Humanoid Robots and Embodied AI

Build a scalable motion data factory for robot training

For humanoid robots, embodied AI, imitation learning, teleoperation, and Sim2Real research, Semcam Live provides markerless human demonstration capture, Goku rigid-body 6DoF tracking, local synchronization in Active Center, and multi-format data output. Research teams can capture data from people, robots, tools, props, and experimental objects in the same coordinate system and timeline.

Core Value

Robot teams need more than a single human body rig estimate. They need a stable, reviewable, calibrated data pipeline that can connect to training tools. Semcam Live captures natural human motion with edge AI camera arrays, then locally outputs 3D reconstruction, body rig solving, hand motion, root trajectories, joint angles, and retarget data. The Goku optical system can simultaneously track robot end effectors, tools, props, and experimental objects in 6DoF. Together in Active Center, they share one project, coordinate system, and timeline, helping teams organize human demonstrations, robot states, and scene objects into long-term data assets.

Application Scenarios

01

Imitation learning datasets

Capture human demonstrations such as walking, turning, grasping, carrying, assembly, tool operation, and bimanual coordination, then output body rigs, joint angles, hand motion, and target poses for training pipelines.

02

Teleoperation and motion reproduction

Use real-time markerless motion data to drive robots, digital twins, or simulated characters, helping R&D teams quickly validate motion links for remote control, motion rehearsal, and task reproduction.

03

Human-robot collaboration research

Record human motion, robot end effectors, tools, and experimental objects in the same space to support quantitative analysis of collaborative tasks, obstacle avoidance, spatial behavior, and interaction timing.

04

Tool and object tracking

Use Goku optical rigid-body tracking to record 6DoF poses of tools, props, fixtures, end effectors, mobile platforms, and other rigid objects, covering object states that markerless human capture cannot reliably represent on its own.

05

Sim2Real and simulation validation

Import human motion, robot poses, and scene object data into MuJoCo, Isaac, ROS, Unity, or in-house simulation systems for coordinate conversion, IK validation, policy replay, and real-world comparison.

06

University and laboratory teaching

Suitable for robotics, AI, human-computer interaction, motion control, and digital twin courses, allowing students to complete demonstration capture, data parsing, retargeting, and simulation validation without complex wearable setup.

How Semcam Live Supports Robot Data Workflows

01

Markerless human demonstration capture

Demonstrators do not need mocap suits or markers, and can perform walking, running, turning, grasping, carrying, tool operation, and bimanual coordination with lower setup cost and less motion interference.

02

Human + rigid-body data in one scene

Semcam Live captures human motion while Goku tracks robot end effectors, tools, props, experimental objects, and other rigid bodies, putting people, robots, and objects into one data chain.

03

Unified coordinate system and timeline

Active Center manages Semcam Live AI cameras and Goku optical cameras together, synchronizing, calibrating, recording, and replaying human motion, rigid-body poses, and external device data in one project.

04

Data output for training

Outputs include body rigs, root trajectories, joint angles, hand motion, retargeted target poses, and rigid-body 6DoF data for imitation learning, teleoperation, dataset construction, and policy validation workflows.

05

Live Preview and HPE Recalculation

The 120fps real-time pipeline supports on-site preview, capture checks, teleoperation validation, and online system debugging, while the HPE High-Precision Extension can generate higher-quality training data after capture.

06

Local deployment and developer interfaces

Motion capture, reconstruction, recording, project management, and data output can all run locally, with interfaces for ROS, C++, Python, MATLAB, MuJoCo, Isaac, and related workflows.

Differentiated Capabilities

The value of Semcam Live for robot data capture is not offline pose estimation from a single video clip. It forms a stable data pipeline with synchronized multi-camera capture, edge AI, precise calibration, and local 3D reconstruction. Cameras process complex visual features at the edge, while the central system receives structured data for cross-camera fusion, triangulation, body rig solving, IK, filtering, and retargeting. This avoids continuously sending massive raw video streams back to a server and makes large multi-camera matrices more practical in bigger experimental spaces.

Relationship with traditional robot data methods

Gamepad teleoperation, wearable devices, and traditional optical marker setups still have value in robotics R&D, but they often introduce wearable burden, motion restrictions, marker preparation, or separation between rigid-object data and human data. Semcam Live lowers the barrier to human demonstration capture, while Goku optical tracking complements it with rigid-body tracking for tools, end effectors, and scene objects, allowing human motion, robot state, and object trajectories to be recorded and reproduced in the same coordinate system and timeline.

Related technologies and integrations

250 Key Points120fpsImitation LearningTeleoperationSim2RealROSMuJoCoIsaacRetarget6DoFC++ / Python

Recommended product model

Depending on your scene scale and needs, we recommend the following configurations

Semcam Live
PRO+

A professional edge AI mocap camera for high-quality motion data capture, with a 55 TOPS embedded GPU and support for the HPE High-Precision Extension.

For life sciences, university labs, robotics training, and animation/VFX

Built-in computing power
55 TOPS
System power consumption
15W
maximum resolution
2.3 million pixels
Maximum number of people caught
4 people
furthest tracking distance
18m
Joint 3D rotation angle
support
finger capture
✓ Supported
facial capture
—
HPE
✓ Supported
Semcam Live
ULTRA

A flagship edge AI mocap camera for large spaces and multi-person capture, with a 100 TOPS embedded GPU, support for 12-person capture, and HPE high-precision processing.

For large-space labs, LBE, simulation training, and multi-person data capture

Built-in computing power
100 TOPS
System power consumption
18W
maximum resolution
5.2 million pixels
Maximum number of people caught
12 people
furthest tracking distance
45m
Joint 3D rotation angle
support
finger capture
✓ Supported
facial capture
✓ Supported
HPE
✓ Supported
Customized solution

Not sure which one is right for you?

Contact our technical team to customize the optimal camera array solution based on your scene size, number of people, space and accuracy requirements.

Turn robot motion data capture from one-off demos into continuous production.

Contact the Semcam Live team to plan a camera array and data workflow for humanoid robotics and embodied AI research based on your robot type, experimental space, participant count, tool objects, synchronization devices, and training data format requirements.

FAQ

FAQ

ProductsIndustries

What is Semcam Live?

Semcam Live is an edge computing AI markerless motion capture system. Target detection, 2D keypoints, ReID, tracking, segmentation, cropping and confidence calculation are completed on the camera side, and Active Center locally completes cross-camera matching, triangulation, 3D keypoints, body rig solving, IK, filtering, retargeting and exporting.

IndustriesWorkflow

Does body capture require wearable equipment or patches?

Human motion capture does not require motion capture suits, inertial sensors or reflective markers. If your project requires tracking rigid body objects, such as tools, props, robot ends, handles, or training equipment, you can use Goku for 6DoF rigid body tracking.

ProductsIndustriesWorkflow

When should I use real-time capture versus HPE?

The live link delivers 120fps motion output and sub-100ms end-to-end latency, making it suitable for live previews, live drives, interactions and acquisition inspections. HPE is a post-harvest high-precision processing pipeline suitable for research, training data production, key shots and projects requiring higher quality keypoint results.

ProductsIndustriesWorkflow

Can Semcam Live be used with Goku?

Yes. Semcam Live is responsible for unmarked human motion, and Goku is responsible for rigid body objects such as robot joints, end effectors, tools, props, handles, training equipment, rackets, drones, and cameras. Both share the coordinate system, timeline, and project in Active Center.

IndustriesData Output

Which analysis workflows are supported for life sciences and university labs?

It can be connected to OpenSim, Visual3D, C3D, mainstream force plate, electromyography and pressure platform processes, and can be extended with index calculation, action quality evaluation and data export through Semcam Biomechanics Plugin and Semcam Action Quality Assessment Plugin.