Kazanzides, Peter},
+    booktitle={IEEE Intl. Conf. on Robotics and Automation (ICRA)},
+    year={2026}
+}
+```
+We have two desktop: one for dVRK running and one for all the other pipelines. They are connected using ROS network and its NTP. 
+
+We have two modes when data collection, and the performance is highly dependent on the hardware.
+
+**Online(-matching) Recorder**: (not uploaded yet)
+
+The design enforces strict time synchronization using multi-threading: only samples that fall within a user-defined time tolerance are admitted. This yields slightly uneven inter-sample intervals (irregular dt), 
+but it retains the natural continuity of smooth teleoperation segments and avoids label/feature drift. In our study, a time tolerance of 10 ms is selected to ensure both high time 
+alignment tightness and consecutive recorder output. 
+
+As a result, the recorder time latency is 6.36 (+/- 4.72) ms with a median of 5.58ms. The recording frequency is 4.04 (+\- 1.69) Hz. We will assemble everything assume 10 FPS.
+
+**Offline(-matching) Recorder**: (already fully uploaded)
+
+Our offline-matching approach decouples recording from time alignments to maximize
+the recording system efficiency. Therefore, this recorder produces synchronized datasets in two stages: (i) a lightweight
+recorder logs camera streams to videos (pre-synced) and raw kinematic streams to binary files to disk with minimal processing;
+(ii) an offline post-processing pipeline reconstructs a fixed-rate frame sequence and, for each frame, gathers the five
+closest samples using nearest-timestamp lookup for subsequent interpolation. Compared to the online-matching recorder (which
+pairs visual and kinematic data in real time), this two-stage design avoids tolerance-based dropping of data during capture
+yielding a higher throughput and uniform intervals between synchronized data packets at the cost of requiring more storage
+and substantial time for post-collection time-matching and interpolation.
+
+As a result, the recorder time latency is 1.35 (+/- 0.81) ms with a median of 1.33ms. And the recording frequency is a solid 10.00 FPS.
+
+**Note:** *Both recorders may have few outliners due to our suboptimal workstation hardware.*
+
+---
+
+## 👥 Attribution & Contact
+
+*Please provide attribution for the dataset creators and a point of contact.*
+
+| |                                                                                                                                                                                                                                                                                                                                                                                                           |
+| :--- |:----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
+| **Dataset Lead** | `[Haoying Zhou (student lead), Peter Kazanzides (faculty PI)]`                                                                                                                                                                                                                                                                                               |
+| **Institution** | `[Johns Hopkins University, Worcester Polytechnic Institute (Haoying Zhou)]`                                                                                                                                                                                                                                                                                                                              |
+| **Contact Email** | `[hzhou6@wpi.edu, hzhou62@jh.edu, pkaz@jhu.edu]`                                                                                                                                                                                                                                                                                                                                                          |
+| **Citation (BibTeX)** | <pre><code>@misc{[SurgSyncExt],<br>  author = {[Haoying Zhou, Peter Kazanzides, Chang Liu, Junlin Wu, Yimeng Wu, Hongjun Wu]},<br>  title = {[SurgSyncExt: Time-Synchronized Multi-Modal Dataset for Surgical Robotics]},<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> |

Surgical/jhu/lcsr/smarts/SurgSync-stitch-coldcut/P3/meta/README.md

@@ -1,2 +1,229 @@
-Placeholder
-Please refer to the README.md uploaded in the parent folder, thanks!
+# SurgSync-stitch-coldcut P3
+
+Canonical SMARTS leaf metadata README.
+
+- Canonical path: `Surgical/jhu/lcsr/smarts/SurgSync-stitch-coldcut/P3/`
+- Legacy source mapping: `Surgical/jhu/lscr/smarts/offline_recorder_extracted/offline_data_part3`.
+- This leaf is one canonical part of the broader JHU SMARTS dataset.
+
+The broader SMARTS dataset includes two sub-datasets: **offline_recorder** and **online_recorder**.
+
+---
+
+## 📋 At a Glance
+
+*Provide a one-sentence summary of your dataset.*
+
+Teleoperated demonstrations of a da Vinci robot (dVRK-Si) performing multiple canonical tasks on ex-vivo tissue or table-top phantoms; ~50% data collected within the Intuitive abdominal dome model. All data collection uses a modern stereo chip-on-tip endoscope. Both endoscope camera calibration and robot hand-eye calibration will be provided.
+
+---
+
+## File Structure
+
+For the dataset, it should
+
+```text
+./offline_recorder or online_recorder
+├── calibration/
+│   ├── case-*...
+│   │   ├── camera calibration
+│   │   │   ├── left.yaml
+│   │   │   ├── right.yaml
+│   │   │   └── stereo_calib_params.json
+│   │   └── hand_eye_calibration
+│   │   │   ├── PSM1/2-registration-dVRK.json
+│   │   │   └── PSM1/2-registration-open-cv.json
+├── data/
+│   └── case-*...
+├── videos/
+│   └── case-*...
+├── meta/
+│   ├── episodes.jsonl
+│   ├── episodes_stats.jsonl
+│   ├── tasks.jsonl
+│   ├── info.json
+│   └── README.md
+└── total_time.json
+```
+
+---
+
+## 📖 Dataset Overview
+
+*Briefly describe the purpose and content of this dataset. What key skills or scenarios does it demonstrate?*
+
+This dataset comprises 2500+ teleoperated expert demonstrations of multiple canonical tasks on ex-vivo tissue or table-top phantoms using a surgical robotic system. The tasks include simple interrupted stitch, cold-cut dissection, peg transfer and tissue manipulation. It includes stereo endoscope calibration and robot hand–eye calibration (use dVRK camera registration format), along with high-fidelity, sharp stereo endoscopic video captured via a modern chip-on-tip endoscope. In addition to full task executions, the dataset provides consecutive subtask-level trajectories, enabling analysis of skill composition and procedural structure. Overall, it supports research in imitation learning and skill learning, surgical gesture/subtask recognition, and advanced perception (e.g., tool/tissue interaction understanding and visual tracking) in realistic surgical scenarios
+
+| |                                                                                                                                                  |
+| :--- |:-------------------------------------------------------------------------------------------------------------------------------------------------|
+| **Total Trajectories** | `1087 ex-vivo (offline) + 1061 ex-vivo (online) + 361 phantom (online)`                                                                          |
+| **Total Hours** | `2.83 ex-vivo (offline) + 3.31 ex-vivo (online) + 0.35 phantom (online)`                                                                         |
+| **Data Type** | `[ ] Clinical` `[X] Ex-Vivo` `[X] Table-Top Phantom` `[ ] Digital Simulation` `[ ] Physical Simulation` `[ ] Other (If checked, update "Other")` |
+| **License** | CC BY 4.0                                                                                                                                        |
+| **Version** | `[1.0]`                                                                                                                                          |
+
+**Note:** The user study experiments have been conducted under HIRB00000701 at Johns Hopkins University.
+
+---
+
+## 🎯 Tasks & Domain
+
+### Domain
+
+*Select the primary domain for this dataset.*
+
+- [X] **Surgical Robotics**
+- [ ] **Ultrasound Robotics**
+- [ ] **Other Healthcare Robotics** (Please specify: `[Your Domain]`)
+
+### Demonstrated Skills
+
+*List the primary skills or procedures demonstrated in this dataset.*
+
+The primary skills or procedures demonstrated in this dataset include but not limited to:
+
+- simple interrupted stitching and its subtasks
+- cold cut dissection and its subtasks
+- peg transfer and its subtasks
+- tissue manipulation and its subtasks
+- ...
+
+---
+
+## 🔬 Data Collection Details
+
+### Collection Method
+
+*How was the data collected?*
+
+- [X] **Human Teleoperation**
+- [ ] **Programmatic/State-Machine**
+- [ ] **AI Policy / Autonomous**
+- [ ] **Other** (Please specify: `[Your Method]`)
+
+### Operator Details
+
+| | Description                                                                                                                                                                 |
+| :--- |:----------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
+| **Operator Count** | `[13]`                                                                                                                                                                      |
+| **Operator Skill Level** | `[4] Expert (e.g., Surgeon, Sonographer)` <br> `[5] Intermediate (e.g., Trained Researcher)` <br> `[4] Novice (e.g., ML Researcher with minimal experience)` <br> `[ ] N/A` |
+| **Collection Period** | From `[2025-08-01]` to `[2026-01-08]`                                                                                                                                       |
+
+### Recovery Demonstrations
+
+*Does this dataset include examples of recovering from failure?*
+
+- [ ] **Yes**
+- [X] **No**
+
+**If yes, please briefly describe the recovery process:**
+
+**Note:** For your reference, this dataset includes recovery and failure demonstrations, but they are not labeled.
+
+---
+
+## 💡 Diversity Dimensions
+
+*Check all dimensions that were intentionally varied during data collection.*
+
+- [X] **Camera Position / Angle**
+- [X] **Lighting Conditions**
+- [X] **Target Object** (e.g., different phantom models, suture types)
+- [X] **Spatial Layout** (e.g., placing the target suture needle in various locations)
+- [ ] **Robot Embodiment** (if multiple robots were used)
+- [X] **Task Execution** (e.g., different techniques for the same task)
+- [X] **Background / Scene**
+- [ ] **Other** (Please specify: `[Your Dimension]`)
+
+*If you checked any of the above please briefly elaborate below.*
+
+The camera (endoscope) can move with dVRK-Si ECM (endoscopy camera manipulator) as needed. The lighting conditions will be changed due to the camera movements. For the stitch, it has both ex-vivo tissue (chicken breast) and silicon phantom cases. The orientation of the wounds could vary from cases to cases. For cold-cut dissection, the target tissue can be different. The options include chicken drumsticks, thin-slide beef/pork, pork belly. The target suture needle could be in various locations. The background can be ex-vivo tissue or the surgical drapes. The detailed subtask approach can vary from cases to cases. For example, the knot-typing could be double throw or single throw.
+
+
+---
+
+## 🛠️ Equipment & Setup
+
+### Robotic Platform(s)
+
+*List the primary robot(s) used.*
+
+- **Robot 1:** `dVRK-Si (the next generation da Vinci Reseach Kit)`
+
+
+### Sensors & Cameras
+
+*List the sensors and cameras used. Specify model names where possible. (Add and remove rows as needed)*
+
+| Type | Model/Details                                                                                                           |
+| :--- |:------------------------------------------------------------------------------------------------------------------------|
+| **Primary Camera** | `Endoscopic Camera from Cornerstone Robotics Limited, 1920x1080 @ 60fps, recoreded in 10 FPS`                           |
+| **Room/3rd Person Camera** | `Intel RealSense RGBD camera, only using RGB channel as a mono side-view camera, 1920x1080 @ 30fps, recorded in 10 FPS` |
+| **Force/Torque Sensor** | `N/A`                                                                                                                   |
+| **Medical Imager** | `N/A`                                                                                                                   |
+| **Other** | `[Specify]`                                                                                                             |
+
+**Note** The camera calibration files and hand-eye calibration matrices are just for the primary camera.
+
+---
+
+## 🎯 Action & State Space Representation (will update if needed)
+
+*Describe how actions and robot states are represented in your dataset. This is crucial for understanding data compatibility and enabling effective policy learning.*
+
+**Please refer to the subfolder README.md for more details.**
+
+---
+
+## ⏱️ Data Synchronization Approach
+
+*Describe how you achieved proper data synchronization across different sensors, cameras, and robotic systems during data collection. This is crucial for ensuring temporal alignment of all modalities in your dataset.*
+
+We use a time-synchronized multi-modal data collection framework from an accepted ICRA 2026 paper:
+```
+@inproceedings{zhou2026surgsync,
+    title={SurgSync: Time-Synchronized Multi-modal Data Collection Framework and Dataset for Surgical Robotics},
+    author={Zhou, Haoying and ... and Kazanzides, Peter},
+    booktitle={IEEE Intl. Conf. on Robotics and Automation (ICRA)},
+    year={2026}
+}
+```
+We have two desktop: one for dVRK running and one for all the other pipelines. They are connected using ROS network and its NTP. 
+
+We have two modes when data collection, and the performance is highly dependent on the hardware.
+
+**Online(-matching) Recorder**: (not uploaded yet)
+
+The design enforces strict time synchronization using multi-threading: only samples that fall within a user-defined time tolerance are admitted. This yields slightly uneven inter-sample intervals (irregular dt), 
+but it retains the natural continuity of smooth teleoperation segments and avoids label/feature drift. In our study, a time tolerance of 10 ms is selected to ensure both high time 
+alignment tightness and consecutive recorder output. 
+
+As a result, the recorder time latency is 6.36 (+/- 4.72) ms with a median of 5.58ms. The recording frequency is 4.04 (+\- 1.69) Hz. We will assemble everything assume 10 FPS.
+
+**Offline(-matching) Recorder**: (already fully uploaded)
+
+Our offline-matching approach decouples recording from time alignments to maximize
+the recording system efficiency. Therefore, this recorder produces synchronized datasets in two stages: (i) a lightweight
+recorder logs camera streams to videos (pre-synced) and raw kinematic streams to binary files to disk with minimal processing;
+(ii) an offline post-processing pipeline reconstructs a fixed-rate frame sequence and, for each frame, gathers the five
+closest samples using nearest-timestamp lookup for subsequent interpolation. Compared to the online-matching recorder (which
+pairs visual and kinematic data in real time), this two-stage design avoids tolerance-based dropping of data during capture
+yielding a higher throughput and uniform intervals between synchronized data packets at the cost of requiring more storage
+and substantial time for post-collection time-matching and interpolation.
+
+As a result, the recorder time latency is 1.35 (+/- 0.81) ms with a median of 1.33ms. And the recording frequency is a solid 10.00 FPS.
+
+**Note:** *Both recorders may have few outliners due to our suboptimal workstation hardware.*
+
+---
+
+## 👥 Attribution & Contact
+
+*Please provide attribution for the dataset creators and a point of contact.*
+
+| |                                                                                                                                                                                                                                                                                                                                                                                                           |
+| :--- |:----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
+| **Dataset Lead** | `[Haoying Zhou (student lead), Peter Kazanzides (faculty PI)]`                                                                                                                                                                                                                                                                                               |
+| **Institution** | `[Johns Hopkins University, Worcester Polytechnic Institute (Haoying Zhou)]`                                                                                                                                                                                                                                                                                                                              |
+| **Contact Email** | `[hzhou6@wpi.edu, hzhou62@jh.edu, pkaz@jhu.edu]`                                                                                                                                                                                                                                                                                                                                                          |
+| **Citation (BibTeX)** | <pre><code>@misc{[SurgSyncExt],<br>  author = {[Haoying Zhou, Peter Kazanzides, Chang Liu, Junlin Wu, Yimeng Wu, Hongjun Wu]},<br>  title = {[SurgSyncExt: Time-Synchronized Multi-Modal Dataset for Surgical Robotics]},<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> |