Foxtech Provides Industrial Drone Solutions & UAV Payload Systems.
Introducing HeadGo Ego: A New Head-Mounted Data Collection Solution for Embodied AI and Robot Learning
Building capable embodied AI systems requires more than recording ordinary video. Researchers need structured, synchronized data that helps robots understand what a person sees, how their hands move, and how an operation develops over time.
To address this challenge, Foxtech is introducing HeadGo Ego, a new head-mounted human data collection device designed for embodied AI, robot learning, imitation learning, and dexterous manipulation training.
Combining first-person visual capture, hand keypoint tracking, motion sensing, real-time monitoring, and an integrated backpack-style DataCube system, HeadGo Ego helps research teams collect human operation data in a mobile and practical workflow.
What Problem Does HeadGo Ego Solve?
One of the main challenges in embodied AI development is collecting sufficient high-quality human demonstration data.
A conventional camera can record what happens during a task, but video alone may not provide the structured information required for robot learning. Researchers may also need to understand the spatial position of the hands, individual hand keypoints, movement trajectories, depth information, and the exact timing of each action.
Collecting these data streams with separate devices can create additional difficulties. The equipment may restrict the operator’s movement, require repeated configuration, or make it difficult to keep different data sources aligned.
HeadGo Ego is designed to simplify this process by collecting first-person operation data directly from the operator’s perspective. It combines visual information with hand pose, spatial coordinates, depth, timestamps, and motion trajectories, providing a richer data foundation for embodied AI training.
Why Is First-Person Data Important for Robot Learning?
Robots must learn not only what a completed task looks like, but also how a person performs it.
During activities such as picking up an object, organizing items, operating a tool, or completing a multi-step manipulation task, the operator’s viewpoint contains valuable contextual information. It shows which objects are visible, how the hands approach them, and how the surrounding environment changes throughout the operation.
Because HeadGo Ego is worn on the head, the cameras follow the operator’s natural viewing direction. This allows the system to record the task from a perspective closely connected to the operator’s attention and actions.
The resulting first-person data can support research into:
- Embodied AI
- Robot learning
- Imitation learning
- First-person human operation analysis
- Dexterous manipulation
- Human demonstration dataset development
What Data Can the System Collect?
HeadGo Ego goes beyond standard visual recording by producing multiple forms of operation-related data.
The system can capture 21 hand keypoints, along with hand pose, spatial coordinates, depth, timestamps, and motion trajectories. These data help describe where the operator’s hands are located and how they move throughout a task.
For research teams, this makes it possible to study an operation as a sequence of structured movements rather than only as a video clip. The information can be used to build datasets for training and evaluating robot learning models.
The system is equipped with three cameras:
- Two binocular cameras
- One RGB camera
This configuration supports RGB and depth-related visual data acquisition. A 9-axis IMU provides additional motion-related sensing, helping capture how the operator moves while completing the task.
How Does the Wide Field of View Improve Data Collection?
Human manipulation does not always take place directly in the center of the camera image. The operator may reach toward objects on either side, look down at a work surface, or move between different parts of the environment.
HeadGo Ego provides a field of view of:
- 150° diagonal
- 128° horizontal
- 80° vertical
This wide viewing range helps capture more of the operator’s workspace and reduces the likelihood that important hand movements or nearby objects will fall outside the recorded area.
For tasks involving two-handed manipulation, large work surfaces, or frequent changes in viewing direction, a broader first-person view can provide more complete visual context for subsequent data processing and model training.
How Does HeadGo Ego Support Natural Human Demonstrations?
Human demonstration data is most useful when the operator can perform tasks naturally.
Because the main sensing device is worn on the head and the processing and power components are integrated into a backpack-style system, the operator can move through the environment without being restricted to a fixed recording station.
This mobile configuration is suitable for collecting data across different workspaces and everyday operating environments. Researchers can demonstrate tasks using their own hands while the system records the operator’s viewpoint, hand movements, and motion-related information.
The backpack integrates the main processing unit and power supply. It also incorporates a shock-resistant structure and active ventilation design to support mobile data collection.
How Can Operators Monitor the Collection Process?
A common problem during data collection is discovering too late that a recording was incomplete, storage was insufficient, or part of the system was not operating correctly.
HeadGo Ego addresses this issue with a main control screen that provides real-time system monitoring. Operators can check:
- Live video
- Device temperature
- CPU and memory usage
- Available storage
- Frame rate
- Device health
- Data collection progress
This visibility helps the operator identify potential issues while the task is being recorded. It can reduce the risk of completing a long demonstration only to find that important data is missing or unusable.
Real-time monitoring is especially valuable when research teams need to collect many demonstrations across multiple operators, tasks, or environments.
How Does the System Reduce Setup Time?
Repeated network configuration can slow down data collection, particularly when moving between locations or changing networks.
HeadGo Ego supports QR-code-based Wi-Fi configuration. The operator can scan a QR code using the Ego camera to configure the network connection, reducing the need for lengthy manual setup.
This feature can be useful for teams that collect data in different laboratories, workspaces, or field environments. Faster network configuration helps researchers spend more time recording demonstrations and less time preparing equipment.
Can HeadGo Ego Be Used Independently?
Yes. HeadGo Ego can operate as an independent first-person human data collection system.
When used independently, it can collect visual information, hand pose data, 21 hand keypoints, spatial coordinates, depth, timestamps, and motion trajectories. This makes it suitable for projects focused on egocentric vision, human operation analysis, imitation learning, and embodied AI dataset development.
The standard system includes:
- One HeadGo Ego head-mounted data collection device
- One DataCube backpack data processing system
Can It Work with Other Human Data Collection Devices?
HeadGo Ego can also be used together with GripperGo Pro.
In a combined workflow, HeadGo Ego can supplement the dataset with first-person visual information, hand pose, keypoint data, and global operation information. This allows research teams to capture human demonstrations from multiple complementary perspectives.
The choice between using HeadGo Ego independently or integrating it with GripperGo Pro depends on the project’s data requirements. Teams that primarily need mobile first-person data collection can use HeadGo Ego on its own, while projects requiring a more comprehensive human operation dataset can consider a combined setup.
What Types of Tasks Can Be Recorded?
HeadGo Ego is intended for first-person human operation capture. It can be used to record demonstrations in which the operator interacts naturally with objects and the surrounding workspace.
Potential research tasks include:
- Object picking and placement
- Sorting and organization
- Two-handed manipulation
- Tool-use demonstrations
- Multi-step operation recording
- Human hand trajectory analysis
- Dexterous manipulation dataset collection
- Everyday task demonstrations for imitation learning
The specific task design and dataset structure can be determined according to the research team’s robot platform, learning framework, and training objectives.
Who Is HeadGo Ego Designed For?
HeadGo Ego is primarily intended for organizations developing embodied intelligence and robot learning systems, including:
- Universities and robotics laboratories
- Embodied AI research teams
- Robot learning and imitation learning researchers
- Dexterous manipulation developers
- AI and robotics startups
- Dataset collection teams
- Companies developing general-purpose robotic systems
It can be particularly useful for teams that need to collect first-person human demonstrations across multiple tasks or environments without building a complex fixed capture system for every project.
How Can HeadGo Ego Improve the Data Collection Workflow?
HeadGo Ego brings sensing, processing, monitoring, and mobile deployment into a more integrated system.
Instead of assembling separate cameras, hand-tracking devices, motion sensors, processing hardware, and power supplies for each collection session, teams can work with a unified head-mounted and backpack-based configuration.
A typical workflow can include:
- Preparing the HeadGo Ego and DataCube system.
- Connecting to the required Wi-Fi network through QR-code scanning.
- Checking video, storage, frame rate, and device status on the monitoring screen.
- Starting the human demonstration task.
- Collecting first-person visual, hand keypoint, depth, timestamp, and trajectory data.
- Monitoring the collection progress in real time.
- Organizing the captured data for subsequent robot learning and model development.
This approach helps make human demonstration collection more repeatable and manageable, especially when a project requires large numbers of task samples.
A More Practical Way to Build Human Demonstration Datasets
The launch of HeadGo Ego provides embodied AI developers with a new option for collecting structured first-person human operation data.
By combining binocular and RGB vision, 21 hand keypoints, spatial pose information, depth-related data, a 9-axis IMU, real-time monitoring, and a mobile DataCube backpack, the system is designed to help researchers capture not only what the operator sees, but also how the operation is performed.
Whether used independently or alongside GripperGo Pro, HeadGo Ego can support the development of richer datasets for robot learning, imitation learning, and dexterous manipulation research.
