Foxtech Provides Industrial Drone Solutions & UAV Payload Systems.
Introducing GripperGo Pro: A High-Precision Human Data Collection System for Embodied AI and Robot Learning
High-quality human demonstration data is becoming an essential foundation for embodied AI, imitation learning, and robotic manipulation research. However, recording a person completing a task is only the beginning. Research teams also need accurately aligned information about hand movement, gripper status, visual context, depth, motion, and spatial trajectories.
When these data streams are captured by separate devices, differences in timing, coordinate systems, and data quality can make the resulting dataset difficult to use. Problems may only become visible after collection is complete, leading to repeated demonstrations and additional post-processing work.
To address these challenges, Foxtech is introducing GripperGo Pro, a professional human data collection system designed for high-precision, gripper-based operation capture. By combining Vive base-station positioning, millisecond-level multimodal synchronization, bimanual collection devices, and DataCube edge-side processing, the system helps research teams build more structured and consistent datasets for robot learning.
What Problem Does GripperGo Pro Solve?
Robots must learn more than the final result of an operation. They need data describing how the task was performed.
For a grasping or manipulation task, researchers may need to capture:
- What the operator saw
- How both hands moved through space
- When the grippers opened or closed
- How the devices were oriented
- How quickly each movement occurred
- How the grippers approached and handled the object
- Whether contact or pressure changed during grasping
Ordinary video cannot provide all of this information. Building a custom collection system from multiple cameras, motion sensors, tracking devices, and processing computers can also introduce synchronization and calibration difficulties.
GripperGo Pro brings these functions into an integrated collection workflow. It records gripper operation together with visual, depth-related, motion, trajectory, and manipulation data, helping teams convert human demonstrations into structured datasets suitable for subsequent robot-learning workflows.
Why Is High-Precision Spatial Positioning Important?
Small differences in hand position can significantly affect a manipulation task.
When a person picks up a small object, inserts a component, uses a tool, or coordinates both hands, the spatial trajectory of each movement becomes part of the training data. If the tracking system introduces excessive drift or inconsistency, the recorded demonstration may not accurately represent the original operation.
GripperGo Pro uses a Vive base-station positioning system to provide 1 mm-level spatial positioning accuracy within a standardized collection environment.
This positioning method is especially useful for tasks that require:
- Repeatable human demonstrations
- Accurate bimanual trajectories
- Consistent spatial coordinates
- Detailed grasping and manipulation analysis
- Standardized data collection across multiple operators
- Comparison between repeated task sequences
Rather than focusing on unrestricted, base-station-free portability, GripperGo Pro is designed for high-precision collection in controlled environments such as laboratories, university research facilities, standardized collection rooms, and industrial training spaces.
How Does 1 ms Synchronization Improve Dataset Quality?
A robot-learning dataset may include images, depth-related information, IMU measurements, gripper opening distance, and spatial trajectories. Even when each sensor works correctly, the data can become difficult to use if the streams are not accurately synchronized.
For example, an image may show the gripper touching an object while the corresponding trajectory or opening-distance data represents a slightly earlier moment. When these timing differences accumulate across a dataset, they can reduce the consistency of the training samples.
GripperGo Pro supports 1 ms multimodal hardware synchronization, helping align visual, motion, and manipulation data along a shared timeline.
This synchronization allows researchers to study each action as a coordinated event. The position of the gripper, its opening status, its orientation, and the corresponding visual information can be matched more accurately, providing a better foundation for imitation learning and robotic manipulation training.
What Data Can GripperGo Pro Collect?
GripperGo Pro is designed to capture multiple forms of human operation data during grasping and manipulation tasks.
The system can collect:
- Video data
- Depth-related information
- Three-dimensional movement trajectories
- Gripper opening distance
- IMU attitude angles
- IMU acceleration
- IMU angular velocity
- Optional tactile and pressure-distribution data
The collection devices incorporate binocular vision and wide-angle visual capture to record the task environment and the relationship between the grippers and surrounding objects.
A 200 Hz IMU records high-frequency motion information, while 128 GB of onboard storage supports local data recording during collection sessions.
By combining these data types, GripperGo Pro provides more than a visual record. It describes how the operation develops over time and how the grippers move and interact with objects throughout the task.
How Do the Grippers Support Natural Human Demonstrations?
The system uses a pair of human-operated collection grippers, allowing the operator to demonstrate manipulation tasks directly.
Each gripper provides a 0–100 mm opening range and up to 20 N of gripping force, supporting common object grasping, handling, sorting, and placement demonstrations.
Using two grippers also makes it possible to collect bimanual operation data. This is important because many real tasks cannot be completed effectively with one hand alone. One hand may stabilize an object while the other adjusts it, or both hands may need to coordinate during lifting, assembly, packaging, and tool use.
Potential demonstrations include:
- Picking and placing objects
- Sorting items by type or position
- Opening containers
- Handling tools
- Transferring objects between hands
- Stabilizing and manipulating an object simultaneously
- Completing multi-step bimanual tasks
- Demonstrating basic assembly procedures
The objective is not to force human movement into the exact structure of a robot arm. Instead, the system records the operator’s actions as structured data that can later support robot-learning research and dataset development.
What Role Does DataCube Play in the Workflow?
Collecting data is only useful when operators can confirm that the recorded information is complete and usable.
GripperGo Pro includes the DataCube edge-side processing platform, which supports data acquisition, synchronization, playback, inspection, and preprocessing. It provides researchers with a more integrated way to manage the collection process without relying entirely on later offline inspection.
During or after a collection session, teams can review the recorded data and examine whether the required sensor streams were captured correctly. This helps identify incomplete, abnormal, or problematic samples earlier in the workflow.
Edge-side quality inspection is particularly valuable for projects involving many repeated demonstrations. Discovering a problem immediately may allow the operator to repeat only the affected task. Discovering the same problem after an entire collection campaign could require substantially more rework.
How Does Edge-Side Inspection Reduce Repeated Collection?
Human data collection can require hundreds or thousands of demonstrations. Even a small failure rate can create a significant amount of unusable data.
Potential problems may include:
- Incomplete task sequences
- Abnormal tracking data
- Missing sensor information
- Poor visual coverage
- Incorrect gripper operation
- Misaligned multimodal data
- Interrupted recordings
GripperGo Pro helps operators inspect and mark problematic data at the edge side. This allows collection teams to perform an initial quality check before transferring the dataset into a larger training or processing pipeline.
The system does not eliminate the need for dataset validation, annotation, or model-specific preprocessing. However, earlier inspection can help teams reduce avoidable recollection and organize their data more efficiently.
Why Is Optional Tactile Sensing Valuable?
Position and vision can show where a gripper is located, but they do not fully describe how it contacts an object.
For research involving grasp stability, fragile objects, deformable materials, or contact-rich manipulation, pressure distribution can provide important additional information. It may help researchers understand when contact begins, how the grasp changes, and how force is distributed across the gripping surface.
GripperGo Pro can be equipped with an optional 21-point tactile sensor array for capturing contact and pressure-distribution information during grasping.
The tactile array is not included as a standard built-in feature. Teams can select it when their projects require more detailed contact information, while projects focused primarily on vision, trajectories, motion, and gripper status can use the standard configuration.
How Does GripperGo Pro Connect Data Collection with Robot Learning?
A dataset must eventually move from the collection system into a robot-learning environment.
GripperGo Pro currently supports LeRobot-format data export, helping teams organize the captured demonstrations for downstream robot-learning workflows. This can reduce the effort required to convert raw collection results into a standardized structure before further processing.
The exported data may still need to be filtered, annotated, normalized, or adapted according to the research team’s model and robot platform. Nevertheless, structured export provides a clearer connection between human demonstration capture and subsequent model development.
A typical workflow can include:
- Setting up the Vive base stations in the collection area.
- Preparing the two GripperGo Pro collection devices.
- Connecting and checking the DataCube platform.
- Calibrating the collection environment.
- Defining the required task and demonstration procedure.
- Recording visual, spatial, motion, and gripper-operation data.
- Reviewing and marking problematic samples through edge-side inspection.
- Preprocessing and organizing the collected data.
- Exporting the dataset in LeRobot format.
- Adapting the data for the selected robot-learning pipeline.
This integrated workflow helps research teams move from individual human demonstrations toward repeatable dataset production.
What Applications Is GripperGo Pro Suitable For?
GripperGo Pro is designed for high-precision human operation capture in standardized environments.
Potential applications include:
Embodied AI Data Collection
The system can record human interactions with objects and workspaces, providing multimodal demonstrations for embodied intelligence research.
Imitation Learning
Researchers can collect examples showing how a person completes a task and use the structured data to support learning from demonstration.
Robotic Manipulation Training
Gripper trajectories, opening distance, motion information, visual context, and optional tactile data can support research into object handling and manipulation.
Bimanual Coordination Research
The paired collection devices allow researchers to study and record tasks requiring coordinated use of both hands.
Dexterous and Contact-Rich Tasks
When equipped with the optional tactile sensor array, the system can provide additional contact and pressure information for more detailed grasping research.
Standardized Dataset Development
Universities, laboratories, and data collection teams can use the system to record repeated demonstrations under controlled spatial and procedural conditions.
Industrial Task Demonstration
The system can support the recording of structured handling, sorting, tool-use, and basic assembly procedures for industrial robot-learning research.
Who Is GripperGo Pro Designed For?
GripperGo Pro is intended for organizations that require accurate and repeatable human demonstration data, including:
- Embodied AI research teams
- Robotics laboratories
- Universities and vocational training institutions
- Imitation-learning researchers
- Robotic manipulation developers
- Dexterous manipulation research teams
- AI and robotics startups
- Dataset collection and annotation teams
- Companies developing general-purpose robots
- Industrial research and training facilities
It is particularly suitable for teams that already have a defined collection space and need consistent data across repeated tasks, operators, or experimental sessions.
When Should Customers Choose GripperGo Pro?
GripperGo Pro is the appropriate choice when spatial accuracy, synchronization, repeatability, and edge-side data management are the main priorities.
Teams should consider the Pro system when they need:
- 1 mm-level spatial positioning
- Bimanual human operation capture
- 1 ms multimodal hardware synchronization
- A standardized, fixed collection environment
- Edge-side data playback and quality inspection
- Integrated preprocessing support
- LeRobot-format data export
- Optional tactile and pressure-distribution data
For projects that primarily require portable collection across homes, factories, supermarkets, outdoor areas, or other open and unstructured environments, a base-station-free solution may be more appropriate. GripperGo Pro is specifically optimized for controlled environments where high-precision and repeatable collection are more important than unrestricted mobility.
What Is Included with the System?
The standard GripperGo Pro package includes:
- Two GripperGo Pro data collection devices
- One DataCube edge-side processing platform
- One Vive base-station system
The two-device configuration supports bimanual data collection as standard. The optional 21-point tactile sensor array can be selected separately when contact and pressure-distribution information is required.
Building More Consistent Human Demonstration Datasets
The launch of GripperGo Pro provides embodied AI and robotics teams with a more integrated way to capture high-precision human manipulation data.
By combining 1 mm-level Vive positioning, 1 ms multimodal synchronization, bimanual gripper-based collection, binocular and wide-angle vision, a 200 Hz IMU, DataCube edge-side processing, and LeRobot-format export, the system connects human demonstration capture with downstream robot-learning workflows.
Its main value lies not in collecting more data streams individually, but in helping those streams remain spatially accurate, temporally aligned, inspectable, and easier to organize.
For research teams developing manipulation policies, imitation-learning models, bimanual robots, or embodied AI datasets, GripperGo Pro offers a practical platform for turning human operations into structured and repeatable training data.
