Summary: NOKOV Motion Capture has launched ShadowEngine, an end-to-end embodied AI training platform that unifies motion data, multimodal sensing, robot training, teleoperation, and Sim2Real validation into a closed-loop infrastructure for robotics research and deployment.
NOKOV Motion Capture recently announced the launch of ShadowEngine, an end-to-end embodied AI training platform that connects data collection, multimodal sensing, robot training, teleoperation, and Sim2Real validation into a unified workflow. Built on NOKOV's long-standing expertise in optical motion capture, ShadowEngine helps robotics teams accelerate robot learning from data collection to real-world deployment.
Why embodied AI training needs a closed loop
Recent advances in reinforcement learning, imitation learning, and humanoid robotics have significantly increased the demand for high-quality training data. However, many robotics teams still rely on fragmented workflows, where motion capture, multimodal sensing, data management, simulation, and validation are handled independently. This fragmentation makes it difficult to build reusable datasets or evaluate robot performance consistently across the entire development cycle.
As robots move from controlled demonstrations to real-world tasks, training is no longer only an algorithm problem. Teams need reliable data collection, synchronized multimodal signals, human-to-robot motion retargeting, expert teleoperation data, simulation training, and measurable validation on real robot systems.
In many workflows, these steps are handled by separate tools. Data may be difficult to align, motion demonstrations may not transfer cleanly to robot models, and simulation results may not be directly comparable with real-machine behavior. ShadowEngine is designed to connect these steps into one traceable data-training-validation loop.
What ShadowEngine provides
Rather than being a standalone training application, ShadowEngine serves as an embodied AI data infrastructure that connects data acquisition, multimodal management, robot training, and validation into one continuous workflow.
Module | Role | Key capability |
NKV-SE Source | Standardized data assets | Includes 1,000+ hours and 100,000 cleaned human motion source data items, plus 100,000 retargeted data items for Unitree G1. |
NKV-SE Fusion | Multimodal data platform | Synchronizes and manages data from motion capture, IMUs, force plates, EMG, EEG, robot joints, cameras, LiDAR point clouds, and audio. |
NKV-SE ReTarget | Motion retargeting | Uses dynamics optimization to retarget human motion to robot models, with real-time and file-based workflows and URDF/MJCF support. |
NKV-SE TeleOp | Teleoperation platform | Supports immersive, low-latency teleoperation and expert demonstration recording for robot learning. |
NKV-SE Train | AI model training workshop | Connects high-fidelity physics simulation with mainstream reinforcement learning frameworks. |
NKV-SE Valid | Validation and quality assessment | Compares position, velocity, acceleration, and custom metrics for Sim2Real validation and consistency testing. |
Core strengths for robotics teams
End-to-end closed loop: connects data collection, multimodal management, retargeting, training, teleoperation, and validation.
High-precision data foundation: uses NOKOV's optical motion capture capabilities to provide accurate motion, pose, and trajectory data.
Ready-to-use data assets: provides standardized motion datasets and retargeted robot data to reduce data preparation work.
Retargeting and teleoperation: helps transfer human motion to robot models and record expert demonstrations for imitation learning.
Sim2Real validation: feeds real-machine evaluation data back into the training workflow for continuous optimization.
Open integration: supports URDF/MJCF model formats and integration with the ROS/ROS2 ecosystem.
Application value
For embodied AI research teams, ShadowEngine helps build reusable motion datasets and manage human skills, robot states, and multimodal sensor data in a unified workflow.
For robot companies and developers, it supports humanoid robot training, motion retargeting, teleoperation demonstrations, policy debugging, and consistency evaluation, helping reduce the gap between offline data and real-machine execution.
For universities, vocational colleges, and robotics training labs, ShadowEngine can support embodied AI teaching and research workflows that combine motion capture, data management, robot retargeting, real-system observation, and performance evaluation.
By reducing data fragmentation and simplifying robot development workflows, ShadowEngine enables faster iteration from data collection to real-world deployment across different robotics applications.
ShadowEngine is built on NOKOV's self-developed motion capture hardware and software ecosystem. The hardware side can include optical motion capture systems, ASTRA markerless motion capture, and integrated data gloves, while the software side provides the six ShadowEngine modules.
NOKOV Motion Capture also provides technical support, regular software updates, and quarterly motion capture application training.
Through initiatives such as the New Generation Star Project and NOKOV Rising Talent Fund, NOKOV continues to support young robotics researchers and international academic exchange.
The launch of ShadowEngine marks NOKOV's expansion from motion capture technology into embodied AI data infrastructure. By connecting data acquisition, robot training, and real-world validation within a unified workflow, ShadowEngine enables robotics teams to build reusable datasets, accelerate robot skill development, evaluate real-world performance, and shorten the path from research to deployment.
ShadowEngine is a robot AI training platform from NOKOV Motion Capture. It covers motion data assets, multimodal fusion, motion retargeting, teleoperation, validation, and AI model training.
No. Humanoid robots are an important application scenario, but the platform also supports embodied AI robots, dexterous manipulation, robotics education, Sim2Real validation, and engineering testing.
ShadowEngine provides retargeted data for Unitree G1 and supports training, teleoperation, and validation workflows for G1 and other robot models.
Motion capture provides high-precision position, posture, trajectory, and time-series data for humans, tools, robot end-effectors, and test objects. These data can support demonstration, retargeting, training, and validation.
Single tools usually cover only simulation or model training. ShadowEngine connects data collection, management, retargeting, teleoperation, validation, and AI training into one workflow.
Room820, China Minmetals Tower, Chaoyang Dist., Beijing
info@nokov.cn
+ 86-10-64922321