The hidden engine behind humanoid robots: motion capture reshaping robot evolution!

The humanoid robot industry in March felt like a spectacular sci-fi blockbuster. Unitree Robotics’ G1 delivered a seamless performance, showcasing a stunning 720-degree spinning kick. Zongxing Robotics captivated the audience with smooth dance moves, perfectly recreating the iconic "Axe Gang" dance. Not to be outdone, Zhiyuan Lingxi’s X2 amazed the crowd with its exceptional interaction and motion capabilities—not only communicating with humans effortlessly but also riding bicycles and scooters like a person. Robots from various companies displayed their unique skills and innovations, making it a dazzling showcase of advanced technology.

Behind the rapid development of humanoid robots lies not only the thinking, decision-making, and deep interaction capabilities empowered by large models but also an invisible driving force quietly supporting their progress—motion capture technology. Known as the "motion mentor" for robots, it can capture high-precision joint angles of the human body for training humanoid robots to perform skills such as walking, running, jumping, grasping, holding, and placing objects. Additionally, it enables the creation of highly realistic virtual environments, allowing for testing across various conditions, including different environments, scenarios, and terrains, while reducing the cost of real-world experiments. Through extensive datasets and training, motion capture technology enhances the robots' dexterity and enables them to mimic and execute a wide range of tasks with greater proficiency.

With the application of motion capture technology, Tesla's humanoid robot Optimus has achieved higher-quality training results. The third-generation Optimus robot, recently unveiled, can not only dance and interact but also engage in conversations with humans, play rock-paper-scissors, and even mix cocktails—just like a robotic butler straight out of a sci-fi movie.

In the training and research of humanoid robots, comprehensive data support is an indispensable component. This includes motion capture data, robot body data, scene training data, and environmental terrain data. Such demands place strict requirements on the performance of motion capture systems, including high precision, ultra-low latency, strong robustness, and the ability to meet data collection needs across various scenarios. These capabilities are essential to effectively support robot motion evaluation and training.

Currently, CHINGMU possesses a high-precision motion capture system and comprehensive solutions to support robotics research. With sub-millimeter capture accuracy (0.02 mm), millisecond-level latency (2.5 ms), proprietary algorithm support, and outdoor enhancement technology, it provides highly accurate and reliable technical support for robot data collection. Furthermore, its innovative optical finger-tracking technology enables the precise capture of fine hand movements, making it applicable to dexterous hand research and robotic arm positioning measurements. Building on the success of this technology, CHINGMU has established collaborations with Honor, Tsinghua University, Zhejiang University, Soochow University, and others to advance the research and development of dexterous hands.

CHINGMU & Yinshi Robotics Teleoperation Demonstration

In addition, CHINGMU's partner, Humanoid Robot (Shanghai) Co., Ltd., utilizes CHINGMU's optical motion capture system to train "Qinglong," the world's first full-size humanoid robot. By collecting data from "Qinglong," they conduct posture analysis, motion evaluation, and other research activities.

Currently, leveraging its industry-leading technological advantages and rich service experience, CHINGMU provides robot training research solutions for robotics and embodied intelligence research. These solutions include motion capability assessment, performance comparison analysis, multi-scenario adaptability testing, spatial perception evaluation, and high-fidelity environmental simulation. CHINGMU offers multidimensional, high-quality training datasets to support precise learning and training of intelligent agents, ultimately contributing to the advancement of humanoid robot technology and industrial upgrades.